Tailored recovery of rare earth elements from fluorescent lamps industry leachates towards new catalytic applications
Abstract
As terras raras são elementos químicos conhecidos como REE (do inglês, Rare Earth Elements), essenciais em diversas aplicações tecnológicas que fazem parte do quotidiano. A procura destes elementos químicos tem vindo a aumentar devido ao seu grande consumo e aplicações. Os zeólitos são aluminossilicatos sólidos porosos, bastante utilizados na recuperação de metais de águas contaminadas, podendo ainda ser aplicados como catalisadores heterogéneos em reações de interesse industrial. Esta tese tem como objetivos o desenvolvimento de um sistema de recuperação de terras raras de águas contaminadas em múltiplos ciclos de adsorção e dessorção e a validação da utilização das terras raras suportadas em zeólitos em reações catalíticas. A modificação química da superfície de diferentes zeólitos foi efetuada com vista à otimização da recuperação de terras raras de águas contaminadas. Verificou-se que o melhor zeólito modificado consegue remover mais de 80 % de todos as terras raras presentes na solução de ensaio e obtiveram-se recuperações acima de 90 % por posterior lixiviação dos sorventes. A estes resultados foram aplicadas, com sucesso, técnicas de machine learning (ML), nomeadamente supervised e unsurpervised learning. Em regime de supervised learning foram aplicados algoritmos de classificação aos dados experimentais e a possibilidade de regressão. Em unsupervised learning, foram aplicados algoritmos para redução da dimensionalidade dos dados utilizados na elaboração dos clusters. A segunda parte é a continuação natural da anterior, em que o melhor zeólito passa de testes batch para ensaios em coluna, nos quais se atingiu mais de 70 % de remoção e uma recuperação, após a otimização, acima de 80 % para todas as REE testadas. As mesmas análises de ML foram aplicadas com sucesso aos resultados obtidos neste sistema aberto. A capacidade catalítica das REE suportadas nos zeólitos foi testada em reações tipo Fenton para a degradação de dois corantes, a tartrazina e o índigo de carmim, após a adição de ferro aos referidos zeólitos com REE. A degradação obtida para a tartrazina foi superior a 80 %, enquanto para o índigo de carmim foi superior a 95 %. Algoritmos de ML foram aplicados eficazmente na análise dos resultados obtidos de degradação.
Full text
Universidade do Minho Escola de Engenharia October 2023 Óscar José Maciel Barros Tailored recovery of rare earth elements from fluorescent lamps industry leachates towards new catalytic applications Óscar José Maciel Barros UMinho | 2023Tailored recovery of rare earth elements from fluorescent lamps industry leachates towards new catalytic applications
Óscar José Maciel Barros Tailored recovery of rare earth elements from fluorescent lamps industry leachates towards new catalytic applications Dissertation for PhD degree in Chemical and Biological Engineering Work carried out under the supervision of: Maria Teresa de Jesus Simões Campos Tavares, PhD Maria Isabel Pontes Correia Neves, PhD October 2023 Escola de Engenharia
DIREITOS DE AUTOR E CONDIÇÕES DE UTILIZAÇÃO DO TRABALHO POR TERCEIROS Este é um trabalho académico que pode ser utilizado por terceiros desde que respeitadas as regras e boas práticas internacionalmente aceites, no que concerne aos direitos de autor e direitos conexos. Assim, o presente trabalho pode ser utilizado nos termos previstos na licença abaixo indicada. Caso o utilizador necessite de permissão para poder fazer um uso do trabalho em condições não previstas no licenciamento indicado, deverá contactar o autor, através do RepositóriUM da Universidade do Minho. Licença concedida aos utilizadores deste trabalho Atribuição-Não Comercial-Sem Derivações CC BY-NC-ND https://creativecommons.org/licenses/by-nc-nd/4.0/
| iii Acknowledgements Developing this thesis would not be possible without the support of several people, for which I am and will be deeply thankful. Each one contributed to my growth not only at a personal level but also as a scientist through the share of knowledge and advice. I would start by expressing my gratitude to my supervisor, Professor Teresa Tavares, for allowing me to join her research group as a researcher and later as a PhD student. Also, I want to thank her for the support these last 4 years regarding my work and progress, but primarily for believing in my work and my capacity to overcome all the difficulties during this journey. To my co-supervisor, Professor Isabel Neves, for all the support, help and confidence boost given me during this journey, especially when my confidence levels were low. To Professor Pier Parpot, for all the availability to help me and teaching regarding the Machine Learning algorithms. To the Centre of Biological Engineering (CEB) and to all its technical staff for the excellent support. To all the people from the LEQ lab at CEB for all the help given, healthy discussions and friendship created. I also want to thank my friends who accompanied me from the start of this new adventure that we called university. I want to thank my friends from LTEB lab at CEB for all the support, conversations, help and friendship created. To a new friend I made during the last months of this adventure, I want to thank you for your patience and comprehension during this short time we have known each other. My truly and deeply felt thanks to my parents, who gave unconditional love, education, support, and the hours they listened to me. Thank you for all of these and for making me the person I am today. To my brothers, for all the love and understanding and for being at my side, always making me push forward and never give up. All of them were essential to me during this time. Adding to them, I also want to thank myself for fighting, working, believing, dreaming and realizing what I aimed to do. Finally, I would like to thank the Portuguese Foundation for Science and Technology (FCT) for my PhD grant (SFRH/BD/140362/2018) and all the funding entities that supported the work. FCT under the scope of the strategic funding of UID/BIO/04469/2020 and UID/QUI/0686/2020 units and BioTecNorte operation (NORTE-01-0145-FEDER-000004) funded by the European Regional Development Fund under the scope of Norte2020—Programa Operacional Regional do Norte, Portugal.
| iv STATEMENT OF INTEGRITY I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho.
| v Resumo Recuperação de terras raras de lixiviados de lâmpadas fluorescentes para aplicações catalíticas As terras raras são elementos químicos conhecidos como REE (do inglês, Rare Earth Elements), essenciais em diversas aplicações tecnológicas que fazem parte do quotidiano. A procura destes elementos químicos tem vindo a aumentar devido ao seu grande consumo e aplicações. Os zeólitos são aluminossilicatos sólidos porosos, bastante utilizados na recuperação de metais de águas contaminadas, podendo ainda ser aplicados como catalisadores heterogéneos em reações de interesse industrial. Esta tese tem como objetivos o desenvolvimento de um sistema de recuperação de terras raras de águas contaminadas em múltiplos ciclos de adsorção e dessorção e a validação da utilização das terras raras suportadas em zeólitos em reações catalíticas. A modificação química da superfície de diferentes zeólitos foi efetuada com vista à otimização da recuperação de terras raras de águas contaminadas. Verificou-se que o melhor zeólito modificado consegue remover mais de 80 % de todos as terras raras presentes na solução de ensaio e obtiveram-se recuperações acima de 90 % por posterior lixiviação dos sorventes. A estes resultados foram aplicadas, com sucesso, técnicas de machine learning (ML) , nomeadamente supervised e unsurpervised learning . Em regime de supervised learning foram aplicados algoritmos de classificação aos dados experimentais e a possibilidade de regressão. Em unsupervised learning , foram aplicados algoritmos para redução da dimensionalidade dos dados utilizados na elaboração dos clusters . A segunda parte é a continuação natural da anterior, em que o melhor zeólito passa de testes batch para ensaios em coluna, nos quais se atingiu mais de 70 % de remoção e uma recuperação, após a otimização, acima de 80 % para todas as REE testadas. As mesmas análises de ML foram aplicadas com sucesso aos resultados obtidos neste sistema aberto. A capacidade catalítica das REE suportadas nos zeólitos foi testada em reações tipo Fenton para a degradação de dois corantes, a tartrazina e o índigo de carmim, após a adição de ferro aos referidos zeólitos com REE. A degradação obtida para a tartrazina foi superior a 80 %, enquanto para o índigo de carmim foi superior a 95 %. Algoritmos de ML foram aplicados eficazmente na análise dos resultados obtidos de degradação. Palavras-chave: Terras Raras; Adsorção; Dessorção; Machine learning; Catálise
| vi Abstract Tailored recovery of rare earth elements from fluorescent lamps industry leachates towards new catalytic applications Rare earth elements, known as REE, are essential chemical elements with diverse technological applications that are part of everyday life. The demand for these elements has increased due to their high consumption and applications. Zeolites are porous solid aluminosilicates, widely used to recover metals from contaminated water and can also be applied as supports for heterogeneous catalysis. This thesis aims to develop a system for recovering REE in multiple adsorption and desorption cycles and to verify the possible use of REE supported on zeolites in catalytical reactions. Finally, machine learning (ML) algorithms were tested to select and to predict the behavior of REE and zeolite systems. Chemical surface modifications of different zeolites were carried out to optimize the recovery REE from contaminated water. The best modified zeolites can remove more than 80 % of all tested REE present in the test solution and recover above 90 % have been obtained by leaching the sorbents. ML techniques were successfully applied to these results namely unsupervised and supervised learning. Within the supervised learning, classification algorithms were applied to the collected data to select the best modified zeolites and to test the possibility of regression, validating the predictive ability of these algorithms on the removal of REE from wastewater. Within the unsupervised learning, algorithms were applied to reduce the dimensionality of the data used to create the clusters. The second part is the natural follow up of the previous work, where the best zeolites were used in continuous flow assays. A total REE removal above 70 % was obtained and a total REE recovery above 80 % for all REE tested after optimization. The same ML analyses were successfully applied. The catalytical capacity of the REE/zeolite was tested in Fenton-type reactions for the degradation of two dyes, tartrazine and indigo carmine, after the addition of iron to the REE/zeolite. The degradation for tartrazine was above 80%, while for indigo carmine, it was higher than 95 %. As previously described, ML algorithms were successfully applied to analyze the obtained results. Keywords: Rare Earths Elements; Adsorption; Desorption; Machine learning; Catalysis
| vii List of Contents Acknowledgements ........................................................................................................................ iii Resumo .......................................................................................................................................... v Abstract ......................................................................................................................................... vi List of Figures ................................................................................................................................. x List of Tables ................................................................................................................................ xvi List of Abbreviations and Acronyms ............................................................................................... xx Chapter 1 – Motivation and Outline .................................................................................. 1 1.1. Context and Motivation ........................................................................................................ 2 1.2. Thesis outline ...................................................................................................................... 4 1.3. Scientific outputs ................................................................................................................. 5 1.4. References .......................................................................................................................... 6 Chapter 2 – Literature Review ........................................................................................... 7 2.1. Rare Earth Elements ........................................................................................................... 8 2.1.1. REE applications ........................................................................................................ 11 2.1.2. Recovery and recycling importance of REE ................................................................. 12 2.2. Zeolites ............................................................................................................................. 15 2.2.1. Applications ............................................................................................................... 16 2.3. Sorption Processes ............................................................................................................ 17 2.4. Catalysis ........................................................................................................................... 18 2.4.1. Catalysts using zeolites and REE ................................................................................ 19 2.5. Machine Learning .............................................................................................................. 20 2.5.1. ML algorithms ........................................................................................................... 21 2.5.2. ML applications ......................................................................................................... 22 2.6. References ........................................................................................................................ 23 Chapter 3 – Chemical modification of zeolites for the recovery of Rare Earth Elements .. 32 3.1. Introduction....................................................................................................................... 33 3.2. Materials and Methods ...................................................................................................... 34 3.2.1. Materials ................................................................................................................... 34 3.2.2. Zeolite modifications .................................................................................................. 34 3.2.3. Characterization ........................................................................................................ 35
| xiv and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent. ........................................................................... 110 Figure S-4.14: Elbow method for selecting the best desorption: A) PCA, B) K-Means and C) KNN Classifier ........................................................................................................................................ 113 Figure S-4.15: PCA maps. Each map represents the distinct distribution for the selected features. 114 Chapter 5 Figure 5.1: Tar degradation in the presence of La10Fe10ZSM5 (MFI): Z15, prepared by ion exchange ( ) and Z1, prepared by impregnation ( ); La10Fe10NaY (FAU), Z2, prepared by impregnation ( ), La10Fe10NaX powder (FAU), A3, prepared by impregnation ( ), and La10Fe10NaX pellet (FAU), A7, prepared by impregnation ( ). Conditions of the reaction: 20 mg of catalyst/25 mL of a 30 ppm solution of Tar; 0.5 mL of H2O2 90 mM; pH=3; T=40 0C; t = 180 min. ..................................................................... 122 Figure 5.2: FTIR spectra of the REE/Fe-zeolite catalysts in the spectral region of 2000 to 450 cm-1. ...................................................................................................................................................... 123 Figure 5.3: Graphical distribution of the ML analysis for the different catalysts: A) PCA analysis; B) KMeans algorithm. The IS and CT values are referred to initial screening and catalytical tests, respectively, for tartrazine (Tar) and for indigo carmine (IC). ................................................................................ 125 Figure 5.4: Degradation of Tar and IC using the REE/Fe-zeolite catalysts for A) La, B) Ce and C) Pr, after IS test. The catalysts are divided into ZSM5 (MFI) with a REE concentration of: 10 mg/L ( ) and 25 mg/L ( ); NaY (FAU) with a REE concentration of: 10 mg/L ( ) and 25 mg/L ( ). Conditions of the reaction: 20 mg of catalyst/25 mL at 30 ppm of dye; 0.5 mL of 90 mM of H2O2 for Tar and of 12 mM of H2O2 for IC; pH=3; T=40 0C and 3 h of reaction. .................................................................... 127 Figure 5.5: IS conversion of the two dyes by La/Fe catalysts prepared by different methods. The catalysts are divided into ZSM5 (MFI): impregnation ( ) and ion exchange ( ); NaY (FAU): impregnation ( ) and ion exchange ( ). Conditions of the reaction: 20 mg of catalyst/25 mL at 30 ppm of dye; 0.5 mL of 90 mM H2O2 for Tar and 12 mM H2O2 for IC; pH=3; T=40 0C and 3 h of reaction. ...................................................................................................................................................... 129 Figure 5.6: Classification of all different REE/Fe-zeolite catalysts using ML algorithms: A) KNN classifier, Decision Tree classifier and Logistic Regression B) Random Forest classifier. The 1 represent a good catalyst, while the 0 is a bad catalyst accordantly to the evaluation performed. The different colors, violet and orange, represent the zone of a good or bad catalyst, respectively. ........................................... 132
| xv Figure 5.7: Confusion matrix for the test data for KNN, Decision Tree and Logistic Regression (A) and Random Forests (B). The values shown refer to the fraction of the true correct predictions (when the model got it right) and false incorrect predictions (when the model got it wrong). ............................. 134 Figure 5.8: Heatmap representing the Pearson correlation between the different features considered on the degradation assays. The left scale represents the different correlation values and the respective colors. ...................................................................................................................................................... 135 Figure 5.9: Conversion by Fenton-like reaction over time of Tar (A) and of IC (B) using the selected catalysts (Z15 and Z16) and the controls. The degradation assisted with 0.5 mL of H2O2 is represented with a full line, while the reaction with 5 mL of H2O2 used a dashed line. Conditions of reaction: 200 mg of catalyst/250 mL at 30 ppm of dye; 0.5 mL or 5 mL of 90 mM H2O2 for Tar and 12 mM H2O2 for IC; pH=3; T=40 0C and 300 min of reaction. ........................................................................................ 137 Figure S-5.1: Elbow method to select the best option for PCA (A), K-Means (B) and accuracy for the training and test sets for the K-Neighbors Classifier (C). .................................................................. 144
| xvi List of Tables Chapter 3 Table 3.1: Description of the modification of each tested zeolite with the respective name. .............. 35 Table 3.2: The binary classification for the different REE. The C/C0 values were given a classification accordantly. The mean value of these intervals was taken and given the respective binary classification. ........................................................................................................................................................ 38 Table 3.3: EDS surface analysis of modified 13X zeolite and controls. ............................................. 40 Table 3.4: EDS analyses of the zeolite 4A, the alkali pre-treated samples and controls. .................... 40 Table 3.5: Si/Al ratios based on FTIR and EDS analyses. ................................................................ 44 Table 3.6: Uptake, q (mg/g), at 24 h sampling, for each REE tested by Z13X and by modified zeolites with alkali treatment. ........................................................................................................................ 50 Table 3.7: Uptake, q (mg/g), at 24h sampling, for each REE tested by 4A and modified zeolites with alkali treatment. ............................................................................................................................... 50 Table 3.8: Regression parameters for the C/C0 values prediction using Pr (as x) and the other REE (as y). .................................................................................................................................................... 57 Table 3.9: Scoring results for the 24 h adsorption assay regression between the different REE (as y) with Y (as x). ..................................................................................................................................... 59 Table 3.10: Scoring results for the 125 h adsorption assay regression between the different REE (as y) with Pr (as x). ................................................................................................................................... 59 Table 3.11: Scoring results for estimating the C/C0 values for the 24 h assay, using the model trained with the 125 h adsorption assay for the different REE (as y) based on Pr (as x). ................................. 60 Table 3.12: Fitting parameters for PFO for the modified 13 X zeolites and respective controls. ........ 63 Table 3.13: The best kinetic model parameters for different inorganic adsorbents for the same REE.64 Table S-3.1 Statistical differences using the Bonferroni’s multiple comparisons test between the modified zeolite and the controls (Z13X and ZX_H2O). ....................................................................... 69 Table S-3.2: Statistical differences using the Bonferroni’s multiple comparisons test between the modified zeolite and the controls (Z4A and ZA_H2O). ......................................................................... 71 Table S-3.3: Statistical resume for the desorption from Z13X. A multi-comparison test of the results was performed where Y, yes, is used when there is a statistical difference between the conditions tested. N,
| xvii no, means that there is no statistical difference. This test was performed for the different REE. The order of the results is: La, Ce, Y, Tb, Pr and Eu. ......................................................................................... 73 Table S-3.4: Fitting parameters and square errors for Pseudo-First Order, PFO, and Pseudo-Second Order, PSO, models for the selected modified zeolite 13 X and respective controls. ........................... 74 Table S-3.5: Confidence intervals for the parameters of Pseudo-First Order, PFO, and Pseudo-Second Order, PSO, kinetic models for every zeolite and REE tested. ............................................................. 75 Chapter 4 Table 4.1: Column designations for the continuous flow assays ....................................................... 80 Table 4.2: Operational parameters of the assays. Each column had 4 adsorption and desorption cycles, only two of them had a washing step between the desorption and adsorption. ................................... 80 Table 4.3: Binary classification used for each sample evaluated regarding the data from the adsorption (removal) and the desorption (recovery) assays. ................................................................................ 82 Table 4.4: Total removal of each REE for each zeolite tested after 4 cycles. ..................................... 89 Table 4.5: Total recovery percentage of each REE for each zeolite sample tested after 4 cycles. ...... 90 Table S-4.1: Two-Way ANOVA for the total removal percentage of REE after 4 cycles. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step................................................................................................................................................ 106 Table S-4.2: Results for the Two-Way ANOVA for the total recovery percentage after 4 cycles. The NW refers to the assays without the washing step and the WW refers to the assays with the NaOH 0.01 M washing step. ................................................................................................................................. 110 Table S-4.3: Statistical tests performed for all desorption cycles for each REE. The Two-Way ANOVA compares the total REE recovery from the zeolite. A comparison between each cycle for each tested condition was performed. The NW refers to the assays without the washing step and the WW refers to the assays with the NaOH 0.01 M washing step. The cycle with increased acid concentration is referred to as cycle 5. ...................................................................................................................................... 111
| xviii Chapter 5 Table 5.1: Designation and details of REE/Fe-zeolite catalysts and respective method of preparation. ...................................................................................................................................................... 118 Table 5.2: Binary classification used for each REE/Fe-catalyst. ..................................................... 121 Table 5.3: Framework Si/Al ratios obtained from the FTIR analysis. .............................................. 124 Table 5.4: Chemical analysis of the solid REE/Fe-zeolite catalysts. ................................................ 130 Table 5.5: Scores obtained for the different classification algorithms. ............................................ 133 Table 5.6: Classification of catalysts by classifier algorithms for the test set. The results include precision, recall and f1-score. ......................................................................................................... 133 Table 5.7: Kinetic results for the degradation of both dyes using the non-linear (eq. 1) and the linear (eq. 4) forms of the pseudo-first order model equation. ................................................................... 138 Table S-5.1: Conversion results with respective errors for tartrazine (Tar) and Indigo Carmine (IC) for the initial screening assays (IS) and the catalytic tests (CT). Conditions of the reaction: 20 mg of catalyst/25 mL at 30 ppm of Tar; 0.5 mL of 90 mM H2O2; pH=3; T=40 0C and 3 h of reaction for IS and 5 h for CT. The same conditions for the IC for both IS and CT, with the exception of the H2O2 concentration used, which was 12 mM. ................................................................................................................ 142 Table S-5.2: One Way ANOVA results of the comparison with the catalysts based in ZSM5. NaY and NaX zeolites are in the same conditions for the degradation of Tar. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. ............................................................................... 143 Table S-5.3: One Way ANOVA results using Bonferroni’s multiple comparison tests with the REE/FEcatalysts based in NaY or in ZSM5 for the same REE. These tests were performed for both dyes. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. ..................................... 145 Table S-5.4: One Way ANOVA results for the degradation comparing the different REE concentrations of the starting solutions tested. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. ........................................................................................................................................ 146 Table S-5.5: One Way ANOVA results for the degradation comparing the different REE for the same support and the same REE concentration. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. ....................................................................................................................... 147 Table S-5.6: One Way ANOVA results for the degradation comparing the different preparation methods for La on both supports. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. ................................................................................................................................................ 148
