Influence of Al2O3 Nanoparticles Addition in ZA-27 Alloy-Based Nanocomposites and Soft Computing Prediction

Citation: Vencl, A.; Svoboda, P.; Klanˇcnik, S.; But, A.; Vorkapi´c, M.; Harniˇcárová, M.; Stojanovi´c, B. Influence of Al2O3Nanoparticles Addition in ZA-27 Alloy-Based Nanocomposites and Soft Computing Prediction. Lubricants 2023,11, 24. https://doi.org/10.3390/ lubricants11010024 Received: 30 November 2022 Revised: 28 December 2022 Accepted: 30 December 2022 Published: 7 January 2023 Copyright: © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). lubricants Article Influence of Al2O3Nanoparticles Addition in ZA-27 Alloy-Based Nanocomposites and Soft Computing Prediction Aleksandar Vencl 1,2,* , Petr Svoboda 3, Simon Klanˇcnik 4, Adrian But 5, Miloš Vorkapi´c 6, Marta Harniˇcárová7,8 and Blaža Stojanovi´c 9,* 1University of Belgrade, Faculty of Mechanical Engineering, Kraljice Marije 16, 11120 Belgrade, Serbia 2South Ural State University, Lenin Prospekt 76, 454080 Chelyabinsk, Russia 3Faculty of Mechanical Engineering, Brno University of Technology, Technická2896/2, 616 69 Brno, Czech Republic 4Faculty of Mechanical Engineering, University of Maribor, Smetanova 17, 2000 Maribor, Slovenia 5Faculty of Mechanical Engineering, Politehnica University of Timis ,oara, Bulevardul Mihai Viteazu 1, 300222 Timis ,oara, Romania 6University of Belgrade, Institute of Chemistry, Technology and Metallurgy, Njegoševa 12, 11000 Belgrade, Serbia 7Institute of Electrical Engineering, Automation, Informatics and Physics, Faculty of Engineering, Slovak University of Agriculture, Tr. A. Hlinku 2, 949 76 Nitra, Slovakia 8Faculty of Technology, Institute of Technology and Business in ˇ CeskéBudˇejovice, Okružní517/10, 370 01 ˇ CeskéBudˇejovice, Czech Republic 9University of Kragujevac, Faculty of Engineering, Sestre Janji´c 6, 34000 Kragujevac, Serbia *Correspondence: [email protected] (A.V.); [email protected] (B.S.) Abstract: Three different and very small amounts of alumina (0.2, 0.3 and 0.5 wt. %) in two sizes (approx. 25 and 100 nm) were used to enhance the wear characteristics of ZA-27 alloy-based nanocomposites. Production was realised through mechanical alloying in pre-processing and compocasting processes. Wear tests were under lubricated sliding conditions on a block-on-disc tribometer, at two sliding speeds (0.25 and 1 m/s), two normal loads (40 and 100 N) and a sliding distance of 1000 m. Experimental results were analysed by applying the response surface methodology (RSM) and a suitable mathematical model for the wear rate of tested nanocomposites was developed. Appropriate wear maps were constructed and the wear mechanism is discussed in this paper. The accuracy of the prediction was evaluated with the use of an artificial neural network (ANN). The architecture of the used ANN was 4-5-1 and the obtained overall regression coefficient was 0.98729. The comparison of the predicting methods showed that ANN is more efficient in predicting wear. Keywords: ZA-27 alloy; Al 2 O 3 nanoparticles; nanocomposites; wear; response surface methodology; artificial neural network 1. Introduction Metal matrix nanocomposites (MMnCs) can be defined as composites with metal base and one or several incorporated nano-sized secondary phases (mainly in particulate form). As in metal matrix composites, these secondary phase/phases usually have physical and mechanical properties very different from those of the matrix. Nowadays MMnCs can be produced by using various processing techniques [ 1 ], but the uniform distribution of the secondary phase/phases throughout the matrix is still a challenging task. The MMnCs can possess higher fracture strength and toughness, hardness and wear resistance over the matrix material. This is obtained through several strengthening mechanisms, i.e., load transfer effect, Hall–Petch strengthening, Orowan strengthening, coefficient of thermal expansion (CTE) and elastic modulus (EM) mismatch [2]. Regarding the matrix material, the majority of the researchers study aluminiumor magnesium-based nanocomposites, and significantly fewer researchers deal with