| xix Table S-5.7: Two Way ANOVA results from the comparison between the different volumes (0.5 or 5 mL) of H2O2. The H2O2 concentration for the Tar degradation was 90 mM, while for IC was 12 mM. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. ........................................ 148 Table S-5.8: Two Way ANOVA results from the comparisons between the control and the catalysts for the same volume (5 mL) of H2O2. The H2O2 concentration for the Tar degradation was 90 mM, while for IC was 12 mM. The 95 % confidence interval of the difference is also included as 95.00% CI of diff. 148
| xx List of Abbreviations and Acronyms A A3 – catalyst La10Fe10NaX, FAU (NaX in powder) with the metals added via impregnation method A7 – catalyst La10Fe10NaX, FAU (NaX in pellet) with the metals added via impregnation method AOP – Advanced Oxidation Processes ATR-FTIR Attenuated Total Reflectance Fourier Transform Infrared spectroscopy B BET – Brunauer-Emmett-Teller C C/C0 out – C/C0 concentration of REE that leave the continuous flow assays C0 – initial concentration (mg/L) of the REE Conc. – concentration CT – Catalytic Tests Ct – concentration (mg/L) of REE at a time (t) D dH2O – distilled water (dH2O) DVD – Digital Versatile Disc E E a – activation energy EU – European Union (EU) F FAU – Faujasite structure FCC – Fluid Cracking Catalysis FTIR – Fourier-transform infrared spectroscopy H HDD – hard disk drives HREE – heavy REE I ICP-AES – Inductively Coupled Plasma Atomic Emission Spectroscopy ICP-OES – Inductively Coupled Plasma – Optical Emission Spectrometry IS – Initial Screening IUPAC – International Union of Pure and Applied Chemistry K k1 – rate constant (min−1) for PFO k2 – parameter related to the initial concentration of solute (g/(mg × min)) ka – rate constants of adsorption kd – rate constants of desorption KNN – K-nearest neighbors Classifier L L.E.D. – light-emitting diodes LREE – light REE LTA – Linde Type A structure M m – the mass (g) of the adsorbent used. MAE – Mean absolute error MFI – Mobil-type five structure ML – Machine learning MSE – Mean squared error N
| xxi NW – without washing with NaOH 0.01 M – between the desorption and adsorption cycles P PCA – Principal Component Analysis PFO – Pseudo-first order pH_assay – pH value of the adsorption solution pH_treatment – pH value of the treatment solution pHfinal – final pH pHinitial – initial pH pHPZC – pH of zero point charge ppm – parts per million PSO – Pseudo-second order Q q – uptake, mass of adsorbate per unit mass of adsorbent (mg/g) qe – mass of adsorbate per unit mass of adsorbent at equilibrium (mg/g) qt – mass of solute retained per mass of solid at time (t) R Rc – recovery percentage during the continuous flow assays REE – Rare Earth Elements REE2O3 or REEOrare earth oxides Rm – removal percentage during the continuous flow assays RMSE – Root mean squared error S SEM-EDS – Scanning Electron Microscopy/Energy Dispersive X-Ray spectroscopy SSR – Sum of Squares due to Regression Sy.x – Standard Deviation of the Residuals V V0 – the initial volume (L) of the REE solution Vt – volume (L) of the solution at a given time (t) W WW – with washing with NaOH 0.01 M between the desorption and adsorption cycles X XPS – X-ray Photoelectron Spectroscopy XRD – X-ray diffraction Z Z1 – catalyst La10Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via impregnation method Z10 – catalyst Pr10Fe10NaY, FAU (NaY in powder) with the metals added via ion exchange method Z11 – catalyst Pr25Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via ion exchange method Z12 – catalyst Pr25Fe10NaY, FAU (NaY in powder) with the metals added via ion exchange method Z13X – zeolite 13X without modification (control for FAU 13X)
| xxii Z13X_NW – zeolite 13X with no chemical modification and without washing with NaOH 0.01 M between the desorption and adsorption cycles Z13X_WW – zeolite 13X with no chemical modification and with washing with NaOH 0.01 M between the desorption and adsorption cycles Z15 – catalyst La10Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via ion exchange method Z16 – catalyst La25Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via ion exchange method Z17 – catalyst La25Fe10NaY, FAU (NaY in powder) with the metals added via ion exchange method Z2 – catalyst La10Fe10NaY, FAU (NaY in powder) with the metals added via impregnation method Z3 – catalyst La25Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via impregnation method Z4 – catalyst La25Fe10NaY, FAU (NaY in powder) with the metals added via impregnation method Z4A – zeolite 4A without modification (control for LTA 4A) Z5 – catalyst Ce10Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via ion exchange method Z6 – catalyst Ce10Fe10NaY, FAU (NaY in powder) with the metals added via ion exchange method Z7 – catalyst Ce25Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via ion exchange method Z8 – catalyst Ce25Fe10NaY, FAU (NaY in powder) with the metals added via ion exchange method Z9 – catalyst Pr10Fe10ZSM5, MFI (ZSM5 in powder) with the metals added via ion exchange method ZA _H2O – zeolite 4A only washed with H2O (control for LTA 4A) ZA_KOH 0.10 M – zeolite 4A modified with KOH solution at 0.10 M ZA_KOH 0.25 M – zeolite 4A modified with KOH solution at 0.25 M ZA_KOH 0.50 M – zeolite 4A modified with KOH solution at 0.50 M ZA_NaOH 0.10 M – zeolite 4A modified with NaOH solution at 0.10 M ZA_NaOH 0.25 M – zeolite 4A modified with NaOH solution at 0.25 M ZA_NaOH 0.50 M – zeolite 4A modified with NaOH solution at 0.50 M ZNaOH_NW – zeolite 13X with chemical modification with NaOH 0.1 M and without washing with NaOH 0.01 M between the desorption and adsorption cycles ZNaOH_WW – zeolite 13X with chemical modification with NaOH 0.1 M and with
| xxiii washing with NaOH 0.01 M between the desorption and adsorption cycles ZSM5 – Zeolite Socony Mobil-type 5 ZX_H2O – zeolite 13X only washed with H2O (control for FAU 13X) ZX_H2SO4 0.25 M – zeolite 13X modified with H2SO4 solution at 0.25 M ZX_HCl 0.25 M – zeolite 13X modified with HCl solution at 0.25 M ZX_HNO3 0.25 M – zeolite 13X modified with HNO3 solution at 0.25 M ZX_KOH 0.10 M – zeolite 13X modified with KOH solution at 0.10 M ZX_KOH 0.25 M – zeolite 13X modified with KOH solution at 0.25 M ZX_KOH 0.50 M – zeolite 13X modified with KOH solution at 0.50 M ZX_NaOH – zeolite 13X modified with NaOH solution at 0.10 M for the continuous flow assays ZX_NaOH 0.10 M – zeolite 13X modified with NaOH solution at 0.10 M ZX_NaOH 0.25 M – zeolite 13X modified with NaOH solution at 0.25 M ZX_NaOH 0.50 M – zeolite 13X modified with NaOH solution at 0.50 M
Chapter 1 Motivation and Outline Barros, O. | 2023 | 6 ➢ Assila, O.; Barros, O.; Zerrouq, F.; Kherbeche, A.; Fonseca, A. M. F.; Parpot, P.; Tavares, T.; Neves, I. C.; Optimization of Fenton-type reaction for water treatment using bimetallic catalysts based in porous materials by Box-Behnken Design , at XVth International Symposium on Environment, Catalysis and Process Engineering, 23 to 25 November 2021, Morocco; ➢ Assila, O.; Barros, O.; Zerrouq, F.; Kherbeche, A.; Fonseca, A. M. F.; Parpot, P.; Tavares, T.; Neves, I. C.; Box-Behnken Desing for optimization of Fenton-type reaction for water treatment using heterogeneous catalysts , XXVII Encontro Nacional da Sociedade Portuguesa de Química, 14 to 16 July 2021, Braga, Portugal. Poster communications in scientific meetings ➢ Barros, O.; Assila, O.; Neves, I. C.; Tavares, T.; Comparison of the catalytic behaviour of rare earth elements loaded in zeolites as heterogeneous catalysts, XI National Meeting on Catalysis and Porous Materials (XI ENCMP) and the II Meeting of the Carbon Group (II RGC), 9 and 10 December 2021, Aveiro, Portugal. 1.4. References 1. American Chemical Society The Periodic Table of Endangered Elements - American Chemical Society Available online: https://www.acs.org/greenchemistry/research-innovation/endangeredelements.html (accessed on Jul 1, 2023). 2. Church, C.; Crawford, A. Green Conflict Minerals: The fuels of conflict in the transition to a lowcarbon economy. IISD Report ; 2018; 3. IEA Global EV Outlook 2019 - Análise - IEA ; 2019; 4. United Nations Climate Change Conference Paris Declaration on Electro-Mobility and Climate Change & Call to Action Available online: https://unfccc.int/news/the-paris-declaration-on-electromobility-and-climate-change-and-call-to-action (accessed on Jul 1, 2023). 5. Ballinger, B.; Stringer, M.; Schmeda-Lopez, D.R.; Kefford, B.; Parkinson, B.; Greig, C.; Smart, S. The vulnerability of electric vehicle deployment to critical mineral supply. Appl. Energy 2019, 255 , 113844. 6. Lèbre, É.; Stringer, M.; Svobodova, K.; Owen, J.R.; Kemp, D.; Côte, C.; Arratia-Solar, A.; Valenta, R.K. The social and environmental complexities of extracting energy transition metals. Nat. Commun. 2020, 11 , 4823. 7. Visual Capitalist All the Metals We Mined in 2021: Visualized - Visual Capitalist Available online: https://www.mining.com/web/all-the-metals-we-mined-in-2021-visualized/ (accessed on Jul 1, 2023). 8. Zheng, B.; Fan, J.; Chen, B.; Qin, X.; Wang, J.; Wang, F.; Deng, R.; Liu, X. Rare-Earth Doping in Nanostructured Inorganic Materials. Chem. Rev. 2022, 122 , 5519–5603. 9. Barrett, S.D.; Dhesi, S.S. The Structure of Rare-Earth Metal Surfaces ; PUBLISHED BY IMPERIAL COLLEGE PRESS AND DISTRIBUTED BY WORLD SCIENTIFIC PUBLISHING CO., 2001; ISBN 9781-86094-165-8. 10. UN Sustainable Development Solutions Network (UNSDSN) Mapping Mining to the Sustainable Development Goals: An Atlas ; 2016; 11. Pérez-Botella, E.; Valencia, S.; Rey, F. Zeolites in Adsorption Processes: State of the Art and Future Prospects. Chem. Rev. 2022, 122 , 17647–17695.
| 7 2. Chapter 2 – Literature Review This chapter presents an overview of the Rare Earth Elements, especially in what concerns their characteristics, processing, applications and the importance of recovering and recycling these elements. The definition, characteristics, importance and applications of zeolites will be discussed. The concepts and importance underlining adsorption processes will be explained. An overview on heterogeneous catalysis and on Fenton-type reactions will be forwarded. Finally, the application of machine learning techniques to adsorption and to catalysis will be reviewed to understand how these algorithms can benefit for the selection of the best outcomes.
Chapter 2 Literature Review Barros, O. | 2023 | 8 2.1. Rare Earth Elements The rare earth group includes seventeen elements (lanthanide group of the periodic table with scandium (Sc) and yttrium (Y)) as defined by the International Union of Pure and Applied Chemistry (IUPAC). The fifteen lanthanides are the elements with the atomic numbers from 57 to 71, being listed as: lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), promethium (Pm), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dysprosium (Dy), holmium (Ho), erbium (Er), thulium (Tm), ytterbium (Yb) and lutetium (Lu). These elements are divided into light REE (LREE) and heavy REE (HREE), with Z of the LREE ranging from 57 to 64,while the atomic number of the HREE goes from 65 to 71, including Y [1,2]. The main difference between the HREE and the LREE is related to the paired electrons (both clockwise and counter-clockwise spinning electrons) present in the HREE [3]. Yttrium is included in the HREE since this element presents some similarities, both in ionic radius and chemical properties with that group. Sometimes the REE are divided into three categories: (i) light REE - La, Ce, Pr, and Nd; (ii) medium REE – Sm, Eu, and Gd and (iii) heavy REE - Tb, Dy, Ho, Er, Tm, Yb, Y and Lu [4–6]. This work will consider the REE division into light and heavy. Sc is considered an REE as it was discovered simultaneously with other REE and its characteristics resemble more the ones of Y rather than those of Al or Ti [6]. However, this element is not considered a LREE or a HREE [3,7]. Pm is considered a REE as it is the last element of the lanthanide group to assume double c-axis hcp ( dhcp ) structure at ambient conditions [8]. This element is the only one that does not occur in nature due to its radioactivity instability [9,10]. Understanding the REE and their various applications, which will be summarized in section 2.1.1 REE applications, will be more straightforward when the characteristics of these elements are known. Diverse studies and reviews have been published to explain the versatility of REE [3,7,11]. The REE possess different oxidation states, although the most common oxidation is 3+, as it is more stable, and due to that, most of the REE are represented as REE2O3 [3,12]. A critical characteristic of all metals is their precipitation pH, mainly for purification purposes, and the REE are no exception. It is essential to understand that the REE precipitation may occur at different pH, depending on the anion present in the solution and its concentration. The work of Han [13] shows that it is possible to induce REE precipitation even at low pH values by using different precipitants. Some examples of REE precipitation described in the literature use carbonate [14,15], fluoride [16,17], phosphate [18,19], sulfate [20–24] and oxalate [7,25–28].
Chapter 2 Literature Review Barros, O. | 2023 | 9 The REE are elements that do not exist as individual native metals such as gold (Au), copper (Cu) or silver (Ag), due to their reactivity. Instead, they occur together in diverse minerals as minor or major constituents [29]. REE are found in a great variety of minerals, over 250 minerals [30], although the primary sources of REE of economic interest are bastnaesite, monazite, loparite and the lateritic ionadsorption clays [29]. The bastnaesite ore is found in China and the United States and constitutes the most significant percentage of the world’s REE source. In contrast, the monazite ore is found in countries such as Australia, Brazil, China, India, Malaysia, South Africa, Sri Lanka, Thailand and the United States. It constitutes the second largest deposit of REE [31,32]. Although with minor impact , other sources of REE are apatite, cheralite, eudialyte, loparite, phosphorites, rare-earth-bearing (ion adsorption) clays, secondary monazite, spent uranium solutions and xenotime [33]. Apatite ore does not present a high concentration, varying from 0.1 % to 1 %, of rare earth oxides (REE2O3 or designed as REEO) [34–36]. However, this ore is abundant and is found worldwide, making it an essential source of REE [37,38]. The work by Liu et al. [39] summaries the updated state regarding the different REE deposits, active mines and 146 projects at an advanced stage, shown in Figure 2.1. Figure 2.1: Current global distribution of REE projects, including active mines and advanced projects. The REE deposit types are represented by various colors accordingly. Active mines ( ) and advanced projects ( ) are also marked. Adapted from Liu et al. [39]. Over the most recent years, the REE production suffered enormous growth, from 1.814 kilotonnes (kt) before 1960 to 112.491 kt in 2010, with an average growth rate of around 10 % per year [40,41]. The global production of rare earth oxide (REO) was 263.054 kt in 2021 and 272.155 kt in 2022
Chapter 2 Literature Review Barros, O. | 2023 | 10 [42]. The global REE market reached 9.5 billion US $ in 2022 [43], with a global consumption of REE oxides around 164 kt [44]. China is the major REE producer, followed by the United States, Australia and Burma, as the percentage of the total production from 2022 [42] reveals in Figure 2.2. Figure 2.2: World production of REE in 2022. Adapted from the U.S. Geological Survey 2023 [42]. The REE purification from ores involves significant energy and resource consumption, high pollution levels and relevant environmental impacts. The REE processing is much more complex than the production of other metals [45,46], due to their similar reactivity. Adding to that, REE production is associated with radioactive pollution and toxicity due to the presence of radioactive elements in the REE deposits like uranium (U) or thorium (Th) [47]. For these reasons, REE production is attracting more attention globally [48,49]. For the extraction of 0.907 t of REE in Bayan Obo in Mongolia, it is required 4.001 t of sulfuric acid (H2SO4), 11.177 t of sodium chloride (NaCl), 1.488 t of sodium hydroxide (NaOH), 1.061 t of hydrochloric acid (HCl), 1.724 t of water and the grinding of almost 45.359 t of mineral ore [50,51]. Additionally, the energy required to obtain 1 ton of an individual REE varies from 38 to 48 GJ, except for Sc and Y, which requires 148 or 75 GJ per ton, respectively [50]. In 2011, REE were sold at record high prices, which had driven some countries to re-open mines, as the Mountain Pass mine in California re-opened in 2012 [52]. Some countries like Japan and most of
Chapter 2 Literature Review Barros, O. | 2023 | 11 the European Union (EU) do not have any REE deposits, so they have to invest in developing alternative sources of REE [52]. Kato and co-workers reported a rich REE mud around Minamitorishima Island in 2011 [53] and further studies verified that the referred mud had almost 8000 ppm of total REE , corresponding to more than 14.515 million tonnes (Mt) of REEO in the studied area [54]. In 2021, Ohta et al. [55] reported a rich REE mud with contents between 2000 and 4500 ppm in the Central Pacific Basin. Although the mud presents high concentrations of REE, it will be a challenge to isolate them. The challenge begins with collecting, concentrating and transporting the mud from the seabed by the lowest cost possible. New technologies are required as well as the adaptation of other ones to process this deep mud and to recover the valuable REE. Another possible source of REE is coal, mainly coal ash or coal-based materials [56–58]. Coal ash is reported to eventually release toxic elements, such as As, Se, Cr and Cd by naturally leaching from the ash deposit sites, which is harmful to the environment and to human health [59]. Some critical metals can also be found in coal like Ge, Ga, Nb, Zr, V, Re, Au and base metals such as Al [59–61]. Luttrel et al. [62] reported that more than 80 % of REE associated with the run-of-mine coals are refused streams after coal preparation. Therefore, the coal refuse or even the coal ash could be a more suitable REE source as higher concentrations of REE mean a more cost-effective and efficient recovery. However, it is essential to note that the feasibility of using such a source depends on various factors, such as the actual concentration of REE in the refuse, the efficiency of extraction methods, environmental considerations and economic viability. Some developments have been reported such as the methods used by Wang et al. [59] to recover REE from coal ash. While for coal refuse, physical and hydrometallurgical approaches usually provide low recoveries [57], a previous calcination step before the acid leaching has encouraging results [63,64]. 2.1.1. REE applications REE are a group of chemical elements with a huge application potential and this is related to their diversified properties of interest, which vary from chemical, optical, electrical, metallurgical and catalytical to magnetic [42,65–67]. The diverse electronic, manufacturing, and technological applications of the properties of the REE show that these elements are present in a significant part of the products that surround us in our daily lives. Due to their importance in diverse areas, several international institutions and governments coded REE as critical materials [1,29,68,69]. REE do have diverse applications such as in metal alloys, lasers, lighter flints, batteries, fluorescent materials (lamps), information storage (Digital Versatile Discs or computer memories),
Chapter 2 Literature Review Barros, O. | 2023 | 12 catalysts, magnets and super magnets, glass additives, military technologies, transport of energy, phosphors, mobile phones, medical applications (nanomedicine, imaging), light-emitting diodes (L.E.D.), light bulbs, high-temperature superconductors, renewable energy sources (solar panels, wind turbines), polishing compounds, military and aerospace systems [29,42,70–72]. The primary applications of these elements in 2021 are shown in Figure 2.3. Figure 2.3: Distribution of rare earth element consumption in 2021 (adapted from Statista [73]). 2.1.2. Recovery and recycling importance of REE REE are relevant for different applications and are nonrenewable strategic resources [74]. The undirected development in recent years has caused a severe depletion of REE and irreversible damage to such resources since the supply of these elements is widening year by year [75]. Therefore, recycling the REE from urban or industrial waste is critical to moving from a linear economy to a circular economy [72]. However, the recycling of REE is quite complex due to the number of such elements and the quantities of those elements (that range from mg to several kgs) present in the end-products [72]. Combining the varying amounts with the complexity of their uses, the inherent difficulty of separating individual REE, the possible extended life usage of specific applications, it is justifiable that less than 1 % of the REE used today is recycled [72,76]. Some relevant efforts have been made to develop new REE recycling methods [77–79]. Over the last years, extensive efforts have been made regarding the recycling
Chapter 2 Literature Review Barros, O. | 2023 | 13 of fluorescent lamps [77,80], of NdFeB magnets from hard disk drives (HDD) [24,81], from wind turbines [82,83] or from magnets production waste [84]. One of the most important secondary sources is electronic waste, e-waste [85]. This waste consists of electric and electronic devices (equipment that required electricity to work) which were discarded at the end of their economic life span when no longer used by consumers [86,87]. Those devices include computers, audio-visual and communication equipment [88,89]. In 2016, each inhabitant produced, on average, 6.1 kg of this waste, a total of 44.7 Mt worldwide and it was estimated to reach 52.2 Mt in 2021 [90]. In 2016, it was estimated that this waste was worth around 55 billion euros [91] and the European market for REE recycling could be worth 1 billion euros [92]. Some factors play a crucial role in the growth of this waste such as the decreasing economic lifespan of electronic devices [93], lack of international consensus on e-waste management [94] and inadequate use awareness [87]. One big challenge for its disposal is its quantification since there is no appropriate waste tracking systems. As an alternative to disposal, e-waste could be redirected for material recycling and reprocessing [95], resulting in new raw materials able to substitute primary materials, more commonly designed as virgin materials obtained from ores. This idea may help to construct a new economy based on circular zero-waste [96]. One important concept is the technosphere, a material stockpile established by human activities and technological processes where different metals and alloys can be found [97]. These metal deposits changed the mining perspective since some metals can be recycled from previous applications and used in new ones [97]. This idea will help to supply the markets with the needed metals, as technosphere mining is a secondary source of many metals, promoting a circular economy and reducing the amount of end-of-life products in e-waste. Jowitt et al. [98] state that the recycling of REE could be divided into different types, namely direct recycling of manufacturing scrap or residues, recycling of solid and liquid industrial wastes and urban mining or technosphere mining of end-of-life products also in accordance to the work of Li et al. [99]. Technosphere mining can be applied for recycling REE from end-of-life products such as permanent magnets, lamp phosphors from fluorescent lamps, batteries, urban solid waste, stocks of landfilled industrial processes residues such as mine tailing bauxite residue, REE catalysts used within the chemical industry and some others [52,98,100–104]. One big obstacle to recycling REE from endof-life products is the vestigial elemental amounts present in most of them. On the other hand, significant amounts of those products could raise the total quantity of REE potentially recovered. Other problems
Chapter 2 Literature Review Barros, O. | 2023 | 14 have held back the recycling of REE up to date such as the difficulty of collecting, extracting and recovering [98]. With correct e-waste management, as summarized in Figure 2.4, it could be possible to perform a technosphere mining of REE, in other words, the REE recycling. On the other hand, tracking the amounts of e-waste produced and recycled could enhance that recycling as it is. However, new recycling methods with good recovery yields even starting from low concentrations, high selectivity, economic, non-toxic and not harmful to human health and to the environment are needed. Diverse processes have been studied and used. However, some of them require chemicals to allow a high recovery capacity, which will lead to a growth of waste while the idea is to reduce it as much as possible. The different sorption processes may be a possible solution for this problem. The focus of this thesis is the recovery of the soluble REE using adsorption techniques, being some of the main reasons for the possible use of this technique with lower concentration of the pollutant and other characteristics that will be explored in 2.3 Sorption Processes. Figure 2.4: REE life cycle: REE are obtained from ores and transformed into different applications (blue), resulting in waste. In green, it is possible to get REE through recycling the e-waste by technosphere mining and re-use those elements into new applications. One main problem after the solubilization of the REE will be their separation from the solution, overcame by precipitation with specific anions. However, as explained in the section 2.1 Rare Earth Elements, precipitation depends on several factors to be successful and economically viable.
Chapter 2 Literature Review Barros, O. | 2023 | 15 2.2. Zeolites Zeolites are crystalline microporous aluminosilicates solids [105], which can be used for incorporating within their structure different cations from a solution by ion exchange reaction. Zeolites which belong to the tectosilicate-type minerals and some of them occur naturally, present a high specific surface area, and are used as catalysts, adsorbents, ion-exchangers, and other aplications [105–107]. The variation of the synthesis conditions, use of more reactive silica sources and more alkaline media led to the obtention of zeolites A and X and 14 new zeolite materials. In this work, it was used MFI, Mobil-type five (Zeolite Socony Mobil-type 5, ZSM5), faujasite, FAU (Y and X) and Linde Type A, LTA (4A) structures. Therefore, these zeolite structures will be the focus and the respective structures are shown in Figure 2.5. Figure 2.5: Front view of the structure of the selected zeolites for this work. Each zeolite has the official name given to each structure by the International Zeolite Association, while their common name is shown in parentheses, adapted from [105]. Zeolites have a three-dimensional structure of SiO4 and AlO4 tetrahedras, in which replacing the Si4+ with Al3+ give the zeolite a negative charge [108]. This negative charge justifies the strong affinity for cations, such as transition metals, and a little affinity for anions and non-polar organic molecules [108,109]. The zeolites can be classified according to their Si/Al ratio as low-silica (Si/Al ratios below 2, highly polar), medium-silica zeolites (Si/Al ratio between 2 and 5, intermediate polarity) and high-silica zeolites (Si/Al ratios above 5) [110]. Also, by the size of the pore diameters [110–112]: ➢ Small pores: between 3 and 4 Å, corresponding to 8 or 9 oxygen rings. ➢ Medium pore: between 5 and 6 Å, corresponding to 10 oxygen rings. ➢ Large pore: between 6 and 7.5 Å, corresponding to 12 oxygen rings. ➢ Extra-large: above 7.5 Å, corresponding to more than 12 oxygen rings.
Chapter 2 Literature Review Barros, O. | 2023 | 22 ➢ Clustering algorithms: it is the classification method of objects into different groups. ➢ Dimensionality Reduction algorithms: aims to remove irrelevant and redundant data to reduce the computational cost and improve data quality for efficient data organization strategies. Figure 2.7: Representation of the classification of all ML algorithms (adapted from [170]). 2.5.2. ML applications ML has been successfully applied in diverse fields such as pattern recognition, medicine, science, computer vision, spacecraft engineering, engineering, biomedicine, psychology, catalysis, neurobiology and many other disciplines [166,169,171]. This broad application allows a faster treatment of tremendous amounts of data since ML is able to analyze and correlate those data to achieve better interpretations and, therefore, to take better decisions. ML models and their importance have been recognized and appreciated in wastewater treatment [172,173]. Some developments were achieved in ML algorithms or deep learning neural networks for the optimization of the adsorption of antibiotics [174,175], of organic compounds [176,177] and of metals [178–180]. The capacity of ML algorithms to evaluate large amounts of data from catalytic reactions and catalyst characterization can help to design the best catalyst for a given reaction [181,182]. Various homogeneous and heterogeneous catalytic applications have been using ML [183].
Chapter 2 Literature Review Barros, O. | 2023 | 23 2.6. References 1. Anastopoulos, I.; Bhatnagar, A.; Lima, E.C. Adsorption of rare earth metals: A review of recent literature. J. Mol. Liq. 2016, 221 , 954–962. 2. Moldoveanu, G.A.; Papangelakis, V.G. Recovery of rare earth elements adsorbed on clay minerals: I. Desorption mechanism. Hydrometallurgy 2012, 117 – 118 , 71–78. 3. Voncken, J.H.L. Physical and Chemical Properties of the Rare Earths. In The Rare Earth Elements - An Introduction ; Springer, Cham, 2016; pp. 53–72 ISBN 978-3-319-26809-5. 4. Xiao, Y.; Huang, L.; Long, Z.; Feng, Z.; Wang, L. Adsorption ability of rare earth elements on clay minerals and its practical performance. J. Rare Earths 2016. 5. CHEN, Z. Global rare earth resources and scenarios of future rare earth industry. J. Rare Earths 2011, 29 , 1–6. 6. Gupta, C.K.; Krishnamurthy, N. Extractive Metallurgy of Rare Earths ; Taylor & Francis: Abingdon, UK, 2004; ISBN 978-0-203-68517-4. 7. Ganguli, R.; Cook, D.R. Rare earths: A review of the landscape. MRS Energy Sustain. 2018, 5 , 6. 8. Samudrala, G.K.; Vohra, Y.K. Structural Properties of Lanthanides at Ultra High Pressure. In Including Actinides ; Bünzli, J.-C.G., Pecharsky, V.K.B.T.-H. on the P. and C. of R.E., Eds.; Elsevier, 2013; Vol. 43, pp. 275–319 ISBN 0168-1273. 9. Lambert, C.E.; Ledrich, M.-L. Lanthanide Series of Metals. In Encyclopedia of Toxicology ; Wexler, P.B.T.-E. of T. (Third E., Ed.; Elsevier: Oxford, 2014; pp. 43–47 ISBN 978-0-12-386455-0. 10. Zhang, J.; Liu, Q.; He, Q.; Nozaki, Y. Rare Earth Elements and Their Isotopes in the Ocean. In Encyclopedia of Ocean Sciences ; Cochran, J.K., Bokuniewicz, H.J., Yager, P.L.B.T.-E. of O.S. (Third E., Eds.; Elsevier: Oxford, 2019; pp. 181–197 ISBN 978-0-12-813082-7. 11. Barrett, S.D.; Dhesi, S.S. The Structure of Rare-Earth Metal Surfaces ; PUBLISHED BY IMPERIAL COLLEGE PRESS AND DISTRIBUTED BY WORLD SCIENTIFIC PUBLISHING CO., 2001; ISBN 9781-86094-165-8. 12. Choppin, G.R.; Rizkalla, E.N. Chapter 128 Solution chemistry of actinides and lanthanides. In Handbook on the Physics and Chemistry of Rare Earths ; 1994; pp. 559–590. 13. Han, K.N. Characteristics of Precipitation of Rare Earth Elements with Various Precipitants. Minerals 2020, 10 , 178. 14. de Vasconcellos, M.E.; da Rocha, S.M.R.; Pedreira, W.R.; Queiroz, C.A. da S.; Abrão, A. Solubility behavior of rare earths with ammonium carbonate and ammonium carbonate plus ammonium hydroxide: Precipitation of their peroxicarbonates. J. Alloys Compd. 2008, 451 , 426–428. 15. Kim, P.; Anderko, A.; Navrotsky, A.; Riman, R.E. Trends in Structure and Thermodynamic Properties of Normal Rare Earth Carbonates and Rare Earth Hydroxycarbonates. Minerals 2018, 8 . 16. Tokunaga, S.; Haron, M.J.; Wasay, S.A.; Wong, K.F.; Laosangthum, K.; Uchiumi, A. Removal of fluoride ions from aqueous solutions by multivalent metal compounds. Int. J. Environ. Stud. 1995, 48 , 17–28. 17. Yan, C.; Jia, J.; Liao, C.; Wu, S.; Xu, G. Rare Earth Separation in China. Tsinghua Sci. Technol. 2006, 11 , 241–247. 18. Wu, S.; Zhao, L.; Wang, L.; Huang, X.; Zhang, Y.; Feng, Z.; Cui, D. Simultaneous recovery of rare earth elements and phosphorus from phosphate rock by phosphoric acid leaching and selective precipitation: Towards green process. J. Rare Earths 2019, 37 , 652–658. 19. Zakharova, B.; Komissarova, L.; Traskin, V.; Naumov, S.; Melnikov, P. Precipitation of Rare Earth Phosphates from H3PO4 Solutions. Phosphorus. Sulfur. Silicon Relat. Elem. 1996, 111 , 2. 20. Moeller, T.; Kremers, H.E. Observations on Rare Earths Double Sodium Sulfate Precipitation for Separation of the Terbium and Yttrium Earths. Ind. Eng. Chem. Anal. Ed. 1945, 17 , 44–45.
Chapter 2 Literature Review Barros, O. | 2023 | 24 21. Vijayalakshmi, R.; Mishra, S.L.; Singh, H.; Gupta, C.K. Processing of xenotime concentrate by sulphuric acid digestion and selective thorium precipitation for separation of rare earths. Hydrometallurgy 2001, 61 , 75–80. 22. Han, K.N. Effect of Anions on the Solubility of Rare Earth Element-Bearing Minerals in Acids. Mining, Metall. Explor. 2019, 36 , 215–225. 23. Han, K.N. Effect of Metal Complexation on the Solubility of Rare Earth Compounds. In Critical and Rare Earth Elements ; CRC Press, 2019; pp. 59–84 ISBN 9780429023545. 24. Ciro, E.; Alzate, A.; López, E.; Serna, C.; Gonzalez, O. Neodymium recovery from scrap magnet using ammonium persulfate. Hydrometallurgy 2019, 186 , 226–234. 25. Gupta, C.K.; Krishnamurthy, N. Extractive metallurgy of rare earths. Int. Mater. Rev. 1992, 37 , 197–248. 26. Chi, R.; Xu, Z. A solution chemistry approach to the study of rare earth element precipitation by oxalic acid. Metall. Mater. Trans. B 1999, 30 , 189–195. 27. Khawassek, Y.M.; Eliwa, A.A.; Gawad, E.A.; Abdo, S.M. Recovery of rare earth elements from ElSela effluent solutions. J. Radiat. Res. Appl. Sci. 2015, 8 , 583–589. 28. He, L.; Xu, Q.; Li, W.; Dong, Q.; Sun, W. One-step separation and recovery of rare earth and iron from NdFeB slurry via phosphoric acid leaching. J. Rare Earths 2022, 40 , 338–344. 29. Balaram, V. Rare earth elements: A review of applications, occurrence, exploration, analysis, recycling, and environmental impact. Geosci. Front. 2019, 10 , 1285–1303. 30. Dostal, J. Rare Earth Element Deposits of Alkaline Igneous Rocks. Resources 2017, 6 . 31. King, H. REE - Rare Earth Elements - Metals, Minerals, Mining, Uses Available online: https://geology.com/articles/rare-earth-elements/ (accessed on Apr 27, 2020). 32. Gholz, E. Rare Earth Elements and National Security. Energy Rep. 2014, 1–20. 33. King, H. REE - Rare Earth Elements - Metals, Minerals, Mining, Uses ; 2016; 34. Habashi, F. The recovery of the lanthanides from phosphate rock. J. Chem. Technol. Biotechnol. Chem. Technol. 1985, 35 , 5–14. 35. Preston, J.S.; Cole, P.M.; Craig, W.M.; Feather, A.M. The recovery of rare earth oxides from a phosphoric acid by-product. Part 1: Leaching of rare earth values and recovery of a mixed rare earth oxide by solvent extraction. Hydrometallurgy 1996. 36. Ihlen, P.M.; Schiellerup, H.; Gautneb, H.; Skår, Ø. Characterization of apatite resources in Norway and their REE potential - A review. Ore Geol. Rev. 2014. 37. U.S. Geological Survey Mineral Commodity Summaries ; 2019; Vol. 3;. 38. Emsbo, P.; McLaughlin, P.I.; Breit, G.N.; du Bray, E.A.; Koenig, A.E. Rare earth elements in sedimentary phosphate deposits: Solution to the global REE crisis? Gondwana Res. 2015. 39. Liu, S.-L.; Fan, H.-R.; Liu, X.; Meng, J.; Butcher, A.R.; Yann, L.; Yang, K.-F.; Li, X.-C. Global rare earth elements projects: New developments and supply chains. Ore Geol. Rev. 2023, 157 , 105428. 40. Gschneidner Jr., K.A. The rare earth crisis – the supply/demand situation for 2010-2015 ; 2011; Vol. 6;. 41. Voncken, J.H.L. The Rare Earth Elements ; SpringerBriefs in Earth Sciences; Springer International Publishing: Cham, 2016; Vol. 53; ISBN 978-3-319-26807-1. 42. U.S. Geological Survey Mineral commodity summaries 2023 ; 2023; 43. IMARC Group Rare Earth Elements Market Share, Size & Forecast 2028 | ; 44. Supply and Demand - Arafura Available online: https://www.arultd.com/products/supply-anddemand/ (accessed on Aug 8, 2023). 45. Khan, A.M.; Bakar, N.K.A.; Bakar, A.F.A.; Ashraf, M.A. Chemical speciation and bioavailability of rare earth elements (REEs) in the ecosystem: a review. Environ. Sci. Pollut. Res. 2017, 24 , 22764–22789.
Chapter 2 Literature Review Barros, O. | 2023 | 25 46. Yin, X.; Martineau, C.; Demers, I.; Basiliko, N.; Fenton, N.J. The potential environmental risks associated with the development of rare earth element production in Canada. Environ. Rev. 2021, 29 , 354–377. 47. Cardoso, C.E.D.; Almeida, J.C.; Lopes, C.B.; Trindade, T.; Vale, C.; Pereira, E. Recovery of rare earth elements by carbon-based nanomaterials—a review. Nanomaterials 2019. 48. Gwenzi, W.; Mangori, L.; Danha, C.; Chaukura, N.; Dunjana, N.; Sanganyado, E. Sources, behaviour, and environmental and human health risks of high-technology rare earth elements as emerging contaminants. Sci. Total Environ. 2018, 636 , 299–313. 49. Ma, L.; Dang, D.H.; Wang, W.; Evans, R.D.; Wang, W.-X. Rare earth elements in the Pearl River Delta of China: Potential impacts of the REE industry on water, suspended particles and oysters. Environ. Pollut. 2019, 244 , 190–201. 50. Talens Peiró, L.; Villalba Méndez, G. Material and Energy Requirement for Rare Earth Production. JOM 2013, 65 , 1327–1340. 51. Oladipo, H.J.; Tajudeen, Y.A.; Taiwo, E.O.; Muili, A.O.; Yusuf, R.O.; Jimoh, S.A.; Oladipo, M.K.; Oladunjoye, I.O.; Egbewande, O.M.; Sodiq, Y.I.; et al. Global Environmental Health Impacts of Rare Earth Metals: Insights for Research and Policy Making in Africa. Challenges 2023, 14 , 20. 52. Binnemans, K.; Jones, P.T.; Blanpain, B.; Van Gerven, T.; Pontikes, Y. Towards zero-waste valorisation of rare-earth-containing industrial process residues: a critical review. J. Clean. Prod. 2015, 99 , 17–38. 53. Kato, Y.; Fujinaga, K.; Nakamura, K.; Takaya, Y.; Kitamura, K.; Ohta, J.; Toda, R.; Nakashima, T.; Iwamori, H. Deep-sea mud in the Pacific Ocean as a potential resource for rare-earth elements. Nat. Geosci. 2011, 4 , 535–539. 54. Takaya, Y.; Yasukawa, K.; Kawasaki, T.; Fujinaga, K.; Ohta, J.; Usui, Y.; Nakamura, K.; Kimura, J.-I.; Chang, Q.; Hamada, M.; et al. The tremendous potential of deep-sea mud as a source of rare-earth elements. Sci. Rep. 2018, 8 , 5763. 55. Ohta, J.; Yasukawa, K.; Nakamura, K.; Fujinaga, K.; Iijima, K.; Kato, Y. Geological features and resource potential of deep-sea mud highly enriched in rare-earth elements in the Central Pacific Basin and the Penrhyn Basin. Ore Geol. Rev. 2021, 139 , 104440. 56. Wang, Z.; Xu, G.; Lin, R.; Wang, H.; Ren, J. Energy performance contracting, risk factors, and policy implications: Identification and analysis of risks based on the best-worst network method. Energy 2019, 170 , 1–13. 57. Zhang, W.; Noble, A.; Yang, X.; Honaker, R. A Comprehensive Review of Rare Earth Elements Recovery from Coal-Related Materials. Minerals 2020, 10 , 451. 58. Huang, Z.; Fan, M.; Tian, H. Rare earth elements of fly ash from Wyoming’s Powder River Basin coal. J. Rare Earths 2020, 38 , 219–226. 59. Wang, Z.; Dai, S.; Zou, J.; French, D.; Graham, I.T. Rare earth elements and yttrium in coal ash from the Luzhou power plant in Sichuan, Southwest China: Concentration, characterization and optimized extraction. Int. J. Coal Geol. 2019, 203 , 1–14. 60. Dai, S.; Finkelman, R.B. Coal as a promising source of critical elements: Progress and future prospects. Int. J. Coal Geol. 2018, 186 , 155–164. 61. Dai, S.; Yan, X.; Ward, C.R.; Hower, J.C.; Zhao, L.; Wang, X.; Zhao, L.; Ren, D.; Finkelman, R.B. Valuable elements in Chinese coals: a review. Int. Geol. Rev. 2018, 60 , 590–620. 62. Luttrell, G.H.; Kiser, M.J.; Yoon, R.-H.; Noble, A.; Rezaee, M.; Bhagavatula, A.; Honaker, R.Q. A Field Survey of Rare Earth Element Concentrations in Process Streams Produced by Coal Preparation Plants in the Eastern USA. Mining, Metall. Explor. 2019, 36 , 889–902. 63. Zhang, W.; Honaker, R. Characterization and recovery of rare earth elements and other critical metals (Co, Cr, Li, Mn, Sr, and V) from the calcination products of a coal refuse sample. Fuel 2020, 267 , 117236.