zinc Lubricants 2023,11, 24. https://doi.org/10.3390/lubricants11010024 https://www.mdpi.com/journal/lubricants Lubricants 2023,11, 24 2 of 13 alloy-based nanocomposites [ 3 ]. Among them, only a few investigated ZA-27 alloy-based nanocomposites with incorporated Al 2 O 3 nanoparticles. Shivakumar et al. [ 4 ] investigated the ZA-27 alloy-base nanocomposites reinforced with 1, 3 and 5 wt. % Al 2 O 3 nanoparticles (average size was 60–70 nm). Nanocomposites were obtained through stir casting accompanied by the squeeze casting technique. Wear tests were performed under the dry sliding conditions at four different normal loads (20, 40, 60 and 80 N) and four sliding speeds (1, 1.33, 1.67 and 2 m/s). The rotating counter-body was a hardened steel disc. Results showed that microhardness and wear resistance increase with the increase in the amount of nanoparticles. Çelebi et al. [ 5 ] produced ZA-27 alloy-base nanocomposites with 1 wt. % Al 2 O 3 nanoparticles (average size of 100 nm) through the hot pressing. Before the hot pressing, mechanical milling with variable time (from 1 to 8 h) was applied to mix the ZA-27 microparticles (average size of 40 µ m) and Al 2 O 3 nanoparticles. The results showed that an increase in milling time produced nanocomposites with higher porosity, but also higher macrohardness. On the other hand, the tensile strength of the nanocomposite initially increased and then rapidly decreased as the milling time increased. The main reason for these changes in the mechanical properties was the work hardening effect which makes the nanocomposites brittle. The work hardening was more pronounced for higher milling times. Bobi´c et al. [ 6 ] also analysed the influence of milling time (from 30 to 180 min) during the pre-processing of the mixture used in the production of nanocomposites (compocasting process). Nanocomposites were with the ZA-27 alloy base and reinforced with 0.2, 0.3 and 0.5 wt. % Al 2 O 3 nanoparticles of two different average sizes (25 and 100 nm). They found that the optimal time of 60 min provides improved macrohardness and compressive yield strength of the nanocomposites. They also noticed that all nanocomposites exhibited a finer structure of the matrix. The same nanocomposites, obtained with the optimal milling time, were also tested on erosion wear resistance with a solid particle impingement test (particles impact angle of 90 ◦ ) [ 7 ]. The results showed that the presence of Al 2 O 3 nanoparticles slightly increased erosive wear resistance due to the positive effect they have on the ductility of nanocomposites. Based on these positive results, the aim of this paper is to further study the ZA-27 alloy-based composites reinforced with 0.2, 0.3 and 0.5 wt. % Al 2 O 3 nanoparticles of two different average sizes (25 and 100 nm). The novelty is that until now, there has been no tribological research on nanocomposites with such a low amount of nanoparticles and very few that covered ZA-27 alloy-based nanocomposites with incorporated alumina nanoparticles in higher amounts. Therefore, the effect of size and amount of added Al 2 O 3 nanoparticles on microhardness and lubricated sliding wear resistance is investigated and analysed. 2. Experimental Details 2.1. Materials The matrix material used to produce nanocomposites was zinc–aluminium alloy ZA27, whose chemical composition is shown in Table 1. In total, six different nanocomposites were reinforced with Al 2 O 3 particles. The amount of added Al 2 O 3 particles varies (0.2, 0.3 and 0.5 wt. %), as well as their approximate sizes (25 and 100 nm). The equipment used for the manufacture of matrix alloy and nanocomposites is detailed elsewhere [8]. Table 1. Chemical composition (wt. %) of ZA-27 alloy. Element Al Cu Mg Zn Percentage 25–27 2.0–2.5 0.015–0.02 Balance The matrix alloy was processed by the thixocasting (semi-solid processing). In order to remove the slag from the melt surface, the alloy was overheated to 550 ◦ C. After that, the melt was cooled to 490 ◦ C (at a rate of 5 ◦ C/min) and its mixing at a rate of 500 rpm Lubricants 2023,11, 24 3 of 13 began. With further cooling, the