Chapter 2 Literature Review Barros, O. | 2023 | 26 64. Zhang, W.; Honaker, R. Calcination pretreatment effects on acid leaching characteristics of rare earth elements from middlings and coarse refuse material associated with a bituminous coal source. Fuel 2019, 249 , 130–145. 65. Ji, B.; Li, Q.; Zhang, W. Leaching recovery of rare earth elements from the calcination product of a coal coarse refuse using organic acids. J. Rare Earths 2022, 40 , 318–327. 66. Commission, E.; Directorate-General for Internal Market Entrepreneurship and SMEs, I.; Grohol, M.; Veeh, C. Study on the critical raw materials for the EU 2023 : final report ; Publications Office of the European Union, 2023; 67. Burton, J. U.S. Geological Survey Releases 2022 List of Critical Minerals Available online: https://www.usgs.gov/news/national-news-release/us-geological-survey-releases-2022-listcritical-minerals (accessed on Aug 4, 2023). 68. Zhao, F.; Repo, E.; Meng, Y.; Wang, X.; Yin, D.; Sillanpää, M. An EDTA-β-cyclodextrin material for the adsorption of rare earth elements and its application in preconcentration of rare earth elements in seawater. J. Colloid Interface Sci. 2016, 465 , 215–224. 69. Rostami, I.; Rezvani, H.; Hu, Z.; Shahmoradian, S.H. Breakthroughs in medicine and bioimaging with up-conversion nanoparticles. Int. J. Nanomedicine 2019, 14 , 7759–7780. 70. Evans, C.H. Medical Uses of the Rare Earths. In Episodes from the History of the Rare Earth Elements ; Springer Netherlands: Dordrecht, 1996; pp. 205–228. 71. Giese, E. Rare Earth Elements: therapeutic and diagnostic applications in modern medicine. Clin. Med. Reports 2018, 2 . 72. Binnemans, K.; Jones, P.T.; Blanpain, B.; Van Gerven, T.; Yang, Y.; Walton, A.; Buchert, M. Recycling of rare earths: a critical review. J. Clean. Prod. 2013, 51 , 1–22. 73. Statista REE share of consumption worldwide by end use 2021 Available online: https://www.statista.com/statistics/604190/distribution-of-rare-earth-element-consumptionworldwide-by-end-use/ (accessed on Aug 9, 2023). 74. Bonfante, M.C.; Raspini, J.P.; Fernandes, I.B.; Fernandes, S.; Campos, L.M.S.; Alarcon, O.E. Achieving Sustainable Development Goals in rare earth magnets production: A review on state of the art and SWOT analysis. Renew. Sustain. Energy Rev. 2021, 137 , 110616. 75. Yang, Y.; Walton, A.; Sheridan, R.; Güth, K.; Gauß, R.; Gutfleisch, O.; Buchert, M.; Steenari, B.-M.; Van Gerven, T.; Jones, P.T.; et al. REE Recovery from End-of-Life NdFeB Permanent Magnet Scrap: A Critical Review. J. Sustain. Metall. 2017, 3 , 122–149. 76. Eggert, R.; Wadia, C.; Anderson, C.; Bauer, D.; Fields, F.; Meinert, L.; Taylor, P. Rare Earths: Market Disruption, Innovation, and Global Supply Chains. Annu. Rev. Environ. Resour. 2016, 41 , 199–222. 77. Tan, Q.; Li, J.; Zeng, X. Rare Earth Elements Recovery from Waste Fluorescent Lamps: A Review. Crit. Rev. Environ. Sci. Technol. 2015, 45 , 749–776. 78. Fujita, Y.; McCall, S.K.; Ginosar, D. Recycling rare earths: Perspectives and recent advances. MRS Bull. 2022, 47 , 283–288. 79. El Ouardi, Y.; Virolainen, S.; Massima Mouele, E.S.; Laatikainen, M.; Repo, E.; Laatikainen, K. The recent progress of ion exchange for the separation of rare earths from secondary resources – A review. Hydrometallurgy 2023, 218 , 106047. 80. Machacek, E.; Richter, J.L.; Habib, K.; Klossek, P. Recycling of rare earths from fluorescent lamps: Value analysis of closing-the-loop under demand and supply uncertainties. Resour. Conserv. Recycl. 2015, 104 , 76–93. 81. Niskanen, J.; Lahtinen, M.; Perämäki, S. Acetic acid leaching of neodymium magnets and iron separation by simple oxidative precipitation. Clean. Eng. Technol. 2022, 10 , 100544. 82. Pietrantonio, M.; Pucciarmati, S.; Sebastianelli, L.; Forte, F.; Fontana, D. Materials recovery from end-of-life wind turbine magnets. Int. J. Environ. Sci. Technol. 2022, 19 , 8019–8026.
Chapter 2 Literature Review Barros, O. | 2023 | 27 83. Kumari, A.; Sinha, M.K.; Pramanik, S.; Sahu, S.K. Recovery of rare earths from spent NdFeB magnets of wind turbine: Leaching and kinetic aspects. Waste Manag. 2018, 75 , 486–498. 84. Kumar Parhi, P.; Kumari Misra, P.; Kumar Jyothi, R. Integrated hydrometallurgical investigation for extraction and recovery of neodymium (Nd) from waste permanent Nd-Fe-B magnet. Inorg. Chem. Commun. 2023, 148 , 110251. 85. Gutiérrez-Gutiérrez, S.C.; Coulon, F.; Jiang, Y.; Wagland, S. Rare earth elements and critical metal content of extracted landfilled material and potential recovery opportunities. Waste Manag. 2015, 42 , 128–136. 86. Royen, H.; Fortkamp, U. Rare Earth Elements - Purification, Separation and Recycling ; 2016; ISBN 9789188319128. 87. Işıldar, A.; Rene, E.R.; van Hullebusch, E.D.; Lens, P.N.L. Electronic waste as a secondary source of critical metals: Management and recovery technologies. Resour. Conserv. Recycl. 2018, 135 , 296–312. 88. Akcil, A.; Erust, C.; Gahan, C.S.; Ozgun, M.; Sahin, M.; Tuncuk, A. Precious metal recovery from waste printed circuit boards using cyanide and non-cyanide lixiviants – A review. Waste Manag. 2015, 45 , 258–271. 89. Kaya, M. Recovery of metals and nonmetals from electronic waste by physical and chemical recycling processes. Waste Manag. 2016, 57 , 64–90. 90. Baldé, C.P.; Forti, V.; Gray, V.; Kuehr, R.; Stegmann, P. The Global E-waste Monitor 2017 ; 2017; ISBN 9789280890532. 91. Tsamis, A.; Coyne, M. Recovery of Rare Earths from Electronic wastes: An opportunity for HighTech SMEs ; 2015; 92. European Commision Report on Critical Raw Materials for the EU ; 2014; 93. Zhang, Y.; Liu, S.; Xie, H.; Zeng, X.; Li, J. Current Status on Leaching Precious Metals from Waste Printed Circuit Boards. Procedia Environ. Sci. 2012, 16 , 560–568. 94. Friege, H. Review of material recovery from used electric and electronic equipment-alternative options for resource conservation. Waste Manag. Res. J. a Sustain. Circ. Econ. 2012, 30 , 3–16. 95. Jones, P.T.; Geysen, D.; Tielemans, Y.; Van Passel, S.; Pontikes, Y.; Blanpain, B.; Quaghebeur, M.; Hoekstra, N. Enhanced Landfill Mining in view of multiple resource recovery: a critical review. J. Clean. Prod. 2013, 55 , 45–55. 96. Ongondo, F.O.; Williams, I.D.; Whitlock, G. Distinct Urban Mines: Exploiting secondary resources in unique anthropogenic spaces. Waste Manag. 2015, 45 , 4–9. 97. Johansson, N.; Krook, J.; Eklund, M.; Berglund, B. An integrated review of concepts and initiatives for mining the technosphere: towards a new taxonomy. J. Clean. Prod. 2013, 55 , 35–44. 98. Jowitt, S.M.; Werner, T.T.; Weng, Z.; Mudd, G.M. Recycling of the rare earth elements. Curr. Opin. Green Sustain. Chem. 2018, 13 , 1–7. 99. Li, J.; He, X.; Zeng, X. Designing and examining e-waste recycling process: methodology and case studies. Environ. Technol. 2017, 38 , 652–660. 100. Tan, Q.; Deng, C.; Li, J. Innovative Application of Mechanical Activation for Rare Earth Elements Recovering: Process Optimization and Mechanism Exploration. Sci. Rep. 2016, 6 , 19961. 101. Jones, P.T.; Van Gerven, T.; Van Acker, K.; Geysen, D.; Binnemans, K.; Fransaer, J.; Blanpain, B.; Mishra, B.; Apelian, D. CR3: Cornerstone to the sustainable inorganic materials management (SIM2) research program at K.U.Leuven. JOM 2011, 63 , 14–15. 102. Reid, S.; Tam, J.; Yang, M.; Azimi, G. Technospheric Mining of Rare Earth Elements from Bauxite Residue (Red Mud): Process Optimization, Kinetic Investigation, and Microwave Pretreatment. Sci. Rep. 2017, 7 , 15252. 103. Bian, Y.; Guo, S.; Jiang, L.; Liu, J.; Tang, K.; Ding, W. Recovery of Rare Earth Elements from NdFeB Magnet by VIM-HMS Method. ACS Sustain. Chem. Eng. 2016, 4 , 810–818.
Chapter 2 Literature Review Barros, O. | 2023 | 28 104. Rombach, E.; Friedrich, B. Chapter 10 - Recycling of Rare Metals. In Handbook of Recycling ; Elsevier, 2014; pp. 125–150 ISBN 9780123964595. 105. Pérez-Botella, E.; Valencia, S.; Rey, F. Zeolites in Adsorption Processes: State of the Art and Future Prospects. Chem. Rev. 2022, 122 , 17647–17695. 106. Laurino, C.; Palmieri, B. Zeolite: “the magic stone”; main nutritional, environmental, experimental and clinical fields of application. Nutr. Hosp. 2015, 32 , 573–81. 107. Hammond, C.; Padovan, D.; Tarantino, G. Porous metallosilicates for heterogeneous, liquid-phase catalysis: perspectives and pertaining challenges. R. Soc. Open Sci. 2018, 5 , 171315. 108. Silva, B.; Figueiredo, H.; Quintelas, C.; Neves, I.C.; Tavares, T. Zeolites as supports for the biorecovery of hexavalent and trivalent chromium. Microporous Mesoporous Mater. 2008, 116 , 555–560. 109. Vaca Mier, M.; López Callejas, R.; Gehr, R.; Jiménez Cisneros, B.E.; Alvarez, P.J.J. Heavy metal removal with mexican clinoptilolite: multi-component ionic exchange. Water Res. 2001, 35 , 373– 378. 110. Kulprathipanja, S. Zeolites in Industrial Separation and Catalysis ; Kulprathipanja, S., Ed.; Wiley, 2010; ISBN 9783527325054. 111. Breck DW Zeolite molecular sieves, chemistry, and use, ; Wiley, 1974; ISBN 978-0-471-09985-7. 112. McCusker, L.B.; Liebau, F.; Engelhardt, G. Nomenclature of structural and compositional characteristics of ordered microporous and mesoporous materials with inorganic hosts(IUPAC Recommendations 2001). Pure Appl. Chem. 2001, 73 , 381–394. 113. Flanigen, E.M. Molecular sieve zeolite technology - the first twenty-five years. Pure Appl. Chem. 1980, 52 , 2191–2211. 114. Flanigen, E.M.; Broach, R.W.; Wilson, S.T. Chapter 1: Introduction. In Zeolites in Industrial Separation and Catalysis ; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, Germany, 2010; pp. 1–26 ISBN 9783527325054. 115. Townsend, R.P.; Coker, E.N. Chapter 11: Ion exchange in zeolites. In Introduction to Zeolite Science and Practice ; van Bekkum, H., Flanigen, E.M., Jacobs, P.A., Jansen, J.C.B.T.-S. in S.S. and C., Eds.; Elsevier, 2001; Vol. 137, pp. 467–524 ISBN 0167-2991. 116. Dyer, A. Chapter 16: Ion-Exchange Properties of Zeolites and Related Materials. In Introduction to Zeolite Science and Practice ; Čejka, J., van Bekkum, H., Corma, A., Schüth, F.B.T.-S. in S.S. and C., Eds.; Elsevier, 2007; Vol. 168, pp. 525–553 ISBN 0167-2991. 117. Inglezakis, V.J. The concept of “capacity” in zeolite ion-exchange systems. J. Colloid Interface Sci. 2005, 281 , 68–79. 118. Milton, R.M. Chapter 1: Molecular Sieve Science and Technology. In Zeolite Synthesis ; Occelli, M.L., Robson, H.E., Eds.; American Chemical Society (ACS), 1989; pp. 1–10. 119. Breck, D.W.; Milton, R.M. Cadmium-loaded molecular sieve 1961. 120. Breck, D.W.; Milton, R.. Metal Loading of Molecular Sieves 1961. 121. Rabo, J.A.; Pickert, P.E.; Boyle, J.E. Hydrocarbon Conversion Process with the Use of a Y Type Crystalline Zeolite 1966. 122. Rabo, J.A.; Pickert, P.E.; Boyle, J.E. Hydrocarbon Conversion Catalysts 1968. 123. Gadd, G.M. Biosorption: critical review of scientific rationale, environmental importance and significance for pollution treatment. J. Chem. Technol. Biotechnol. 2009, 84 , 13–28. 124. Costa, F.; Lago, A.; Barros, Ó.; Rocha, V.; Vipotnik, Z.; Silva, B.; Tavares, T. Retention of organic micro-pollutants by sorption processes. In Current Developments in Biotechnology and Bioengineering ; Elsevier, 2020; pp. 331–362. 125. Tareq, R.; Akter, N.; Azam, M.S. Biochars and Biochar Composites. In Biochar from Biomass and Waste ; Elsevier, 2019; pp. 169–209. 126. Yang, R.T. Adsorbents: Fundamentals and Applications ; Wiley, 2003; ISBN 9780471297413.
Chapter 2 Literature Review Barros, O. | 2023 | 29 127. Ata, A.; Nalcaci, O.O.; Ovez, B. Macro algae Gracilaria verrucosa as a biosorbent: A study of sorption mechanisms. Algal Res. 2012, 1 , 194–204. 128. Lee, S.-Y.; Choi, H.-J. Persimmon leaf bio-waste for adsorptive removal of heavy metals from aqueous solution. J. Environ. Manage. 2018, 209 , 382–392. 129. Nguyen, T.C.; Loganathan, P.; Nguyen, T.V.; Kandasamy, J.; Naidu, R.; Vigneswaran, S. Adsorptive removal of five heavy metals from water using blast furnace slag and fly ash. Environ. Sci. Pollut. Res. 2018, 25 , 20430–20438. 130. Burakov, A.E.; Galunin, E. V; Burakova, I. V; Kucherova, A.E.; Agarwal, S.; Tkachev, A.G.; Gupta, V.K. Adsorption of heavy metals on conventional and nanostructured materials for wastewater treatment purposes: A review. Ecotoxicol. Environ. Saf. 2018, 148 , 702–712. 131. Djilani, C.; Zaghdoudi, R.; Modarressi, A.; Rogalski, M.; Djazi, F.; Lallam, A. Elimination of organic micropollutants by adsorption on activated carbon prepared from agricultural waste. Chem. Eng. J. 2012, 189 – 190 , 203–212. 132. Mailler, R.; Gasperi, J.; Coquet, Y.; Derome, C.; Buleté, A.; Vulliet, E.; Bressy, A.; Varrault, G.; Chebbo, G.; Rocher, V. Removal of emerging micropollutants from wastewater by activated carbon adsorption: Experimental study of different activated carbons and factors influencing the adsorption of micropollutants in wastewater. J. Environ. Chem. Eng. 2016, 4 , 1102–1109. 133. Kim, E.; Jung, C.; Han, J.; Her, N.; Min Park, C.; Son, A.; Yoon, Y. Adsorption of selected micropollutants on powdered activated carbon and biochar in the presence of kaolinite. Desalin. Water Treat. 2016, 57 , 1–13. 134. Henley, E.J.; Seader, J.D.; Keith Roper, D. Separation process principles, 3rd Edition ; 2011; ISBN 9780470646113. 135. Das, N. Recovery of precious metals through biosorption — A review. Hydrometallurgy 2010, 103 , 180–189. 136. Akpomie, K.G.; Dawodu, F.A.; Adebowale, K.O. Mechanism on the sorption of heavy metals from binary-solution by a low cost montmorillonite and its desorption potential. Alexandria Eng. J. 2015, 54 , 757–767. 137. Wilhelm, B.S.; Duncan, J.R. Reusability of immobilised Saccharomyces cerevisiae with successive copper adsorption-desorption cycles. Biotechnol. Lett. 1996, 18 , 531–536. 138. Rodríguez‐Padrón, D.; Puente‐Santiago, A.R.; Balu, A.M.; Muñoz‐Batista, M.J.; Luque, R. Environmental Catalysis: Present and Future. ChemCatChem 2019, 11 , 18–38. 139. Melián-Cabrera, I. Catalytic Materials: Concepts to Understand the Pathway to Implementation. Ind. Eng. Chem. Res. 2021, 60 , 18545–18559. 140. Bartholomew, C.H.; Farrauto, R.J. Fundamentals of Industrial Catalytic Processes ; Wiley, 2005; ISBN 9780471457138. 141. Moulijin, J.A.; Makkee, M.; van Diepen, A.E. Chemical process technology ; 2013; Vol. 51; ISBN 978-1-444-32025-1. 142. Hagen, J. Industrial Catalysis ; Wiley, 2015; ISBN 9783527331659. 143. Polshettiwar, V.; Varma, R.S. Green chemistry by nano-catalysis. Green Chem. 2010, 12 , 743. 144. Centi, G.; Perathoner, S. Catalysis and sustainable (green) chemistry. Catal. Today 2003, 77 , 287–297. 145. Anastas, P.T.; Kirchhoff, M.M.; Williamson, T.C. Catalysis as a foundational pillar of green chemistry. Appl. Catal. A Gen. 2001, 221 , 3–13. 146. Védrine, J.C. Chapter 9: Concluding remarks and challenges of heterogeneous catalysis on metal oxides. In Metal Oxides in Heterogeneous Catalysis ; Elsevier, 2018; pp. 551–569 ISBN 9780128116319. 147. Kalz, K.F.; Kraehnert, R.; Dvoyashkin, M.; Dittmeyer, R.; Gläser, R.; Krewer, U.; Reuter, K.; Grunwaldt, J.-D. Future Challenges in Heterogeneous Catalysis: Understanding Catalysts under
Chapter 2 Literature Review Barros, O. | 2023 | 30 Dynamic Reaction Conditions. ChemCatChem 2017, 9 , 17–29. 148. Fechete, I.; Wang, Y.; Védrine, J.C. The past, present and future of heterogeneous catalysis. Catal. Today 2012, 189 , 2–27. 149. Descorme, C.; Gallezot, P.; Geantet, C.; George, C. Heterogeneous Catalysis: A Key Tool toward Sustainability. ChemCatChem 2012, 4 , 1897–1906. 150. Centi, G.; Ciambelli, P.; Perathoner, S.; Russo, P. Environmental catalysis: trends and outlook. Catal. Today 2002, 75 , 3–15. 151. Sousa-Aguiar, E.F.; Trigueiro, F.E.; Zotin, F.M.Z. The role of rare earth elements in zeolites and cracking catalysts. Catal. Today 2013, 218 – 219 , 115–122. 152. Woltermann, G.M.; Magee, J.S.; Griffith, S.D. Chapter 4: Commercial Preparation and Characterization of FCC Catalysts. In Studies in Surface Science and Catalysis ; Elsevier, 1993; Vol. 76, pp. 105–144 ISBN 9780444890375. 153. Scherzer, J. Chapter 5: Correlation Between Catalyst Formulation and Catalytic Properties. In Studies in Surface Science and Catalysis ; Elsevier, 1993; Vol. 76, pp. 145–182 ISBN 9780444890375. 154. Malleswara Rao, T.V.; Dupain, X.; Makkee, M. Fluid catalytic cracking: Processing opportunities for Fischer–Tropsch waxes and vegetable oils to produce transportation fuels and light olefins. Microporous Mesoporous Mater. 2012, 164 , 148–163. 155. Mante, O.D.; Agblevor, F.A.; Oyama, S.T.; McClung, R. The effect of hydrothermal treatment of FCC catalysts and ZSM-5 additives in catalytic conversion of biomass. Appl. Catal. A Gen. 2012, 445 – 446 , 312–320. 156. Thomas, B.; Sugunan, S. Effect of rare earth metal ions on the structural and textural properties of NaFAU-Y zeolite and vapour phase alkylation of benzene with 1-octene. Indian J. Chem. Technol. 2005, 12 , 676–688. 157. Barros, Ó.; Costa, L.; Costa, F.; Lago, A.; Rocha, V.; Vipotnik, Z.; Silva, B.; Tavares, T. Recovery of Rare Earth Elements from Wastewater Towards a Circular Economy. Molecules 2019, 24 , 1005. 158. Assila, O.; Barros, Ó.; Fonseca, A.M.F.; Parpot, P.; Soares, O.S.G.P.; Pereira, M.F.R.; Zerrouq, F.; Kherbeche, A.; Rombi, E.; Tavares, T.; et al. Degradation of pollutants in water by Fenton-like oxidation over LaFe-catalysts: Optimization by experimental design. Microporous Mesoporous Mater. 2023, 349 , 112422. 159. Figueiredo, H.; Silva, B.; Kuźniarska-Biernacka, I.; Fonseca, A.M.; Medina, R.; Rasmussen, S.; Bañares, M.A.; Neves, I.C.; Tavares, T. Oxidation of cyclohexanol and cyclohexene with triazenido complexes of chromium immobilized in biosorption FAU supports. Chem. Eng. J. 2014, 247 , 134–141. 160. Figueiredo, H.; Neves, I.C.; Quintelas, C.; Tavares, T.; Taralunga, M.; Mijoin, J.; Magnoux, P. Oxidation catalysts prepared from biosorbents supported on zeolites. Appl. Catal. B Environ. 2006, 66 , 274–280. 161. Figueiredo, H.; Silva, B.; Quintelas, C.; Neves, I.C.; Tavares, T. Effect of the supporting zeolite structure on Cr biosorption: Performance of a single-step reactor and of a sequential batch reactor—A comparison study. Chem. Eng. J. 2010, 163 , 22–27. 162. Figueiredo, H.; Silva, B.; Quintelas, C.; Pereira, M.F.R.; Neves, I.C.; Tavares, T. Biosorption of hexavalent chromium based on modified y zeolites obtained by alkali-treatment. Environ. Eng. Manag. J. 2010, 9 , 305–311. 163. Jordan, M.I.; Mitchell, T.M. Machine learning: Trends, perspectives, and prospects. Science (80-. ). 2015, 349 , 255–260. 164. Bi, Q.; Goodman, K.E.; Kaminsky, J.; Lessler, J. What is Machine Learning? A Primer for the Epidemiologist. Am. J. Epidemiol. 2019, 188 , 2222–2239.
Chapter 2 Literature Review Barros, O. | 2023 | 31 165. Samuel, A.L. Some studies in machine learning using the game of checkers. IBM J. Res. Dev. 2000, 44 , 206–226. 166. El Naqa, I.; Murphy, M.J. Chapter 1: What Is Machine Learning? In Machine Learning in Radiation Oncology ; Springer International Publishing: Cham, 2015; pp. 3–11 ISBN 9781119815075. 167. Peng, G.C.Y.; Alber, M.; Buganza Tepole, A.; Cannon, W.R.; De, S.; Dura-Bernal, S.; Garikipati, K.; Karniadakis, G.; Lytton, W.W.; Perdikaris, P.; et al. Multiscale Modeling Meets Machine Learning: What Can We Learn? Arch. Comput. Methods Eng. 2021, 28 , 1017–1037. 168. Dhage, S.N.; Raina, C.K. A review on Machine Learning Techniques. Int. J. Recent Innov. Trends Comput. Commun. 2016, 4 , 395–399. 169. Qin, S.J.; Chiang, L.H. Advances and opportunities in machine learning for process data analytics. Comput. Chem. Eng. 2019, 126 , 465–473. 170. Mithun Sridharan Machine Learning Algorithms Mindmap | Jixta Available online: https://www.kaggle.com/discussions/getting-started/83518 (accessed on Aug 12, 2023). 171. Nasteski, V. An overview of the supervised machine learning methods. HORIZONS.B 2017, 4 , 51–62. 172. Torregrossa, D.; Leopold, U.; Hernández-Sancho, F.; Hansen, J. Machine learning for energy cost modelling in wastewater treatment plants. J. Environ. Manage. 2018, 223 , 1061–1067. 173. Pang, J.-W.; Yang, S.-S.; He, L.; Chen, Y.-D.; Cao, G.-L.; Zhao, L.; Wang, X.-Y.; Ren, N.-Q. An influent responsive control strategy with machine learning: Q-learning based optimization method for a biological phosphorus removal system. Chemosphere 2019, 234 , 893–901. 174. Foroughi, M.; Ahmadi Azqhandi, M.H.; Kakhki, S. Bio-inspired, high, and fast adsorption of tetracycline from aqueous media using Fe3O4-g-CN@PEI-β-CD nanocomposite: Modeling by response surface methodology (RSM), boosted regression tree (BRT), and general regression neural network (GRNN). J. Hazard. Mater. 2020, 388 , 121769. 175. Al-Gheethi, A.A.; Mohd Salleh, M.S.; Noman, E.A.; Mohamed, R.M.S.R.; Crane, R.; Hamdan, R.; Naushad, M. Cephalexin Adsorption by Acidic Pretreated Jackfruit Adsorbent: A Deep Learning Prediction Model Study. Water 2022, 14 , 2243. 176. Zhang, K.; Zhong, S.; Zhang, H. Predicting Aqueous Adsorption of Organic Compounds onto Biochars, Carbon Nanotubes, Granular Activated Carbons, and Resins with Machine Learning. Environ. Sci. Technol. 2020, 54 , 7008–7018. 177. Sigmund, G.; Gharasoo, M.; Hüffer, T.; Hofmann, T. Deep Learning Neural Network Approach for Predicting the Sorption of Ionizable and Polar Organic Pollutants to a Wide Range of Carbonaceous Materials. Environ. Sci. Technol. 2020, 54 , 4583–4591. 178. Sadek, A.H.; Fahmy, O.M.; Nasr, M.; Mostafa, M.K. Predicting Cu(II) Adsorption from Aqueous Solutions onto Nano Zero-Valent Aluminum (nZVAl) by Machine Learning and Artificial Intelligence Techniques. Sustainability 2023, 15 , 2081. 179. Abdi, J.; Mazloom, G. Machine learning approaches for predicting arsenic adsorption from water using porous metal–organic frameworks. Sci. Rep. 2022, 12 , 16458. 180. Ke, B.; Nguyen, H.; Bui, X.-N.; Bui, H.-B.; Choi, Y.; Zhou, J.; Moayedi, H.; Costache, R.; NguyenTrang, T. Predicting the sorption efficiency of heavy metal based on the biochar characteristics, metal sources, and environmental conditions using various novel hybrid machine learning models. Chemosphere 2021, 276 , 130204. 181. Kitchin, J.R. Machine learning in catalysis. Nat. Catal. 2018, 1 , 230–232. 182. Li, Z.; Wang, S.; Xin, H. Toward artificial intelligence in catalysis. Nat. Catal. 2018, 1 , 641–642. 183. Yang, W.; Fidelis, T.T.; Sun, W.-H. Machine Learning in Catalysis, From Proposal to Practicing. ACS Omega 2020, 5 , 83–88.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 38 3.2.7. Machine learning The ML analysis was performed using a table, named DataFrame, with labeled axes (rows and columns). The rows are related with the samples used, while the columns are the different elements that are being used to evaluate the different samples. The different features used were the C/C0 results after 24h for each tested REE, the pH of the zeolite treatment and the pH of the adsorption solution after the 24 h time point. The DataFrame was evaluated under unsupervised learner (Principal Component Analysis, KMeans Analysis) and supervised learner (classification and regression). The Principal Component Analysis, PCA, is a method for reducing the dimensionality of data, leading to an increased interpretation and minimizing information lost, while the K-Means divides the samples into groups or clusters that are more compatible with each other accordantly to the studied conditions. K-nearest neighbors Classifier (KNN), Decision Tree Classifier and Random Forest Classifier were used to classify the samples. These classifiers are often used in binary classification, as explained in Table 3.2. The data was divided into two sets, a training set (70 % of the data), which contains a known output and where the model learns to generalize and then apply to other data and a test set (30 % of the data), where the model’s prediction is tested. It also added a stratify option to the data division, allowing both test and train sets to have the same percentage of positive cases (in this case, a good adsorbent) as the complete set. Table 3.2: The binary classification for the different REE. The C/C0 values were given a classification accordantly. The mean value of these intervals was taken and given the respective binary classification. C/C0 intervals C/C0 Classification Binary Classification 0.8 < C/C0 < 1.0 1 > 4.0 is 1 0.6 < C/C0 < 0.8 2 0.4 < C/C0 < 0.6 3 0.2 < C/C0 < 0.4 4 < 4.0 is 0 0.0 < C/C0 < 0.2 5 For the regression, the training test split was 70% for training and 30 % for testing. Different metrics were used to evaluate the regression prediction, which were the mean absolute error (MAE), the mean squared error (MSE), the root mean squared error (RMSE) and the R-Squared (R2). These metrics
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 39 were used to evaluate the test data. MAE is calculated by the sum of the absolute differences between the real and predicted values of each tested observation and then divided by the number of observations. All tests were performed using Spyder (Python 3.9) and the respective needed modules as pandas, numpy, scikit-learn, matplotlib and seaborn. 3.2.8. Statistical analysis The adsorption results were analyzed using the One-Way ANOVA, where the obtained values for each pre-treated zeolite and the respective controls were compared between each other. The Two-Way ANOVA was used for the statistical analysis of the desorption results. The Bonferroni’s multiple comparison test was used for both data sets. The ANOVA analyses were performed using the software Graph Pad Prism version 8.0.2 (Graph Pad Software, Inc, San Diego, CA, USA). The results were only considered significantly different when the probability ( p -value) was lower than 0.05, assuming a 95 % confidence interval. 3.3. Results and Discussion 3.3.1. Modified zeolites characterization Two different zeolite structures, FAU (13X) and LTA (4A), were subjected to chemical treatments with acid and basic solutions at room temperature. From the SEM observations it was concluded that the modifications had slight effects on the zeolite morphology, which were more noticeable at higher concentrations. These results do not give any specific information regarding the modifications and their eventual impact on the surface characteristics or on the zeolite behaviour. The elementary quantification of the pristine zeolites and the modified ones were evaluated using EDS. The results for zeolite 13X and its modified forms are shown in Table 3.3. These data prove that the different chemical treatments affect the surface of the zeolite. In the case of the acid treatments, the sodium present in the framework was entirely replaced by protons of the acid solutions and a dealumination was observed. An opposed effect was observed for the alkali treatments. A decrease in the Si/Al ratio was perceived in the same magnitude for both strong bases, NaOH and KOH, and its values ranged between 1.20 and 1.34, showing that both alkali solutions affect the zeolite structure in the same extension.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 40 Table 3.3: EDS surface analysis of modified 13X zeolite and controls. Chemical treatment Element (wt %) Si/Al Samples Conc. (M) O Si Al Na K ZX_HCl 0.25 56.0 ± 1.0 27.3 ± 1.7 5.5 ± 0.0 0 - 4.8 ZX_HNO3 0.25 57.0 ± 4.3 27.0 ± 2.4 5.9 ± 0.4 0 - 4.4 ZX_H2SO4 0.25 56.6 ± 6.1 27.0 ± 2.9 5.3 ± 0.4 0 - 4.9 ZX_NaOH 0.10 68.0 ± 0.4 15.2 ± 0.3 12.2 ± 0.2 3.6 ± 0.2 - 1.2 0.25 57.6 ± 1.5 14.2 ± 1.0 10.2 ± 0.6 9.6 ± 0.7 - 1.3 0.50 57.7 ± 1.0 16.8 ± 0.6 12.3 ± 0.2 10.8 ± 0.6 - 1.3 ZX_KOH 0.10 58.2 ± 1.0 16.9 ± 0.3 12.2 ± 0.3 7.5 ± 0.4 0 1.3 0.25 58.0 ± 1.5 16.0 ± 0.7 12.1 ± 0.4 6.0 ± 0.2 6.2 ± 0.7 1.3 0.5 56.5 ± 4.7 15.9 ± 1.3 11.5 ± 0.7 4.9 ± 0.9 8.6 ± 3.4 1.3 ZX_H2O 57.3 ± 1.7 18.1 ± 1.3 10.6 ± 1.2 12.0 ± 1.0 - 1.6 Z13X 57.3 ± 1.7 18.1 ± 1.3 10.6 ± 1.2 12.0 ± 1.0 - 1.6 Considering that the acid treatment provokes more modifications to the zeolite framework in the case of Z13X, this treatment was not carried out on the LTA structure. The other reason is related with the poor adsorption capacity that the acid treatment had on the Z13X, as will be analyzed in the 3.3.2 Selection of the most suitable chemical treatment section of this chapter. The EDS results for the zeolite 4A and the modified samples are shown in Table 3.4. In the case of LTA, the alkali treatments do not affect the Si/Al ratio since is similar between the samples and close to the values found for the controls. Table 3.4: EDS analyses of the zeolite 4A, the alkali pre-treated samples and controls. Chemical treatment Element (wt %) Si/Al Samples Conc. (M) O Si Al Na K ZA_NaOH 0.10 56.6 ± 2.3 13.5 ± 2.5 12.5 ± 1.8 9.5 ± 1.7 1.0 0.25 56.7 ± 1.6 16.71 ± 1.2 15.2± 0.7 11.4 ± 0.3 1.1 0.50 60.5 ± 3.1 14.27 ± 1.6 13.33 ± 1.4 11.7 ± 0.1 1.0 ZA_KOH 0.10 57.2 ± 2.7 15.76 ± 0.7 14.1 ± 0.3 7.0 ± 1.0 5.6 ± 2.9 1.1 0.25 55.9 ± 1.9 16.1 ± 0.7 14.5 ± 0.5 6.3 ± 0.5 7.0 ± 1.1 1.1 0.50 56.2 ± 0.5 15.2 ± 0.3 14.2 ± 0.3 4.5 ± 0.1 9.4 ± 0.57 1.0 ZA_H2O 58.1 ± 0.8 16.5 ± 1.2 14.9 ± 0.6 10.4 ± 1.0 1.1 Z4A 55.9 ± 1.2 17.2 ± 0.4 15.5 ± 0.4 11.2 ± 0.4 1.1
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 41 An increase of K+, without the loss of Si or Al from the framework, was observed for the samples treated with KOH solutions with different concentrations. This effect was more pronounced on LTA structure than on FAU structure, where the presence of K+ was detected at the higher concentrations of KOH (Table 3.3). As the washing step of the zeolites preparation may lead to the Na removal and, consequently to a negative charge of the solid surfaces, the presence of the K+ may be related to the counterbalance of such charge. The profiles of pHPZC and the FTIR spectra for both zeolite structures, FAU and LTA, and respective modified samples are shown in Figure 3.1 and Figure 3.2. The alkali treated 13X and 4A zeolites have similar pHPZC to the ones of the respective controls, closer to 10, Figure 3.1. Although there was no difference regarding the pHPZC values between the modified and the control samples, the treated zeolites show a different behavior during the adsorption. It should be mentioned that higher pHPZC values probably could lead to REE precipitation during the adsorption assays, which is not intended. All the acid treated 13X samples decreased their pHPZC compared to the pristine zeolite, 9.6 [9]. The samples ZX_HNO3 and ZX_H2SO4 had lower values, 5.6 and 4.4, respectively, but the effect was not so evident on the ZX_HCl, 8.4. The acid treatment with HNO3 or H2SO4 greatly decreases the pHZPC values as expected as the H+ from the chemical modification was incorporated into the zeolites to reduce their natural negative charge. The pHPZC value for one of the controls, ZX_H2O, is 7.5 while ZA_H2O presents a pHPZC value close to 7.0. As previously stated, the modification of these zeolites was performed using only dH2O, with a pH value of 7.00. However, due the absorption of carbon dioxide from the atmosphere, the pH value will be slightly acidic with a value around 5.8. This acidification of the water used for the chemical modification of the controls can explain the difference of the pHPZC value when compared to the zeolites without modification (Z13X and Z4A).
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 42 Figure 3.1: pHPZC results: A - acid modified 13X, B - alkali modified 13X, and C – alkali modified 4A.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 43 Figure 3.2: FTIR analyses: A - acid modified 13X, B - alkali modified 13X and C - alkali modified 4A.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 44 FTIR spectra of zeolites 13X and 4A, Figure 3.2, are very similar to the ones of FAU zeolite structures. In Figure 3.2 is possible to see a band at 1640 cm-1 characteristic of (H2O) vibration of absorbed water on the zeolite. The characteristic bands of the lattice vibrations of the framework are evidenced in 1330 to 450 cm-1. The band at 960 cm-1 is attributed to the asymmetric stretching of Si-O and Al-O belonging to the TO4 tetrahedras (T = Si or Al) [31–33]. The bands at 670 cm-1 and near 750 cm-1 are related to the Si–O symmetric stretching and oscillations of aluminosilicate oxygen tetrahedral chains [32,33]. At the same time, the band near 550 cm-1 is attributed to the symmetric stretching vibrations of the bridge bonds Si-O-Si and to the bending vibrations of O-Si-O [34]. Comparing the modified samples to the respective controls, the same bands in the same positions are observed indicating that the pre-treatments had no effect on the pristine zeolites. In addition, Si/Al ratio of the FAU samples can be determined by FTIR analysis using the equation 5: 𝑥 =3.857−0.00621𝑊𝐷𝑅 (Eq. 5) In here x = (1+Si/Al)-1 and WDR is the wavenumber at 500-650 cm-1, related to the vibrations of the FAU lattice [35]. The results are shown in Table 3.5. Table 3.5: Si/Al ratios based on FTIR and EDS analyses. Zeolite Conc. (M) Si/Al from FTIR Si/Al from EDS ZX_HCl 0.25 1.50 4.75 ZX_HNO3 0.25 1.42 4.38 ZX_H2SO4 0.25 1.42 4.88 ZX_NaOH 0.10 1.58 1.20 0.25 1.42 1.34 0.50 1.58 1.31 ZX_KOH 0.10 1.50 1.33 0.25 1.42 1.26 0.50 1.50 1.33 ZX_H2O 1.50 1.64 Z13X 1.50 1.64 From the results of Table 3.5 show that the zeolites modification was mostly superficial, as supported by the Si/Al ratio values shown in Table 3.3. The Si/Al ratio given by FTIR gives an indication of the framework ratio. However, the difference between the EDS and the FTIR results supports that the modification affects the most external surface. Any modification of the zeolite structure would become
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 45 irrelevant to the adsorption as this is a surface phenomenon. For the same reason, it is expected that acid modified zeolites may act as poor adsorbents, as acids tend to reduce the available specific surface. 3.3.2. Selection of the most suitable chemical treatment 3.3.2.1. Selection based on adsorption results The REE ionic radii are small enough, between 0.9 to 1.032 Å [36,37], so the zeolites Z13X and Z4A are expected to remove those ions from the liquid solution considering their average pore size, 7 Å and 4 Å, respectively. The adsorption data from the mixed REE solution with the pre-treated zeolites are presented in Figure 3.3. Overall, independently of the REE, the alkali treated zeolites t showed higher adsorption performance than the acid treated ones. Statistical differences (Table S-3.1) were found between the pre-treated sorbents, especially the acid modified zeolites and ZX_KOH 0.50 M, and the controls, for the retention of some REE. As the acid treated 13X zeolite did not show any enhancement of its adsorption ability, in accordance to the data in Table 3.5, it was not considered in the forward experiments. The adsorption process occurs on the material surface and the acid modified zeolites may suffer a reduction of the microporosity and an increase of the mesoporosity, which leads to a reduced specific surface area and explains the poorer results. The C/C0 results obtained after the alkali modification of surfaces are similar for both the zeolites treated with 0.25 M solutions and for ZX_NaOH 0.50 M, when compared to the controls. The ZX_KOH 0.50 M presented worse results, with two significant differences found for Tb and Eu when compared with Z13X (Table S-3.1). The samples modified with 0.10 M of NaOH and KOH were the ones that reached the lowest C/C0 in solution after 24 h, Figure 3.3. ZX_NaOH 0.10 M reached removals over 80 % for five REE, apart from Ce. The results are similar between the controls, Z13X and ZX_H2O, except for Tb and Eu, for which Z13X presented lower C/C0, suggesting that the pristine zeolite washing with distilled water does not improve or even worsened the adsorption capacity, as expected. The increased removal for the ZX_NaOH 0.1 M and ZX_KOH 0.1 M, when compared with the controls, could be related with the increase of the surface area of the zeolite resulting from the treatment. K+ was detected in the EDX analysis (Table 3.3) after the treatment with KOH, which can reduce the negative charge of the zeolite and this explains the lower adsorption for the zeolites treated with 0.25 and 0.5 M. For the treatment with NaOH, there was no increase in the Na+ in the EDX analysis (Table 3.3), probably due to a competition between the REE and the Na+. More Na+ was detected in the zeolite for the higher concentrations. The presence of high quantities of Na+ shown in analysis could be related with a poor performance, as happened for the controls. Therefore, the higher concentrations in the NaOH treatment
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 46 may lead to a low REE removal from the solution, assuming that what happened to the controls could also happen for these samples. Figure 3.3: REE adsorption on Z13X and modified samples. The assays were carried out with a multi solution of REE previously described. The pH solution was monitored during each of the assays with 13X zeolites, as presented in Figure 3.4, and for that, the precipitation of the REE can be discharged alkali treatment solutions with lower concentrations. However, some REE precipitation was observed 3 h after the beginning of the assay with zeolites treated with NaOH or with KOH, 0.50 M. For this reason, the concentration of 0.50 M for
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 47 both alkali treatment solutions are considered inadequate for the REE recovery. The Z13X zeolite presented a constant growth regarding the pH value that could be related with the presence of sodium in the zeolite, which tends to increase the pH value of water. Figure 3.4: pH values during the adsorption assays for the basic and acid pretreatment to the Z13X. The REE adsorption tests were also performed with the zeolite 4A and respective modified samples, Figure 3.5.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 54 Figure 3.8: Representative display of the classification of the different modified zeolites based on the experimental data obtained after 24 h of assay using ML algorithms. The classification algorithms used were KNN classifier (A), Decision Tree classifier (B) and Random Forest classifier (C). The 1 represents a good adsorbent, while the 0 is a bad adsorbent accordantly to the evaluation performed. The different colors, violet and orange, represent the zone of a good or bad sorbent, respectively.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 55 The classification was performed using 3 different classifiers: KNN, Decision Tree and Random Forest to evaluate its sustainability by using ML. The classification will help to select suitable new materials for REE removal from wastewater using zeolites as adsorbents and may be implemented for other pollutants removal to determine the best approach. The zeolites that had a good classification (binary classification of 1) were ZX_KOH 0.1 M, ZX_NaOH 0.1 M and ZA_NaOH, 0.25 and 0.5 M. The selection of the best suitable number of neighbors (n_neighbors) by the KNN Classifier, is shown in Figure S-3.1C, from which 1 neighbor is selected depending on the accuracy values for both training and test sets. The accuracy for the precision, recall and f1-score for each classifier used and for both test and training sets is determined. The precision of each classifier is related to the probability of making good predictions, which was 100 % for every classifier, as happens for the recall (value of the correctly identified positive predictions) and for f1-score (harmonic mean of the precision and the recall). In these approaches, it should be noticed that having 100 % accuracy on the training sets with a relatively low value for the test set means that overfitting occurs for the training set, which is not the bestcase scenario. The idea of the training set is to get a good generalization of the results and then the model is used for unseen data, a test set, which serves to evaluate the capacity of the model to classify new data. All classifiers presented a 100 % score for the training set. The same value was reached forthe test sets, in terms of precision, recall and f1-scores of the prediction done by the model, considering the actual classification. These results were validated by the confusion matrix, Figure 3.9, that confirms that there were only true positives (the model predicted good adsorption results and the results were actually good, that is the model predicted it was true and it was actually true) and true negatives (the model predicted poor adsorption results and the results were actually poor, that is the model predicted it was false and it was actual false). Overall, this study shows that it is possible to use ML algorithms to support the selection of the best suitable zeolite for the removal of REE as the models used were able to select the best adsorbents within the starting list with very good metric results.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 56 Figure 3.9: Confusion matrix for the test data for the different classifiers. The values shown refer to the fraction of the true correct predictions (when the model got it right) and false incorrect predictions (when the model got it wrong). 3.3.2.3. Predicting unseen data using ML algorithms The Pearson correlation was calculated from the DataFrame used in the previous analysis and the results are displayed in a heatmap, Figure 3.10. Figure 3.10: Heatmap representing the Pearson correlation between the different features considered in adsorption assays. The left scale represents the different correlation values and the respective colors. It may be seen that the different pH values (pH_assay and pH_treatment) have a moderate positive correlation (between 0.50 and 0.70) [41]. The pH_assay has a negligible correlation (between