temperature reached 465 ◦ C and at this temperature, the melt was mixed for 25 min. This intensive mixing resulted in the change in the structure of the primary particles of the matrix alloy, i.e., they become nondendritic. The melt was poured into the preheated (at 400 ◦ C) steel mould, and after cooling at room temperature, the casts were additionally hot-pressed (at 370 ◦C and under 250 MPa). The nanocomposites were produced by the compocasting process with mechanical alloying pre-processing (ball milling), which also started with preheating to 550 ◦ C and cooling to 490 ◦ C (with the introduction of mixing at 500 rpm). After that, the melt was cooled to 485 ◦ C and the mixing rate was reduced to 250 rpm. At this point, the infiltration of the previously prepared (during ball milling) mixture of nanoparticles and metal chips [ 6 , 7 ] started and lasted for 4 min. During the infiltration the temperature was cooled to 465 ◦ C, when the homogenisation of the composite melt started (still at 250 rpm) and lasted for 3 min. The mixing rate was then increased to 500 rpm and mixing continued for the next 20 min at a constant temperature. The nanocomposite melt was also poured into the preheated (at 400 ◦ C) steel moulds, and after cooling at room temperature the casts were additionally hot-pressed (at 370 ◦C and under 250 MPa). 2.2. Methods of Characterisation Plate-like samples (16 × 12 × 6 mm) were used for both microhardness and wear tests, i.e., microhardness was measured on the samples that were later used in wear testing. Measurements were according to the ASTM E384 standard. Vickers indenter and a load of 500 g (HV 0.5) with a dwell time of 15 s were used. At least five measurements were made for each sample in order to eliminate possible segregation effects and to obtain a representative value of the material microhardness. Wear tests were carried out on a block-on-disc tribometer, according to the ASTM G77 standard, in lubricated sliding conditions with line contact between the block and disc of 6 mm. Blocks were made of tested materials, whereas the disc (counter-body) was made of quenched and tempered steel EN 42CrMo4 (51–54 HRC). The surface roughness of blocks and counter-body was approximately Ra = 0.14–0.17 µ m and Ra = 0.6–0.7 µ m, respectively. The lubricant was mineral gear oil (ISO VG 220, ISO L-CKC/CKD). Test parameters were: 0.25 and 1 m/s (sliding speeds), 1000 m (sliding distance) and 40 and 100 N (normal loads). The temperature of the oil was continually measured throughout each test. The average steady-state value of the temperature, i.e., the value at the end of the test was in the range of 30 ± 2 ◦ C. The wear scar on the block was measured optically with an accuracy of 0.05 mm, after each test, in order to calculate the volume loss and compute the wear rate. Additionally, wear scars were utilised to determine the average pressure on contact surfaces at the end of the test, which in some cases reached 24 MPa. After testing, the worn surfaces of blocks were examined using the scanning electron microscope (SEM) equipped with an energy dispersive spectrometer (EDS). Samples were prepared through carbon evaporation where the thin carbon layer of 15–25 nm was deposited on samples. 2.3. Experimental Design The used experimental design is presented in Table 2. The dependent variable was the wear rate of the tested nanocomposites, while the factors that affect this variable were Al 2 O 3 amount, Al 2 O 3 size, sliding speed and normal load. The first factor (Al 2 O 3 amount) had four levels, and the other three factors (Al 2 O 3 size, sliding speed and normal load) had two levels. Experimental results were analysed by applying the response surface methodology (RSM). The advantage of RSM is that it can provide quantitative measurements of possible interactions between influencing factors, which are difficult to obtain by other techniques [ 9 ]. The main idea of RSM is to use a sequence of designed experiments to obtain an optimal response [ 10 , 11 ]. It applies several mathematical and statistical techniques to develop an adequate functional