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 57 0.00 and - 0.30) [41] with the C/C0 for La, a low negative correlation (between - 0.30 and - 0.5) [41] with the results for Ce, Y and Pr and a high negative correlation (between -070 and -0.90) [41] with the adsorption of Tb and Eu. It would be expected that C/C0 would be lower as the pH increases till a certain degree (pH values that do not lead to REE precipitation) and this was verified during the adsorption assays. The pH_treatment had moderate negative correlation (between - 0.50 and - 0.70) with the adsorption of Ce, Y and Pr, a high negative correlation (between -0.70 and -0.90) [41] with Tb and Eu and a low negative correlation (between - 0.30 and - 0.5) with La, following the same explanation as before. The C/C0 correlation between the REE is high (between 0.70 and 0.90) or very high positive (between 0.70 to 0.90) [41], with 2 exceptions: between La and Tb, 0.67, and between La and Eu, 0.66, which are moderate positive correlations. These results suggest that the entrapment values of the REE have a direct correlation with each other, meaning that for the samples used in this study it could be possible to predict the final C/C0 of one REE using the known values of another REE. It is essential to mention that the correlation is stronger for REE of the same group, light REE – from La to Gd or heavy REE – from Tb to Lu, including Y [1,2]. Pr presented the best correlation with the other tested REE, with values over 0.9, as shown in Figure 3.10. This leads to the possibility of only using Pr to estimate the adsorption values of other REE present in the solution, considering all zeolites tested during 24h assays. The results for the tested linear regressions are shown in Table 3.8. Table 3.8: Regression parameters for the C/C0 values prediction using Pr (as x) and the other REE (as y). Metrics and scoring Pr vs. La Pr vs. Ce Pr vs. Y Pr vs. Tb Pr vs. Eu Mean absolute error (MAE) 0.058 0.018 0.024 0.088 0.089 Mean squared error (MSE) 0.005 0.000 0.001 0.010 0.011 Root mean squared error (RMSE) 0.069 0.021 0.027 0.101 0.104 R² for the model explaining test data 0.772 0.984 0.977 0.795 0.801 R² for the model explaining training data 0.787 0.931 0.956 0.819 0.804 Table 3.8 displayed different metrics used to evaluate the scoring of the estimation of the C/C0 of the other REE, once the Pr values are known. MAE is related to the difference between the prediction and the real value of one observation, while the MSE metric is used to measure the quality of the model and RMSE helps to understand the performance of the model. The lower the values for these metrics, the closer are predicted values to the real ones, indicating that the model predictions are good. The R2 value is also essential, as it validates a good fitting to the data. As shown, the values are higher than 0.77
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 58 for both test and training sets, which is good since it reveals a strong correlation. It may be concluded from the overall evaluated metrics that it is possible to determine the residual REE concentrations in solution, just knowing one of them. However, a larger dataset with more samples would allow a better training for the model and would mprove prediction capacity of the model. Nevertheless, the REE concentrations prediction was tested for a time period using the data obtained with a single zeolite. Two different adsorption periods were considered, 24 h and 125 h , for the ZX_NaOH 0.10 M, which was the best zeolite. As before, a Pearson correlation was made for the different results of the REE C/C0, as shown in Figure 3.11. Figure 3.11: Heatmap representation of the Pearson correlations for the different REE after 24 h of contact with the sorbent(A) and after 125 h (B). The left scale represents the different correlation values and the respective colors. Figure 3.11A shows the correlations of the C/C0 of the different REE for the 24 h adsorption period. The correlation values are very high for every REE, including the correlation values obtained with Y, with values above 0.98. In this test, the Y will be used as the x value since this REE presented the
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 59 highest correlations with the other REE, with the results shown in Table 3.9. Every combination was tested during 125 h assays and the correlation values are shown in Figure 3.11B. Pr was used as the x value and the results are shown in Table 3.10. Comparing the Pearson Correlation values between the 2 assays, the longer tests present a better correlation between the tested REE. That is translated into better metrics and scorings for the models. The methodology was the same as the one performed for the estimation of the C/C0 values of the zeolite. The data were divided into training and test sets and the metrics were evaluated to assess the model behavior. Table 3.9: Scoring results for the 24 h adsorption assay regression between the different REE (as y) with Y (as x). Metrics and scoring Y vs. La Y vs. Ce Y vs. Pr Y vs. Tb Y vs. Eu Mean absolute error (MAE) 0.072 0.081 0.072 0.072 0.072 Mean squared error (MSE) 0.005 0.008 0.005 0.005 0.005 Root mean squared error (RMSE) 0.071 0.089 0.071 0.071 0.071 R² for the model explaining test data 0.943 0.762 0.943 0.943 0.943 R² for the model explaining training data 0.946 0.907 0.946 0.946 0.946 Table 3.10: Scoring results for the 125 h adsorption assay regression between the different REE (as y) with Pr (as x). Metrics and scoring Pr vs. La Pr vs. Ce Pr vs. Y Pr vs. Tb Pr vs. Eu Mean absolute error (MAE) 0.011 0.004 0.011 0.011 0.011 Mean squared error (MSE) 0.000 0.000 0.000 0.000 0.000 Root mean squared error (RMSE) 0.000 0.000 0.000 0.000 0.000 R² for the model explaining test data 1.000 1.000 1.000 1.000 1.000 R² for the model explaining training data 0.998 1.000 0.998 0.998 0.998 The results are much better for the model that used the 125 h adsorption data. That is related to the minor errors and the higher R2 values for the training and test sets that indicate that the predicted values are closer to the actual values, showing that increasing the measurements number is essential to construct a good model. For the sake of robustness validation, it was decided to use the model obtained from the 125 h assays to predict the C/C0 values of the REE in the 24 h assay. The C/C0 values of Pr were used as x, while the rest of the C/C0 values for the other REE were predicted from the previously
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 60 mentioned model. The metrics and respective scorings between the predicted and actual values are shown in Table 3 . 11. The metrics values in Table 3 . 11 are low, showing that the predicted results are very similar to those obtained experimentally. Comparing these metrics to the ones presented in Table 3.9 it may be concluded that the prediction capacity of the model improved on the base of the 125 h assay. Table 3.11: Scoring results for estimating the C/C0 values for the 24 h assay, using the model trained with the 125 h adsorption assay for the different REE (as y) based on Pr (as x). Metrics and scoring Pr vs La Pr vs Ce Pr vs Y Pr vs Tb Pr vs Eu Mean absolute error (MAE) 0.054 0.091 0.038 0.043 0.048 Mean squared error (MSE) 0.004 0.013 0.003 0.004 0.003 Root mean squared error (RMSE) 0.063 0.114 0.055 0.063 0.055 It is demonstrated that it is possible to predict the final C/C0 in solution for any REE, based on the values for one of the sorbates, as well as to predict the C/C0 values for each REE along time, based on the measurements at one time point. It is demonstrated that it is possible to train and test a model with a robust prediction capacity and reduced associated errors. These models can be further improved and tested to ensure the best possible prediction. In the future, these models could be an excellent help for faster quantification of metal pollutants, REE or heavy metals, eventually some other pollutants in water streams. This could make the quantification of various contaminants in water resources more effortless and faster, leading to quicker treatment. For the following tests, the selected modified zeolites are ZX_NaOH 0.10 M and ZX_KOH 0.10 M. The other zeolites, namely ZA_NaOH 0.25 and 0.50 M, although with a good potential for the REE removal from wastewater by the classification algorithm, they will not be considered for future assays due to the possible REE precipitation with ZA_NaOH 0.50 M, as previously explained, while ZA_NaOH 0.25 M did not really reach outstanding results during the adsorption. Therefore, only the selected modified zeolites will be used, as well as the controls, on the leaching assays and on the adsorption kinetics determination. 3.3.3. Leaching of REE The selected modified zeolites prepared by alkali treatments were used for the evaluation of leaching processes. HCl, H2SO4 and HNO3 solutions, 0.10 M, were used as eluents. The loaded zeolites
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 61 were subjected to leaching for 0.5h and circa 80 to 90% of the entrapped REE were recovered by the eluent. This period of contact is the one needed to achieve relevant recovery rates of the tested REE while avoiding any damage to the adsorbent. The recovery results after 0.5 h leaching are shown in Figure 3.12 for the 13X zeolite. Water leaching REE recovery (data not shown) from both zeolite structures was less than 5 %. This suggests that the presence of H+ in the solution is required to enhance the removal of the REE from the zeolite eventually by a cationic exchange with the REE2+. Figure 3.12: Recovery results for Z13X and respective controls, with HCl, H2SO4 and HNO3 aqueous solutions.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 62 Figure 3.12 shows that the zeolites modified with NaOH and KOH 0.10 M had the best recovery, with values near the 100 % for all the tested REE. They differ 40 to 45% from the Z13X and 20 to 35 % from the ZX_H2O. A statistical analysis of the recovery values was performed and the results are shown in Table S-3.3. A significant difference for every REE and eluent was observed when comparing with the control Z13X and with the modified zeolites (NaOH or KOH 0.10 M). In addition, a significant difference was also found for most of the REE and eluents tested when comparing the referred recoveries with the ones obtained with the ZX_H2O control. As expected, comparing the recovery results obtained with the zeolites treated with NaOH or KOH 0.10 M, no significant difference was observed, Figure 3.12, but some differences are seen between the controls ZXH2O with Z13X. In general, it may be concluded that the initial alkali treatments applied to the zeolites improve their capacity to remove REE from wastewater as well as to recover the sorbates by acid leaching. The tested eluents led to similar recoveries and present no significant difference between them. Therefore, for future assays, the selected eluent will be a solution of HNO3 since this acid is weaker than the others and does not represent an environmental threat as the other ones. A concentration step is the expected step to be performed after the desorption of the REE from the zeolite. Its complexity will depend on its final objective, on the presence of other metals and on the precipitant used. The purification of the individual REE requires a different approach than the recovery of a REE mixture. At the same time, the foreseen application of each element has also to be considered, depending on the anions used for the purpose. Some examples described in the literature include the precipitation with carbonate [42,43], fluoride [44,45], phosphate [46,47], sulfate [48–52] and oxalate [53–57]. 3.3.4. Adsorption kinetics The selected modified zeolites and the respective controls were used in kinetics evaluations to understand the mechanism of the whole process and its dependence on pre-treatments. Two fitting models were tested: the pseudo first-order, PFO, and the pseudo second-order, PSO. The parameters for both models is presented in Table S-3.4 with the respective confidence intervals in Table S-3.5 and the graphical representation in Figure S-3.2. The best fitting model was selected considering the correlation value, R2, and the similarity of the qe values from the model and the ones obtained experimentally. PFO fitted the experimental data better than PSO and assumes that a change of the solute uptake along time is directly proportional to the difference between sorbent saturation and the uptake along time [58]. The fitting parameters are shown in Table 3.12.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements Barros, O. | 2023 | 63 The qe is the theoretical capacity of the zeolite to retain REE at equilibrium, the highest one being obtained with ZX_NaOH 0.10 M. This corroborates that for this zeolite, the alkali treatment increased the REE retention as demonstrated before. The k1 is the affinity constant for the interaction between the REE and the zeolite. Table 3.12: Fitting parameters for PFO for the modified 13 X zeolites and respective controls. ZX_KOH 0.1 M ZX_NaOH 0.1 M ZX_H2O Z13X La DF 22 22 37 21 k1 0.043 0.044 0.062 0.046 qe 3.346 3.380 3.245 3.364 R2 0.975 0.969 0.962 0.984 Ce DF 21 26 26 21 k1 0.063 0.033 0.099 0.059 qe 2.813 4.442 3.220 3.747 R2 0.951 0.986 0.925 0.983 Y DF 22 22 37 32 k1 0.051 0.045 0.059 0.045 qe 2.971 2.934 2.809 2.681 R2 0.964 0.974 0.958 0.970 Tb DF 26 19 29 23 k1 0.046 0.048 0.052 0.032 qe 3.231 3.063 3.391 3.211 R2 0.962 0.967 0.970 0.945 Pr DF 17 26 30 23 k1 0.056 0.040 0.075 0.042 qe 2.822 3.094 2.753 2.823 R2 0.982 0.985 0.977 0.962 Eu DF 19 20 30 20 k1 0.056 0.044 0.055 0.029 qe 3.011 3.824 3.773 3.939 R2 0.969 0.980 0.962 0.949 DF — degrees of freedom; qe — adsorption capacity at equilibrium calculated from the fitting (mg/g); k1 — affinity constant of the pseudo-first order model (min - 1); R2 — coefficient correlation. The kinetic parameters for REE adsorption by ZX_NaOH 0.10 M were compared with the ones obtained with other inorganic materials and the results are shown in Table 3.13. ZX_NaOH 0.10 M has one of the highest qe values among the considered sorbents and the kinetic parameters indicate a faster interaction of that modified zeolite with the REE in solution. The ratio between the different REE concentrations and the adsorbent concentration certainly also determine the differences noticed.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 70 Pr ZX_KOH 0.10 M No (ns) No (ns) ZX_KOH 0.25 M No (ns) No (ns) ZX_KOH 0.50 M No (ns) No (ns) ZX_NaOH 0.10 M No (ns) No (ns) ZX_NaOH 0.25 M No (ns) No (ns) ZX_NaOH 0.50 M No (ns) No (ns) ZX_H2SO4 0.25 M No (ns) Yes (*) ZX_HCl 0.25 M Yes (*) Yes (**) ZX_HNO3 0.25 M Yes (*) Yes (**) Eu ZX_KOH 0.10 M No (ns) No (ns) ZX_KOH 0.25 M No (ns) No (ns) ZX_KOH 0.50 M No (ns) Yes (*) ZX_NaOH 0.10 M No (ns) No (ns) ZX_NaOH 0.25 M No (ns) No (ns) ZX_NaOH 0.50 M No (ns) No (ns) ZX_H2SO4 0.25 M No (ns) Yes (***) ZX_HCl 0.25 M No (ns) Yes (***) ZX_HNO3 0.25 M No (ns) Yes (****)
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 71 Table S-3.2: Statistical differences using the Bonferroni’s multiple comparisons test between the modified zeolite and the controls (Z4A and ZA_H2O). REE Modified zeolite Significantly different from ZX_H2O Z13X La ZA_KOH 0.10 M No (ns) No (ns) ZA_KOH 0.25 M No (ns) No (ns) ZA_KOH 0.50 M No (ns) Yes (**) ZA_NaOH 0.10 M No (ns) No (ns) ZA_NaOH 0.25 M No (ns) Yes (*) ZA_NaOH 0.50 M Yes (*) Yes (****) Ce ZA_KOH 0.10 M No (ns) No (ns) ZA_KOH 0.25 M No (ns) No (ns) ZA_KOH 0.50 M No (ns) Yes (**) ZA_NaOH 0.10 M No (ns) No (ns) ZA_NaOH 0.25 M No (ns) Yes (***) ZA_NaOH 0.50 M Yes (**) Yes (****) Y ZA_KOH 0.10 M No (ns) No (ns) ZA_KOH 0.25 M No (ns) Yes (***) ZA_KOH 0.50 M No (ns) Yes (**) ZA_NaOH 0.10 M No (ns) No (ns) ZA_NaOH 0.25 M Yes (*) Yes (****) ZA_NaOH 0.50 M Yes (**) Yes (****) Tb ZA_KOH 0.10 M No (ns) No (ns) ZA_KOH 0.25 M Yes (**) Yes (***) ZA_KOH 0.50 M No (ns) Yes (**) ZA_NaOH 0.10 M No (ns) No (ns) ZA_NaOH 0.25 M Yes (***) Yes (****) ZA_NaOH 0.50 M Yes (**) Yes (***) Pr ZA_KOH 0.10 M No (ns) No (ns) ZA_KOH 0.25 M No (ns) No (ns) ZA_KOH 0.50 M No (ns) Yes (**) ZA_NaOH 0.10 M No (ns) No (ns) ZA_NaOH 0.25 M Yes (*) Yes (***) ZA_NaOH 0.50 M Yes (**) Yes (****) Eu ZA_KOH 0.10 M No (ns) No (ns) ZA_KOH 0.25 M Yes (**) Yes (***) ZA_KOH 0.50 M Yes (*) Yes (**) ZA_NaOH 0.10 M No (ns) No (ns) ZA_NaOH 0.25 M Yes (****) Yes (****) ZA_NaOH 0.50 M Yes (**) Yes (***)
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 72 Figure S-3.1: Graphical representation of the elbow method to select the best option: A) PCA, B) K-Means and C) KNN Classifier accuracy for the training and test sets.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 73 Table S-3.3: Statistical resume for the desorption from Z13X. A multi-comparison test of the results was performed where Y, yes, is used when there is a statistical difference between the conditions tested. N, no, means that there is no statistical difference. This test was performed for the different REE. The order of the results is: La, Ce, Y, Tb, Pr and Eu. ZX_H2O HCl H2SO4 N; N; N; N; N; N HNO3 N; N; N; N; N; N N; N; N; N; N; N ZX_KOH 0.1 M HCl Y; Y; Y; Y; Y; Y Y; N; Y; Y; Y; Y Y; Y; N; Y; Y; Y H2SO4 Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y N; Y; N; Y; Y; Y N; N; N; N; N; N HNO3 Y; Y; N; Y; N; Y Y; Y; N; Y; Y; Y Y; Y; N; Y; N; Y N; N; N; N; N; N N; N; N; N; N; N Z13X HCl Y; Y; Y; Y; N; N N; N; Y; N; N; N Y; Y; Y; Y; Y; N Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y H2SO4 Y; Y; Y; Y; Y; Y Y; N; Y; N; N; N Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y N; N; N; N; N; N HNO3 Y; N; Y; Y; N; N Y; N; Y; N; N; N Y; Y; Y; Y; Y; N Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y N; N; N; N; N; N N; N; N; N; N; N ZX_NaOH 0.1 M HCl Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; N; Y; Y; Y N; N; N; N; N; N N; N; N; N; N; N N; N; N; N; N; N Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y H2SO4 Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; N; Y; Y; Y N; N; N; N; N; N N; N; N; N; N; N N; N; N; N; N; N Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y N; N; N; N; N; N HNO3 Y; Y; N; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; N; Y; Y; Y N; N; N; N; N; N N; N; N; N; N; N N; N; N; N; N; N Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y Y; Y; Y; Y; Y; Y N; N; N; N; N; N N; N; N; N; N; N HCl H2SO4 HNO3 HCl H2SO4 HNO3 HCl H2SO4 HNO3 HCl H2SO4 HNO3 ZX_H2O ZX_KOH 0.1 M Z13X ZX_NaOH 0.1 M
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 74 Table S-3.4: Fitting parameters and square errors for Pseudo-First Order, PFO, and Pseudo-Second Order, PSO, models for the selected modified zeolite 13 X and respective controls. REE Zeolite designation DF PFO PSO k1 qe R2 k2 qe R2 La ZX_KOH 0.1 M 22 0.043 3.346 0.975 0.010 4.132 0.963 ZX_NaOH 0.1 M 22 0.044 3.380 0.969 0.009 4.325 0.951 ZXH2O 37 0.062 3.245 0.962 0.022 3.714 0.963 Z13X 21 0.046 3.364 0.984 0.011 4.093 0.979 Ce ZX_KOH 0.1 M 21 0.063 2.813 0.951 0.021 3.274 0.922 ZX_NaOH 0.1 M 26 0.033 4.442 0.986 0.005 5.880 0.983 ZXH2O 26 0.099 3.220 0.925 0.036 3.571 0.923 Z13X 21 0.059 3.747 0.983 0.015 4.436 0.972 Y ZX_KOH 0.1 M 22 0.051 2.971 0.964 0.015 3.581 0.945 ZX_NaOH 0.1 M 22 0.045 2.934 0.974 0.011 3.694 0.955 ZXH2O 37 0.059 2.809 0.958 0.023 3.235 0.955 Z13X 32 0.045 2.681 0.970 0.015 3.224 0.963 Tb ZX_KOH 0.1 M 26 0.046 3.231 0.962 0.011 3.998 0.944 ZX_NaOH 0.1 M 19 0.048 3.063 0.967 0.013 3.727 0.949 ZXH2O 29 0.052 3.391 0.970 0.016 3.973 0.977 Z13X 23 0.032 3.211 0.945 0.009 3.980 0.963 Pr ZX_KOH 0.1 M 17 0.056 2.822 0.982 0.017 3.395 0.960 ZX_NaOH 0.1 M 26 0.040 3.094 0.985 0.010 3.822 0.972 ZXH2O 30 0.075 2.753 0.977 0.031 3.116 0.976 Z13X 23 0.042 2.823 0.962 0.006 4.365 0.970 Eu ZX_KOH 0.1 M 19 0.056 3.011 0.969 0.016 3.621 0.947 ZX_NaOH 0.1 M 20 0.044 3.824 0.980 0.009 4.781 0.962 ZXH2O 30 0.055 3.773 0.962 0.015 4.418 0.968 Z13X 20 0.086 2.369 0.950 0.039 2.692 0.949 DF — degrees of freedom; qe — adsorption uptake at equilibrium calculated from the fitting (mg/g); k1 — affinity constant of the pseudo first-order model (min - 1); k2 — affinity constant of the pseudo second-order model (g · mg - 1 · min - 1); R2 — coefficient correlation.