dependence between the response (wear rate), and the influencing factors (Al2O3amount, Al2O3size, sliding speed and normal load). Lubricants 2023,11, 24 4 of 13 Table 2. Influencing factors and their levels. Factor Unit Level 1 2 3 A: Al2O3amount wt. % 0.2 0.3 0.5 B: Al2O3size nm 25 100 C: sliding speed m/s 0.25 1.00 D: normal load N 40 100 The artificial neural network (ANN) simulation started with a “training” process in which a set of inputs are applied to the network. After that, the resulting set of outputs is compared to known values. The training is performed until the error between the output and known values reach a predefined value. ANN is usually used with a high amount of input data, but it can be used successfully for small data sets as well [ 12 ]. A feed-forward backpropagation multilayer ANN is employed, with three inputs, five neurons in the hidden layer and one output used (Figure 1). Networks with 2–25 neurons in the hidden layer were trained. Training and testing of the ANN are conducted using the software MATLAB R2016a. The logarithmic sigmoid transfer function (logsig) and linear transfer function (purelin) are used as activation transfer functions, while the Levenberg–Marquardt backpropagation algorithm (trainlm) is used as the training algorithm. Lubricants2023,11,xFORPEERREVIEW4of14   toobtainanoptimalresponse[10,11].Itappliesseveralmathematicalandstatisticaltech‐ niquestodevelopanadequatefunctionaldependencebetweentheresponse(wearrate), andtheinfluencingfactors(Al 2 O 3 amount,Al 2 O 3 size,slidingspeedandnormalload). Table2.Influencingfactorsandtheirlevels. FactorUnitLevel 123 A:Al 2 O 3 amountwt.%0.20.30.5 B:Al 2 O 3 sizenm25100 C:slidingspeedm/s0.251.00 D:normalloadN40100 Theartificialneuralnetwork(ANN)simulationstartedwitha“training”processin whichasetofinputsareappliedtothenetwork.Afterthat,theresultingsetofoutputsis comparedtoknownvalues.Thetrainingisperformeduntiltheerrorbetweentheoutput andknownvaluesreachapredefinedvalue.ANNisusuallyusedwithahighamountof inputdata,butitcanbeusedsuccessfullyforsmalldatasetsaswell[12].Afeed‐forward backpropagationmultilayerANNisemployed,withthreeinputs,fiveneuronsinthehid‐ denlayerandoneoutputused(Figure1).Networkswith2–25neuronsinthehiddenlayer weretrained.TrainingandtestingoftheANNareconductedusingthesoftwareMATLAB R2016a.Thelogarithmicsigmoidtransferfunction(logsig)andlineartransferfunction (purelin)areusedasactivationtransferfunctions,whiletheLevenberg–Marquardtback‐ propagationalgorithm(trainlm)isusedasthetrainingalgorithm.  Figure1.ArchitectureofthedevelopedANN. 3.ResultsandDiscussion 3.1.Microhardness Themicrohardnesstestingresults(Figure2)showedgoodrepeatabilityoftheresults sincethecalculatedstandarddeviationswerebelow7%.Allnanocompositesshowed more‐or‐lesssimilarmicrohardnessvalues,whichwasexpectedduetotheirsimilarmi‐ crostructures[6,7].Hardnesstypicallyincreaseswiththeincreaseinnanoparticleamount, buttheinterfacebetweenreinforcementsandmatrixaswellasthedistributionofrein‐ forcementshaveasignificantimpactoncompositeshardnessaswell[13].Theuniform dispersionofthenanoparticlesthroughoutthematrixisademandingtask,noteasyto obtain[14].Thelowinfluenceofceramicnanoparticlescanbeexplainedbythefactthat theiramountisverysmallandthattheaverageVickersmicrohardnessindentationdiag‐ onal(approximately85μm)issignificantlysmallerthanthedistancebetweenthenano‐ particles,sothemeasuredhardnessvaluesaremostprobablyvaluesofthenanocomposite matrix. Figure 1. Architecture of the developed ANN. 3. Results and Discussion 3.1. Microhardness The microhardness testing results (Figure 2) showed good repeatability of the results since the calculated standard deviations were below 7%. All nanocomposites showed more-or-less similar microhardness values, which was expected due to their similar microstructures [ 6 , 7 ]. Hardness typically increases with the increase in nanoparticle amount, but the interface between reinforcements and matrix as well as the distribution of reinforcements have a significant impact on composites hardness as well [ 13 ]. The uniform dispersion of the nanoparticles throughout the matrix is a demanding task, not easy to obtain [ 14 ]. The low influence of ceramic nanoparticles can be explained by the fact that their amount is very small and that the average Vickers microhardness indentation diagonal (approximately 85 µ m) is significantly smaller than the distance between the nanoparticles, so the measured hardness values are most probably values of the nanocomposite matrix. 3.2. Wear Rate The experiments were conducted according to the L24 orthogonal array, which is formed with Minitab 19 statistical package. The results of the wear tests for different combinations of factors are presented in Table 3. Wear rates were calculated for the whole testing period, i.e., they represented the total wear rates. Lubricants 2023,11, 24 5 of 13 Lubricants2023,11,xFORPEERREVIEW5of14    Figure2.Microhardnessvaluesoftestednanocompositeswithcorrespondingstandarddeviations(SD). 3.2.WearRate TheexperimentswereconductedaccordingtotheL24orthogonalarray,whichis formedwithMinitab19statisticalpackage.Theresultsoftheweartestsfordifferentcom‐ binationsoffactorsarepresentedinTable3.Wearrateswerecalculatedforthewhole testingperiod,i.e.,theyrepresentedthetotalwearrates. Table3.ExperimentaldesignusingL24orthogonalarrayandobtainedweartestresults. Testno. Al2O3 Amount,wt. % Al2O3Size, nm Sliding Speed,m/s NormalLoad, N WearRate× 10–4,mm3/m 10.2250.25401.671859 20.2250.251002.337679 30.2251400.701434 40.22511001.315875 50.21000.25400.940843 60.21000.251001.911662 70.21001400.150596 80.210011000.825748 90.3250.25400.851294 100.3250.251001.800392 110.3251400.630633 120.32511001.249261 130.31000.25400.573598 140.31000.251001.591235 150.31001400.055426 160.310011000.277412 170.5250.25400.414330 180.5250.251001.532718 190.5251400.141480 200.52511000.281368 210.51000.25400.200666 220.51000.251000.861667 230.51001400.013621 240.510011000.062987 136.1 139.0 132.9 0.2 0.3 0.5 0 20 40 60 80 100 120 140 160 Al 2 O 3 amount.wt.% Al 2 O 3 size:25nm SD=8.77 SD=6.11 MicrohardnessHV0.5 SD=4.99 139.1 135.9 130.7 0.2 0.3 0.5 0 20 40 60 80 100 120 140 160 Al 2 O 3 size:100nm Al 2O3amount.wt.% SD=3.00 SD=6.87 MicrohardnessHV0.5 SD=9.72 Figure 2. Microhardness values of tested nanocomposites with corresponding standard deviations (SD). Table 3. Experimental design using L24 orthogonal array and obtained wear test results. Test no. Al2O3Amount, wt. % Al2O3Size, nm Sliding Speed, m/s Normal Load, N Wear Rate ×10–4, mm3/m 1 0.2 25 0.25 40 1.671859 2 0.2 25 0.25 100 2.337679 3 0.2 25 1 40 0.701434 4 0.2 25 1 100 1.315875 5 0.2 100 0.25 40 0.940843 6 0.2 100 0.25 100 1.911662 7 0.2 100 1 40 0.150596 8 0.2 100 1 100 0.825748 9 0.3 25 0.25 40 0.851294 10 0.3 25 0.25 100 1.800392 11 0.3 25 1 40 0.630633 12 0.3 25 1 100 1.249261 13 0.3 100 0.25 40 0.573598 14 0.3 100 0.25 100 1.591235 15 0.3 100 1 40 0.055426 16 0.3 100 1 100 0.277412 17 0.5 25 0.25 40 0.414330 18 0.5 25 0.25 100 1.532718 19 0.5 25 1 40 0.141480 20 0.5 25 1 100 0.281368 21 0.5 100 0.25 40 0.200666 22 0.5 100 0.25 100 0.861667 23 0.5 100 1 40 0.013621 24 0.5 100 1 100 0.062987 Wear factors are also calculated in order to compare the obtained results with the literature data, using the well-known equation [ 15 ]. The obtained values were in the interval from 10 −8 to 10 −6 mm 3 /Nm which corresponds to the literature data for metallic materials in sliding contact under boundary lubrication conditions [ 16 ]. This is in accordance with the used lubrication system, which delivers lubricant to the contact by the rotation of the disc. The disc is only partially sunk into the oil container and the contact between the block and the disc is above the oil level. This means that the contact is not fully supplied with the lubricant so mixed and boundary lubrication may occur. Applying RSM to the experimental results, a second-order equation that determines the wear rate was created: y= 2.015 −6.3 ·A−0.0077 ·B−0.567 ·C + 0.0236 ·D + 4.4436 ·A2+ 0.01112 ·A·B + + 1.316 ·A·C−0.0139 ·A·D−0.0012 ·B·C−0.000019 ·B·D−0.01135 ·C·D(1) Lubricants 2023,11, 24 6 of 13 where: yis the wear rate and A, B, C and D are the Al 2 O 3 amount, Al 2 O 3 size, sliding speed