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 75 Table S-3.5: Confidence intervals for the parameters of Pseudo-First Order, PFO, and Pseudo-Second Order, PSO, kinetic models for every zeolite and REE tested. REE Designation ZX_KOH 0.1 M ZX_NaOH 0.1 M ZXH2O Z13X La DF 22 22 37 21 PFO k1 0.035 to 0.053 0.034 to 0.055 0.053 to 0.073 0.039 to 0.053 qe 3.187 to 3.523 3.169 to 3.622 3.139 to 3.356 3.233 to 3.500 PSO k2 0.006 to 0.016 0.005 to 0.015 0.017 to 0.028 0.008 to 0.015 qe 3.766 to 4.606 3.834 to 4.993 3.545 to 3.900 3.822 to 4.414 Ce DF 21 26 26 21 PFO k1 0.049 to 0.080 0.028 to 0.038 0.075 to 0.131 0.051 to 0.069 qe 2.640 to 2.993 4.230 to 4.684 3.041 to 3.402 3.596 to 3.906 PSO k2 0.012 to 0.035 0.004 to 0.006 0.02314 to 0.05449 0.011 to 0.021 qe 2.939 to 3.675 5.436 to 6.409 3.318 to 3.853 4.122 to 4.797 Y DF 22 22 37 32 PFO k1 0.041 to 0.064 0.037 to 0.054 0.050 to 0.070 0.038 to 0.052 qe 2.815 to 3.136 2.775 to 3.110 2.709 to 2.913 2.578 to 2.790 PSO k2 0.009 to 0.023 0.007 to 0.018 0.017 to 0.031 0.011 to 0.020 qe 3.246 to 4.001 3.311 to 4.192 3.067 to 3.424 3.019 to 3.463 Tb DF 26 19 29 23 PFO k1 0.037 to 0.056 0.039 to 0.059 0.044 to 0.061 0.023 to 0.044 qe 3.056 to 3.419 2.873 to 3.266 3.253 to 3.537 2.880 to 3.650 PSO k2 0.006969 to 0.01681 0.007903 to 0.02036 0.01229 to 0.02007 0.005 to 0.014 qe 3.618 to 4.471 3.330 to 4.211 3.774 to 4.194 3.529 to 4.579 Pr DF 17 26 30 23 PFO k1 0.047 to 0.066 0.034 to 0.045 0.066 to 0.086 0.033 to 0.053 qe 2.684 to 2.968 2.962 to 3.240 2.671 to 2.837 2.626 to 3.051 PSO k2 0.011 to 0.027 0.007 to 0.012 0.025 to 0.039 0.009 to 0.019 qe 3.064 to 3.792 3.676 to 4.347 2.991 to 3.250 3.150 to 3.840 Eu DF 19 20 30 20 PFO k1 0.045 to 0.068 0.037 to 0.051 0.046 to 0.067 0.021 to 0.040 qe 2.827 to 3.207 3.629 to 4.032 3.607 to 3.950 3.505 to 4.530 PSO k2 0.010 to 0.025 0.0057 to 0.013 0.011 to 0.020 0.003 to 0.010 qe 3.237 to 4.082 4.313 to 5.348 4.168 to 4.701 4.283 to 5.868
Chapter 3 Chemical modification of zeolites for the recovery of Rare Earth Elements using Machine Learning algorithms Barros, O. | 2023 | 76 Figure S-3.2: Two kinetic models fitting to the different REE for the 13X zeolite.
| 77 4. Chapter 4 – Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Unsupervised machine learning (ML) technique is applied to the characterization of the adsorption of Rare Earth Elements (REE) by zeolites in continuous flow. The successful application of PCA and K-means algorithms from ML allowed a wide range assessment of the adsorption results. This global approach permits the evaluation of the different stages of the sorption cycles and the development of possible optimizations and improvements. The results from ML are also used for the definition of a regression model to estimate other REE recoveries based on the known values of the tested REE. Overall, it was possible to remove more than 70% of all REE from aqueous solutions during the adsorption assays and to recover over 80% for the REE entrapped on the zeolite, using the optimized desorption cycle. Adapted from: Barros, O.; Parpot, P.; Neves, I. C.; Tavares, T.; Exploring optimization of zeolites as adsorbents for Rare Earth Elements using Supervised Machine Learning techniques – under revision.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 78 4.1. Introduction The continuous research and progress have resulted in a significant surge of available data, motivating some sectors of our society to reposition themselves and harness the disruptive potential of data analytics and machine learning [1]. Machine learning (ML) is an evolving branch of computational algorithms, whose development led to statistical models that can make predictions and support decisions without being explicitly programmed [2–5]. ML can integrate multimodality multi-fidelity data to reveal correlations between different features [6]. It has been applied successfully in diverse fields such as pattern recognition, medicine, science, computer vision, spacecraft engineering, engineering, biomedicine, psychology, catalysis, neurobiology and many other disciplines [1,5,7]. This wide application allows a faster treatment of great amounts of data and therefore ML can be used to analyze and correlate those data to achieve better interpretations and, therefore, to make better decisions. ML models and their importance have been recognized and appreciated in wastewater treatment [8–10]. Some developments have been made to use ML algorithms or deep learning neural networks for the optimization of the adsorption of antibiotics [11,12], organic compounds [13,14] and metals [15– 17]. The advantages of the ML techniques applied to the recovery of Rare Earth Elements (REE) from aqueous solutions using zeolites as adsorbents are described. REE represents 19% of the metals used in the technology and in the precious metals sectors, which accounts for 0.05% of world metal production and the trend is upwards [18,19]. REE plays a crucial role in the materials industry across various domains such as phosphors, magnets, metallurgy, catalysts and glass since the 1950s. They are frequently employed as additives or dopants in materials formulations. REE are particularly valuable due to their ability to induce significant changes in material properties, even when used in small quantities. Consequently, they have earned the reputation of being the "vitamins" of modern industry and the design of materials doped with rare earths has emerged as indispensable for technological advances [19]. Zeolites are porous aluminosilicate materials known for their highly structured crystalline network composed of alumina and silica tetrahedra (TO4). The presence of alumina induces a negative charge on the structure that is compensated by cations. Within the zeolitic structures available in the commercial market, zeolites of the LTA type and FAU type (faujasite, including zeolites X and Y) are commonly applied in various fields [20].The FAU structure exhibits a low Si/Al ratio, which results in a high cation exchange capacity mainly for cations with high charge density [20], such as rare earth elements (REE ions). So, this structure present goods properties for adsorption of these metals.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 79 The application of ML algorithms has been applied to REE separation techniques [21] and to adsorption [22]. Therefore, the objective of this study is to optimize the removal of REE (La, Eu, Pr, Ce, Tb and Y) by adsorption on FAU structures in continuous flow assays, employing ML techniques for evaluation and system development. 4.2. Material and Methos 4.2.1. Materials The materials used in this chapter (REE, zeolite FAU – 13X) are the same as the ones described in Chapter 3, 3.2.1 Materials. 4.2.2. Analytical quantification of REE The analytical quantification of the REE was performed as explained in the Chapter 3, 3.2.4 Analytical quantification of REE. 4.2.3. Continuous flow assays The continuous flow assays were carried out using 150 g of zeolite with and without modifications as the bed of column (height of 30 cm and diameter of 4.2 cm) set-ups with up-flow feeding, as represented in Figure 4.1. Figure 4.1: Schematic representation of the columns set-up (A) and the real system (B). The zeolite modification was the same as the one selected as the most suitable one in the Chapter 3. The modified zeolite will be designed as ZX_NaOH, while the zeolite without modification will be designed as Z13X. The column designations are shown in Table 4.1.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 86 obtained and from these results, it can be concluded that no condition achieved optimal removal and recovery in the same cycle. Figure 4.3: ML analysis for the second analysis to select the best condition cycle: A) PCA analysis; B) K-Means algorithm. The numbers after the designations are referencing to their respective cycles.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 87 A heatmap showing the Pearson correlations of the tested features is presented in Figure 4.4. Figure 4.4: Heatmap representation of the correlation between different features used for the cycle evaluation. The left scale represents the different correlation values and the respective colors. The heatmap shows 3 crucial relations. In a green square, the first one is the inverse correlation between the starting zeolites (Z13X and ZX_NaOH) and the washed samples (NW and WW, without and with washing, respectively). There is negative correlation for these two groups, which is expected since each sample is one zeolite or the other and the same happens regarding the washing. The second square (light blue) refers to the correlation of the cycles and the different removals and recoveries of the REE. The first cycle is the one with the stronger correlation with both removal and recovery of the REE. As seen in Figure 4.4, the removal and recovery correlation got more negative,
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 88 inducing a poorer performance as the number of cycles grew. This supports that both the adsorption and the desorption lose efficiency over the cycles. The last relation appears between the REE removals and recoveries. Each REE removal has a very high positive correlation (over 0.9) [24] with the removal of other REE, showing how the matrix entrapment of the different REE is similar and correlated. These analyses support some other observations in batch assays reported in the Chapter 3. Each REE removal and recovery have a moderate correlation (between 0.5 and 0.7) [24]. These correlations were expected to be stronger and this could be related to the fact that relatively poor recoveries are translated into a lower removal in the next cycle. A further evaluation of this hypothesis will be performed in the following sections. Each REE recovery has a very high positive correlation with the other recoveries. This is related to the fact that the oxidation number is the same (3+) for all the tested REE and even with some differences regarding the ionic radii, they behave similarly in the desorption process. The rest have a negligible correlation [24], except for the correlation between the different cycles, which has a low negative correlation [24]. This is due to the possible influence that incomplete recoveries would have over the removing in the following cycle. The results of the adsorption and desorption will be analyzed separately in the following sections to understand better what may be improved. This analysis, especially for the desorption, could indeed give an important insight of what happened during the assay. A cycle analysis will be performed, since the overall results for both the adsorption and desorption (Erro! A origem da referência não foi encontrada., Figure S-4.1 and Figure S-4.2) are too overlapped, with dense information that makes the analyze too complex. 4.3.2. Sorption analysis of the continuous flow assays cycles 4.3.2.1. Adsorption analysis The removal (%) is the result measured for specific time points 24, 48 and 72 h. The results are shown in Figure S-4.6 to Figure S-4.9 for the different cycles. The results of the removal for the first cycle are similar in each analyzed time point, with no significant differences. As expected, the removal values increased over time, confirming the REE adsorption by the zeolite samples. For the second cycle, the total removal shows similar results, with most of the statistical tests having no significant difference. The same behavior would be expected for the subsequent cycles. The results of total removal for the third cycle and fourth cycle show that no significant differences were found. These results indicate
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 89 that the ion exchange capacity of the zeolite attains the equilibrium, which is more visible for these cycles than the one obtained for the second cycle since a higher concentration was used . The total removal of the REE for each tested condition was calculated and the results are shown in Table 4.4. Table 4.4: Total removal of each REE for each zeolite tested after 4 cycles. Removal (%) La Ce Y Tb Pr Eu Z13X_NW 81.6 ± 6.5 83.3 ± 5.8 80.4 ± 6.7 83.8 ± 6.0 84.7 ± 5.7 83.6 ± 5.5 Z13X_WW 71.9 ± 0.4 74.7 ± 0.2 71.4 ± 2.2 76.2 ± 1.8 76.4 ± 0.8 75.6 ± 2.0 ZNaOH_NW 73.0 ± 3.0 73.6 ± 2.6 83.0 ± 2.0 72.7 ± 2.3 73.2 ± 2.9 72.7 ± 2.5 ZNaOH_WW 68.2 ± 4.1 70.9 ± 4.8 75.1 ± 4.5 72.3 ± 2.4 70.9 ± 3.6 70.7 ± 3.3 Overall, between 65 to 90 % of the total mass of REE present in the solutions to be treated were removed and the zeolite reveals similar affinity towards the different REE, for each of the four conditions (Table 4.4). No significant difference was found between the tested conditions (Table S-4.1), including the control zeolite, Z13X and the ZX_NaOH, as foreseen by previous batch assays, described in Chapter 3. This suggests that a pre-treatment of the zeolites will not improve the sorbent behavior. A difference between the performance of zeolites with (WW) and without (NW) NaOH washing between cycles was forseen. It was expected that the washed beds (Z13X_WW and ZNaOH_WW) would reach higher removals as the OHfrom the NaOH could neutralize part of the H+ from HNO3, used in the desorption step. As this was not observed, probably the NaOH concentration used, 0.01M, was not enough for the purpose as it is 10 times lower than the acid concentration. 4.3.2.2. Desorption analysis The desorption results for the four different cycles are shown in Figure S-4.10 to Figure S4.13. The desorption results for the first cycle are shallow, below 30 %, with no significant difference found for the different comparisons evaluated. For the second cycle, the desorption recovered below 11 %, with no significant differences. The same occurred for the third cycle, with recoveries below 10 % and for the fourth cycle, with recoveries below 14 %. No significant difference was found. These results were not expected since the same concentration was used for the acid chosen to be the best one in batch assay. The same assessment reported that the NaOH 0.1 M zeolite had very high recovery.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 90 The low recoveries of REE probably are related to the saturation of the zeolite surface [25]. Even though, this was not evident for the second cycle due to the lower initial REE concentration of 10 mg/L, supported by the adsorption results (Figure S-4.7). The third and fourth cycles had higher initial REE concentrations of 60 and 25 mg/L, respectively, but the low removal is more evident, as supported by the adsorption results (Figure S-4.8 and Figure S-4.9). This shows the importance of the desorption step in a multiple cycle assay in continuous flow through a column bed since good desorption may lead to a near total recovery of the REE from the zeolite. After that and during the second adsorption step, the zeolite would be available to remove more REE from the solution. The total recovery was calculated and shown in Table 4.5. Table 4.5: Total recovery percentage of each REE for each zeolite sample tested after 4 cycles. Recovery (%) La Ce Y Tb Pr Eu Z13X_NW 9.5 ± 0.1 9.2 ± 0.2 11.0 ± 0.7 10.9 ± 1.1 9.4 ± 0.1 11.5 ± 0.5 Z13X_WW 7.9 ± 1.1 8.4 ± 1.1 10.7 ± 0.6 10.1 ± 0.7 8.3 ± 1.0 11.2 ± 0.6 ZNaOH_NW 11.7 ± 1.7 12.1 ± 2.2 10.5 ± 4.5 13.6 ± 3.5 13.8 ± 2.1 15.8 ± 2.9 ZNaOH_WW 14.9 ± 1.6 13.9 ± 0.8 14.9 ± 1.8 17.5 ± 0.9 16.0 ± 1.3 18.9 ± 1.1 The overall desorption results were below 20 %, with no significant differences (Table S-4.2). These results were meager and unexpected since higher recoveries were achieved in batch assays. This indicates that a higher concentration of the acid should be used in these assays, so the desorption step needs to be improved. After the desorption process, the REE need to be concentrated so they be reused into new applications. The purification process may be consider the REE precipitation by addition of anions such as carbonate [26,27] or oxalate [28–32]. 4.3.2.3. ML analysis of the desorption optimization In the desorption batch assays from Chapter 3, 0.35 g of loaded zeolite was used with 0.1 L of HNO3 at 0.1 M, which defines a ratio of 28.6 mmol of HNO3 per g of zeolite. In these continuous flow assays, the ratio is 0.67 mmol of HNO3 per g of zeolite. The proportion between the batch ratio and the column ratio is 43, which explains the reduced desorption efficiency in continuous flow assays. So, it was decided to perform new desorption assays in the continuous flow set-up with a ratio of 13.3 mmol of HNO3 per g of the zeolite (2 L of 1 M of HNO3 during 3 h with the same flow rate in close loop). The results obtained from the optimized desorption cycle are shown in Figure 4.5.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 91 Figure 4.5: Total recovery for the different REE from loaded Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the 1 M eluent desorption. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent. The changes implemented for the desorption were definitive to improve the REE recoveries as seen in Figure 4.5. After 1 h of assays, over 70 % of La, Ce and Pr were removed from the zeolite, while the other REE had a lower recovery. This difference could be related to the accessibility of the cations located in the zeolite surface. The structural framework of zeolite Y or X (FAU) are distinguished by three main units: the hexagonal prism, the sodalite cavity and the supercage [33]. The recovery of these cations is facilitated if they are primarily located in the surface sites of the supercage or sodalite cavities. The different REE radii could be justify the observations since a smaller ion could easily enter into smaller structures of the zeolite and therefore, it would require more time to desorb from the zeolite. After a 3 h leaching, the recoveries of REE are similar between them and over 80 %. It is important to add that no significant differences were found between the tested conditions and that the primary source of variation is the time for all tested REE.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 92 The last time point results of all tested desorption cycles (the first 4 and the one with higher acid concentration) were compared for the same tested conditions. A significant difference, Table S-4.3, was found for all REE when any cycle is compared with cycle 5 (desorption with 1 M acid), as expected since this desorption presented recoveries up to 4 times higher than the ones from the previous cycles. Here, it was found that the primary source of variation is the cycle for all tested REE. After this desorption, an improved removal of the REE from the solution would be expected in a possible new adsorption cycle. This new adsorption cycle results could be better than to the ones obtained for cycles 3 and 4 since there will remain almost no REE in the zeolite. However, it is important to remember that the treatment could lead to changes in the structure, as a dealumination, which could reduce the negative charge of the zeolite and lead to a reduction of the REE adsorption. The previous results show a great potential to produce a positive supervised ML analysis, since the results from Figure 4.5 suggest that in this last cycle could have a higher mean value above 3.5, accordingly to Table 4.3. For that reason, a new DataFrame was built just considering the desorption results. This DataFrame consists in the previous 4 cycles and the results obtained from this new desorption cycle, with higher ratio of acid/zeolite. This DataFrame was used in a new ML analysis to investigate the impact of the amount of acid on the desorption efficiency. From the results in Figure S-4.14A, four components were selected to build the PCA, confirmed by the Knee Locator method [23] and the result is shown in Figure S-4.15. The first 2 features explain 63.95 % of the variance of the samples as shown in Figure 4.6A. The PCA distribution results in two main groups, one on the right side, more influenced by cycle 5 (desorption with 1 M acid) and the recoveries of the different REE. The second group, on the left side, is more influenced by the first 4 cycles. This group can be divided into 2 smaller groups: the top group was more influenced by the ZX_NaOH zeolite, while the bottom group was more influenced by the Z13X. Each zeolite group can be further divided into two groups depending on the washing after cycles, with NW in the top and WW in the bottom. The K-Means were made using three groups, as shown in Figure S-4.14B, confirmed by the Knee Locator method [23]. A clear group 1, in blue, which is from the ZX_NaOH zeolite for both NW and WW, can be seen in Figure 4.6B. The group 2, in green, is the fifth cycle of desorption with both zeolites and group 3, in purple is the Z13X zeolite. The other groups are mixed.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 93 Figure 4.6: ML analysis: A) PCA; B) K-Means. The Rec refers to the recovery of each REE.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 94 The ML classification algorithms were used to select the best desorption conditions, using the binary classification (Table 4.3: Binary classification used for each sample evaluated regarding the data from the adsorption (removal) and the desorption (recovery) assays.Table 4.3). For this case, 4 samples were considered good accordingly to Table 4.3. The 4 selected samples are the ones from the cycle 5, as expected, since the results shown in Figure 4.5 show that this cycle was the best one. With this, the classification was carried out using 4 classifiers, KNN, Decision Tree, Random Forest and Logistic Regression. The results of the different classifiers are shown in Figure 4.7. Figure 4.7: Conditions division according to the classifiers, A) KNN and logistic regression and B) decision tree and random forest. The different colors, violet and orange, represent the zone of a good or bad catalyst, respectively. For the KNN, Figure 4.7A, three neighbors were selected accordantly to the accuracy values for both training and test sets, shown in Figure S-4.14C. The Decision Tree classifier, Figure 4.7B, has only one parameter, random state, with a value of 20. For the Random Forest, Figure 4.7B, the n estimator parameter was 10 and the random state was the same as for the Decision Tree classifier. Finally, the same value for the random state parameter was used for the Logic Regression, Figure 4.7A.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 95 It is crucial to avoid overfitting of the training set for the classifiers, as it happens often with the model. A suitable generalization of the model from the training set can lead to a good classification of new and unseen data, which is the test set. All the tested classifiers could separate the 2 groups without any problem. Therefore, it is expected that the respective accuracy scores of the values (x and y values of training and test data) would be 100 %. The scores of the training and test using the four different classification algorithms are 100 % for all tested classifiers. Similar results were obtained using the classification report, which summarizes percentages of precision, recall and f1-scores. It is important to know that the precision is related to the accuracy of making good predictions, the recall is the value of the correctly identified positive predictions and the f1-score is the harmonic mean of the precision and recall. This evaluation used the real classification from the binary classification ( y_real ) and the predicted classification ( y_pred ) after training the model. All classifiers presented a 100 % score for the precision, recall and f1-scores of the prediction of the model. Another vital metric to assess the classification used is the confusion matrix. All the classifiers evaluated are very similar between them and the overall result is shown in Figure 4.8. Figure 4.8: Graphical representation of the confusion matrix for all tested classifiers. The values shown refer to the fraction of the true correct predictions (when the model got it right) and false incorrect predictions (when the model got it wrong).
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 102 Figure S-4.5: PCA maps in 3D.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 103 Figure S-4.6: Total removal for the different REE adsorption for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the first adsorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 104 Figure S-4.7: Total removal for the different REE adsorption for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the second adsorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 105 Figure S-4.8: Total removal for the different REE adsorption for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the third adsorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 106 Figure S-4.9: Total removal for the different REE adsorption for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the fourth adsorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent. Table S-4.1: Two-Way ANOVA for the total removal percentage of REE after 4 cycles. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Bonferroni's multiple comparisons tests REE La Ce Y Tb Pr Eu Z13X_NW vs. Z13X_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_NW vs. ZNaOH_NW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_NW vs. ZNaOH_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_WW vs. ZNaOH_NW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_WW vs. ZNaOH_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) ZNaOH_NW vs. ZNaOH_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns)
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 107 Figure S-4.10: Total recovery of the different REE from Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the first desorption step. The NW refers to the assays without the washing step and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 108 Figure S-4.11: Total recovery for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the second desorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 109 Figure S-4.12: Total recovery for the different REE for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the third desorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent.
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 110 Figure S-4.13: Total recovery for Z13X_NW ( ), Z13X_WW ( ), ZNaOH_NW ( ) and ZNaOH_WW ( ) for the first desorption cycle. The NW refers to the assays without the washing step, and the WW refers to the assays with the NaOH 0.01 M washing step. Samples were taken from the accumulation Erlenmeyer with the outflow eluent. Table S-4.2: Results for the Two-Way ANOVA for the total recovery percentage after 4 cycles. The NW refers to the assays without the washing step and the WW refers to the assays with the NaOH 0.01 M washing step. Bonferroni's multiple comparisons tests REE La Ce Y Tb Pr Eu Z13X_NW vs. Z13X_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_NW vs. ZNaOH_NW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_NW vs. ZNaOH_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_WW vs. ZNaOH_NW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) Z13X_WW vs. ZNaOH_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns) ZNaOH_NW vs. ZNaOH_WW No (ns) No (ns) No (ns) No (ns) No (ns) No (ns)
Chapter 4 Exploring optimization of zeolites as adsorbents for Rare Earth Elements in continuous flow assays Barros, O. | 2023 | 111 Table S-4.3: Statistical tests performed for all desorption cycles for each REE. The Two-Way ANOVA compares the total REE recovery from the zeolite. A comparison between each cycle for each tested condition was performed. The NW refers to the assays without the washing step and the WW refers to the assays with the NaOH 0.01 M washing step. The cycle with increased acid concentration is referred to as cycle 5. REE Bonferroni's multiple comparisons tests Two way ANOVA Z13X_NW Z13X_WW ZNaOH_NW ZNaOH_WW La Cycle 1 vs. Cycle 2 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 2 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 3 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 3 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 4 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Ce Cycle 1 vs. Cycle 2 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 2 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 3 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 3 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 4 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Y Cycle 1 vs. Cycle 2 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 2 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 3 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 3 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 4 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Tb Cycle 1 vs. Cycle 2 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 1 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 2 vs. Cycle 3 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 2 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 3 vs. Cycle 4 No (ns) No (ns) No (ns) No (ns) Cycle 3 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****) Cycle 4 vs. Cycle 5 Yes (****) Yes (****) Yes (****) Yes (****)
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 118 the lack of a filtration. The resulting solution was decanted, leaving the catalysts to be dried and then calcinated. The prepared heterogeneous catalysts are displayed in Table 5.1. Table 5.1: Designation and details of REE/Fe-zeolite catalysts and respective method of preparation. Samples Label Zeolite type pH Method La10Fe10NaX A3 FAU powder 4.00 impregnation La10Fe10NaX A7 FAU pellet 4.00 impregnation La10Fe10ZSM5 Z1 MFI 4.00 impregnation La10Fe10NaY Z2 FAU 4.00 impregnation La25Fe10ZSM5 Z3 MFI 4.00 impregnation La25Fe10NaY Z4 FAU 4.00 impregnation La10Fe10ZSM5 Z15 MFI 4.00 ion exchange La25Fe10ZSM5 Z16 MFI 4.00 ion exchange La25Fe10NaY Z17 FAU without adjustment ion exchange Ce10Fe10ZSM5 Z5 MFI 4.00 ion exchange Ce10Fe10NaY Z6 FAU 4.00 ion exchange Ce25Fe10ZSM5 Z7 MFI 4.00 ion exchange Ce25Fe10NaY Z8 FAU 4.00 ion exchange Pr10Fe10ZSM5 Z9 MFI 4.00 ion exchange Pr10Fe10NaY Z10 FAU 4.00 ion exchange Pr25Fe10ZSM5 Z11 MFI 4.00 ion exchange Pr25Fe10NaY Z12 FAU 4.00 ion exchange 5.2.3. Catalysts characterization Attenuated Total Reflectance Fourier Transform Infrared spectroscopy (ATR-FTIR) analysis was performed at room temperature using a PerkinElmer Spectrum Two spectrometer equipped with an ATR accessory. A diamond prism was used as the waveguide. All spectra were recorded with a resolution of 2 cm-1 in the wavelength region 4000-400 cm−1 by averaging 50 scans. Elemental quantification of the La, Ce, Pr and Fe in the solutions used for metal addition to the zeolites was performed using an ICP-OES spectrometer (Optima 8000, PerkinElmer). The REE quantification is similar to the one described in Chapter 3, 3.2.4 Analytical quantification of REE. The wavelength used for Fe was 238.204 nm with a radial plasma view.
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 119 The chemical analysis of the catalyst was performed to quantify La, Ce, Pr, Fe, Si, Al and Na in the solid samples (0.05 g). The samples were thermally treated at 500 ◦C for 12 h to remove the adsorbed water and subsequently placed in a platinum crucible. Then, the melting agent was added (Li2B4O7:sample = 15:1 by weight), and the alkaline fusion was carried out in a muffle furnace at 1000 ◦C for 40 min. After cooling of the melt, the resultant fusion bead was transferred into a beaker and heated on a plate at 80 ◦C after addition of 100 mL of 5% HNO3 (all the samples were found to completely dissolve within 40 min). Finally, the solution was transferred into a volumetric flask (250 mL) and diluted to the desired final volume with milliQ water. The resulting solution was analyzed with a 5110 Inductively Coupled Plasma Atomic Emission Spectroscopy (ICP-AES) spectrometer (Agilent Technologies). 5.2.4. Fenton-like reaction Catalytic runs were carried out in a semi-batch reactor at atmospheric pressure and 40 °C, under continuous stirring (300 rpm), using a solution of 30 mg/L of tartrazine or of indigo carmine, at pH of 3.00 and a specific concentration of hydrogen peroxide (H2O2). The concentration of H2O2 used for the tartrazine was 90 mM, while for indigo carmine was 12 mM. The concentration of dyes, the pH and temperature were fixed at optimal values determined in a preliminary evaluation of the degradation of organic pollutants using similar zeolite-based LaFe catalysts [20]. The runs were divided into initial screening (IS) and catalytic tests (CT). Once the assay started, samples were taken at fixed time intervals and the reaction was stopped with the addition of an excess of NaHSO3, which instantaneously consumes the unreacted H2O2. The suspension was centrifuged at 12000 rpm for 10 min, and the liquid was then analyzed for the pollutant by UV-vis. The quantification was performed using a microplate spectrophotometer Epoch 2 from Biotek, with the characteristic wavelengths (max) at 427 nm and 610 nm for Tar and IC, respectively. In order to select the best catalyst, an initial screening (IS) was carried out with sampling at the beginning of the run and after 180 min, using a catalyst concentration of 0.8 g/L with 0.5 mL of H2O2 at the respective concentration. The remaining conditions of the assay are equal to the ones described above. The catalytic tests (CT) were carried out with the REE/Fe-zeolite catalysts with liquid samples taken at 0, 15, 30, 60, 90, 120, 180, 240 and 300 min, using a catalyst concentration of 0.8 g/L. The effects of the H2O2 load were assessed by using 5 or 0.5 mL at the respective concentration. Usual Fenton-like reaction kinetics are described by a pseudo first-order model [24] that is used to evaluate the parameters of the degradation of the dyes and its non-linear equation is described by Eq. 5.1:
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 120 𝐶𝑡 𝐶0= 𝑒−𝑘∗(𝑊 𝑉)∗𝑡 (Eq. 5.1) Ct (mg/L) is the concentration of the dye at a given time t ; C0 (mg/L) is the initial concentration of dye; k (L/(g*min)) is the rate constant; V (L) is the volume of the solution; W (g) is the mass of catalyst used for the assay and t (min) is the time. Using Eq. 5.2, it is possible to simplify Eq. 5.1 to a more straightforward form (Eq. 5.3), since W and V are constant during the reaction: 𝑚=𝑘∗(𝑊 𝑉) (Eq. 5.2) 𝐶𝑡 𝐶0= 𝑒−𝑚∗𝑡 (Eq. 5.3) A linear form could be achieved according to the following equation: ln(𝐶𝑡 𝐶0)= −𝑚∗𝑡 (Eq. 5.4) 5.2.5. Machine learning analysis ML analysis was conducted using a table designed as DataFrame, with the different catalysts produced as rows, while the different columns or features were filled with the REE and Fe concentrations obtained by the catalyst chemical analysis, the ratio between these two concentrations and the Si/Al ratio. Adding to that information, the dyes degradation reached in the IS and CT assays was also considered as feature. Supervised and unsupervised ML approaches were applied to all catalytic results. For the unsupervised learner, the Principal Component Analysis, PCA, (method for reducing the dimensionality of data, leading to an increased interpretation and minimizing information loss) was used and K-Means clustering (division of the samples into groups or clusters that are more compatible with each other accordantly to the studied conditions). For this analysis a scaling or normalization of the data is required before using PCA and K-Means. The supervised learner uses a classification to divide the tested catalysts into good or bad ones, accordantly to their results. K-nearest neighbors classifier (KNN), Decision Tree Classifier, Random Forest and Logistic Regressor Classifier were used to classify the samples. These classifiers are often used in a binary system. In this study, a binary classification approach was employed to evaluate the performance of each catalyst with respect to the degradation of the dye. Based on the obtained degradation results, each dye was classified according to the percentage of conversion. A mean value taking into account the
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 121 results obtained for different catalysts and experimental conditions, allows the preliminary selection of the best catalysts. This approach enables the efficiently evaluation and comparison of the performances of the catalysts in a standardized manner (Table 3.2). To develop and evaluate the model's performance, the dataset was divided into two sets: a training set (70% of the data), which contains known output and enables the model to learn how to generalize and apply to new data, and a test set (30% of the data), which was used to evaluate the model's prediction accuracy. The data was divided using a stratified approach, ensuring that both the training and test sets had the same proportion of each class. Table 5.2: Binary classification used for each REE/Fe-catalyst. Degradation intervals, % Degradation Classification Classification means Binary Classification 80 < Deg < 100 1 4.75 1 60 < Deg < 80 2 40 < Deg < 60 3 20 < Deg < 40 4 4.75 0 0 < Deg < 20 5 All tests were performed using Spyder (Python 3.9) and the required modules for the python analysis as pandas, numpy, scikit-learn, matplotlib and seaborn. 5.2.6. Statistical analysis The initial screening (IS) results were statistically analyzed using One-Way ANOVA, through which all samples were compared between themselves. The catalytic tests (CT) results were analyzed using Two-Way ANOVA. Bonferroni’s multiple comparison test was used for the different comparisons performed. The tests were performed using the software Graph Pad Prism version 8.0.2 (Graph Pad Software, Inc, San Diego, CA, USA). The results were considered significantly different only when the probability ( p - value) was lower than 0.05, assuming a 95 % confidence interval.
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 122 5.3. Results and Discussion In order to select the best FAU catalyst and to compare with the MFI-based ones, Tar degradation was carried out with the catalysts prepared from a solution with an initial concentration of La of 10 ppm. In a previous work [10], NaX was used to remove REE from aqueous solutions and it was found that this particular zeolite was an effective adsorbent [10]. For this reason, NaX prepared by impregnation in the same experimental conditions used for the other REE/Fe-zeolite catalysts, is also included for evaluation as a heterogeneous catalyst (in pellet or powder) for the Fenton-like reaction (Figure 5.1). Figure 5.1: Tar degradation in the presence of La10Fe10ZSM5 (MFI): Z15, prepared by ion exchange ( ) and Z1, prepared by impregnation ( ); La10Fe10NaY (FAU), Z2, prepared by impregnation ( ), La10Fe10NaX powder (FAU), A3, prepared by impregnation ( ), and La10Fe10NaX pellet (FAU), A7, prepared by impregnation ( ). Conditions of the reaction: 20 mg of catalyst/25 mL of a 30 ppm solution of Tar; 0.5 mL of H2O2 90 mM; pH=3; T=40 0C; t = 180 min. The best results were obtained with the catalysts based on MFI structure (Z15 and Z1), followed by NaY (Z2) and NaX (A3 and A7), as shown in Table S-5.1. The significant differences calculated by the column analysis performed using One-Way ANOVA are shown in Table S-5.2. The last one, the powder form (A3) favors the Fenton-like reaction in comparison with the pellets (A7), as the external mass transfer limitations are reduced when the average size of the catalysts particles diminishes. The lower degradation efficiency obtained for NaX was expected since this type of zeolite is mainly used for adsorption processes [10] rather than for catalysis. Moreover, Z2 (La10Fe10NaY) reached higher conversion
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 123 than any of the NaX catalysts, as NaY zeolite is widely used in catalytic applications [36–38] since it enhances the catalytic role of the supported metal. In addition, Z15 (La10Fe10ZSM5), prepared by the ion exchange method, reached higher degradation efficiency than Z1 (La10Fe10ZSM5), prepared by the impregnation method, probably due to fact that the metallic active sites are better distributed on the internal surface area. Since NaY and ZSM5 loaded with catalytic metals act as bifunctional catalysts, enhancing the metal role within the desired reaction, they are expected to have advantage over NaX as supports for heterogeneous catalysis. These last REE/Fe-zeolite catalysts were analyzed by FTIR (Figure 5.2). Figure 5.2: FTIR spectra of the REE/Fe-zeolite catalysts in the spectral region of 2000 to 450 cm-1. The characteristic bands of the pristine zeolite structures (FAU and MFI) dominate the spectra of all REE/Fe-zeolite catalysts. The band characteristic of the (H2O) vibration mode of absorbed water on zeolite was identified at 1640 cm-1, whereas the typical bands of the lattice vibrations of the framework are evidenced in the 1330-450 cm-1 range [20,39–41]. The band at 960 cm-1 is attributed to the asymmetric stretching of Si–O and Al–O bonds belonging to SiO4 and AlO4 tetrahedra [39–41], whereas the bands at 670 cm-1 and near 750 cm-1 are related to the Si–O symmetric stretching and oscillations of aluminosilicate oxygen tetrahedral chains [40,41]. The band at about 550 cm-1 is attributed to the symmetric stretching vibrations of bridge bonds, Si–O–Si and bending vibrations of O–Si–O [42].
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 124 The granulometry of the NaX pristine structures may have an impact over the spectra of A3 (powder) and A7 (pellets) while the catalysts preparation method seems not to influence the spectra of the two ZSM5-based catalysts (Z1 and Z15). In addition, the framework Si/Al ratio of the samples based on FAU structure can be determined by FTIR analysis using the following Eq. 5.5 [42]: x=3.857-0.00621WDR (Eq. 5.5) where x = (1+Si/Al)-1 and WDR is the wavenumber at 500-650 cm-1, related to the vibrations of the FAU lattice [42] (Table 5.3). Table 5.3: Framework Si/Al ratios obtained from the FTIR analysis. Label Samples Framework Si/Al - NaY 2.80 - NaX 1.64 A3 La10Fe10NaX 1.44 A7 La10Fe10NaX 1.64 Z2 La10Fe10NaY 2.49 The framework Si/Al values of FAU-based catalysts show that Z2 and A3 were the most affected by the introduction of both metals, La and Fe. The reduction of the Si/Al ratio for A3 may be related with the acid character of the metals solution. The powder form of A3, with larger surface area, makes it more sensitive to its circumstances than A7 with a pelleted form and smaller specific surface area. The lower degradation performance of A3 and Z2 is probably related to the impregnation method used in their preparation, which affects more the zeolite structure than the ion exchange method. The impregnation method results in a weak metal-support interaction and large metal sites are obtained, while ion exchange reaches a finer metal dispersion [43], with reactional advantages. The larger metal sites resulting from the impregnation method lead to a reduced conversion efficiency. Based on the obtained catalytic results, ZSM5 and NaY structures were selected as the supports to prepare the REE/Fe-zeolite catalysts for degrading Tar and IC through the Fenton-like reaction. Tar degradation was carried out in the same reaction conditions as those in Figure 5.1 [20]. IC degradation was instead performed using two different H2O2 concentrations, 12 and 90 mM. It was proven (data not shown) that a 7.5-fold increase in H2O2 concentration is not justified, as it does not lead to any improvement in degradation performance after 180 min of reaction.
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 125 5.3.1. Selection of the best REE/Fe-zeolite catalyst using IS results The results for the PCA analysis are shown in Figure S-5.1A, after the scaling of the DataFrame. Two principal components, PCA 1 and PCA2 (variables created from the linear composites of the original variables with the highest variance), were selected to build the PCA, and the same value was obtained with the Knee Locator method [44]. The results obtained for PCA are shown in Figure 5.3A. These results are shown in a biplot where the bottom x and left y are used for samples distribution, while the top x and right y are used for the distribution of the different features. Figure 5.3: Graphical distribution of the ML analysis for the different catalysts: A) PCA analysis; B) K-Means algorithm. The IS and CT values are referred to initial screening and catalytical tests, respectively, for tartrazine (Tar) and for indigo carmine (IC). The cos of the angle between the analyzed features indicates their correlation. Values close to 1 (angle near 0o) indicate that features are directly correlated, values near -1 (angle near 180o) indicate
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 126 indirect correlation and values next to 0 (angle near 90o) show no correlation. The CT for Tar and IC, the IS for Tar and IS, the Si/Al ratio and the Fe/REE ratio seem to have some direct correlation between each other. This positive correlation suggests that the ratios Si/Al and Fe/REE might have an influence on the degradation of the tested dyes, foreseeing that the catalysts with higher ratios have better catalytic properties. It is reported in different applications that catalysts with higher Si/Al ratio have a higher activity and higher selectivity [45]. Apparently, there is no correlation with the preparation methods or with the Fe concentration. Adding to that, a negative correlation between degradation and REE concentration in the zeolites was found, suggesting that increasing concentrations of REE seem not to improve dyes degradation. A higher amount of REE on the catalysts surface might not imply a better dye degradation, just larger active sites not as efficient as smaller but more dispersed ones. The Fe concentration has a positive correlation with the impregnation preparation method and a negative correlation with the ion exchange, indicating that the catalysts produced by impregnation could have a higher Fe concentration than the ones made via ion exchange. This higher quantity of Fe could be related to the selectivity of this metal by the zeolite. The catalysts group division based on the results of the PCA could be performed in diverse ways and for that reason it was tested by K-Means algorithm, shown in Figure 5.3B. This analysis provided four clusters based on the Elbow method, presented in Figure S-5.1B and confirmed by the Knee Locator method. Group 1 consists in Z5, Z7, Z9, Z11, Z15 and Z16, group 2 in Z2, Z4 and Z17, group 3 in in Z1 and Z3 and finally, group 4 in Z6, Z8, Z10 and Z12. The groups 1 and 3 seem to be more influenced by the values of Si/Al and Fe/REE ratios as well as by the results of the degradation tests for both dyes. This suggests that these groups may include the best catalysts (possibly Z15, Z16 and Z3), mainly due to the stronger influence of the degradation results of Tar and IC. Group 2 appears to be primarily affected by the concentration of La and by the impregnation method, whereas in group 3 the key determinants were the concentrations of Pr and Ce, along with the ion exchange method. Important to mention that group 1 includes catalysts designed with all REE of interest, while group 3 and group 2 include only La catalysts and group 4 includes only Ce and Pr catalysts. The combination shown in Figure 5.3 helps to perform a division between groups considering the zeolite type and the preparation method. For example, the zeolite type division would consist of groups 1 and 3, both related to the higher Si/Al ratio, which is characteristic of the ZSM5 zeolite used as support for these catalysts, while groups 2 and 4 should have a lower Si/Al, characteristic of NaY zeolite. The preparation method division would consist in groups 2 and 3, as both used the impregnation method, while groups 1 and 4 used the ion exchange protocol. It is important to mention that the preparation
Chapter 5 Development of REE/Fe-zeolite catalysts for Fenton-like reaction Barros, O. | 2023 | 127 method division just includes the results for La catalysts, as this was the only REE involved in both methods. Therefore, these observations will help to assess the extent of the possible differences between the catalysts accordantly to the previously mention characteristics. The dye degradation obtained with the catalysts based on the two zeolite types and the REE concentration on the starting solution are highlighted in Figure 5.4. Figure 5.4: Degradation of Tar and IC using the REE/Fe-zeolite catalysts for A) La, B) Ce and C) Pr, after IS test. The catalysts are divided into ZSM5 (MFI) with a REE concentration of: 10 mg/L ( ) and 25 mg/L ( ); NaY (FAU) with a REE concentration of: 10 mg/L ( ) and 25 mg/L ( ). Conditions of the reaction: 20 mg of catalyst/25 mL at 30 ppm of dye; 0.5 mL of 90 mM of H2O2 for Tar and of 12 mM of H2O2 for IC; pH=3; T=40 0C and 3 h of reaction. The best catalytic results for the degradation of both dyes by Fenton-like reaction were obtained for ZSM5 used as support, as shown in Figure 5.4 and validated by the statistical differences presented in Table S-5.3. Remarkably, in the case of IC degradation, the performance of the catalysts is similar between La, Ce or Pr for the same support. The worst results considering these three metals were obtained with the NaY supported catalysts, with a slight IC degradation enhancement with the supported Ce or Pr (Table S-5.3). Overall, these results confirm the observations described in Figure 5.3A, with