and normal load, respectively. The validity of the developed model was analysed using the analysis of variance (ANOVA) technique, for which results are shown in Table 4. The ANOVA was carried out for a significance level of 5%, i.e., for a confidence level of 95%. Sources with a p-value less than the significance level of 0.05 (5%) were considered to have a statistically significant contribution on the nanocomposite wear rate. This means that all considered influencing factors (Al 2 O 3 amount, Al 2 O 3 size, sliding speed and normal load) had a significant impact on wear rate. Regarding their interactions, only the interaction of sliding speed and normal load can be treated to have a significant impact on wear rate. Table 4. Analysis of variance table for wear rate. Source Degree of Freedom Adjusted Sums of Square Adjusted Mean Square F-Value p-Value Model 11 10.2340 0.93037 25.57 0.000 Linear 4 9.1493 2.28733 62.87 0.000 Al2O3amount 1 2.5177 2.51766 69.20 0.000 Al2O3size 1 1.1482 1.14815 31.56 0.000 Sliding speed 1 3.1590 3.15903 86.83 0.000 Normal load 1 2.3245 2.32447 63.89 0.000 Square 1 0.0406 0.04060 1.12 0.312 Al2O3amount ×Al2O3amount 1 0.0406 0.04060 1.12 0.312 2-way interaction 6 0.6295 0.10491 2.88 0.056 Al2O3amount ×Al2O3size 1 0.0649 0.06494 1.78 0.206 Al2O3amount ×sliding speed 1 0.0909 0.09094 2.50 0.140 Al2O3amount ×normal load 1 0.0649 0.06488 1.78 0.207 Al2O3size ×sliding speed 1 0.0069 0.00686 0.19 0.672 Al2O3size ×normal load 1 0.0109 0.01085 0.30 0.595 Sliding speed ×normal load 1 0.3910 0.39099 10.75 0.007 Error 12 0.4366 0.03638 Total 23 10.6706 A normal probability plot of residuals (Figure 3) was used to verify the assumption that the residuals are normally distributed. As could be noticed, the prediction of wear rate with the formulated model is acceptable. The obtained R 2 (R-squared) value of 0.9591 and adjusted R 2 value of 0.9216 indicate a good fit of the model to the input data (R 2 = 1 is a perfect fit). Lubricants2023,11,xFORPEERREVIEW7of14   aresettotheconstantmeanvalue(holdvalue),whilethechangesintheothertwofactors areshownonthediagram.Basedonthepresentedwearmaps,thelowestwearwasob‐ tainedforthenanocompositewith0.5wt.%Al 2 O 3 of100nmsize.Theinfluencesofthemain influencingfactorsonwearratearealsopresentedas3DwearcurvesinFigure4g,h.Gen‐ erally,allnanocompositesshowedrelativelylowwearvalues.Theaveragewearratesofthe nanocompositesreinforcedwithsmaller‐sizeandbigger‐sizenanoparticles,atasliding speedof0.25andanormalloadof100N,aremorethantwotimeslower(approx.56%and 66%,respectively)thanthewearrateofthixocastedZA‐27alloy[3],suggestingthatthead‐ ditionofnanoparticlescansignificantlyimprovewearresistanceofthematrix.Weobtained somethingsimilarinourpreviousresearch[17],wheretheadditionof1wt.%Al 2 O 3 nano‐ particles(withparticlesize20–30nm)resultedinthereductioninthewearrateofmodified ZA‐27alloybyapproximately60%.Testingconditionsandcontactgeometryweresimilar, i.e.,thenormalloadwasthesame,buttheslidingspeedwashigher(0.5m/s)andlubricant viscosity(mineralengineoilSAE15W‐40,ACEAE3)wasmorethantwotimeslower.Ro‐ hatgiandSchultz[1]intheirreviewpapershowedthatforsomeMMnCs,additionofa smallpercentageofnanoparticlescanaffectspecificmaterialpropertiesandthatinsome cases,thedispersednanoparticlesleadtopropertychangesinthematrixitself.  Figure3.NormalprobabilityplotforthedevelopedRSMmodel. Asexpected,testingunderahighernormalloadandlowerslidingspeedcausedhigher wearrates.ThisisaccordingtotheStribeckcurveintheareaofmixedlubrication[18].In thatarea,lubricationconditionsbecomeharshwithloadincreaseand/orspeeddecrease.On theotherhand,theincreaseintheAl 2 O 3 particlesize,aswellastheAl 2 O 3 particleamount, increasedthewearresistanceofnanocomposites.Itwasalreadyshowninourprevious studythatthepresenceofnanoparticlesledtothestrengtheningofthenanocompositesat roomtemperatureandthatthelargestcontributionwasduetotheenhanceddislocation densitystrengtheningmechanism[6].Itwasalsoshownthatthestrengtheningeffectwas higherwiththehigheramount(wt.%)ofnanoparticles.Itismostprobablethatthecontact temperature,togetherwithcontactpressure,inducedotherstrengtheningmechanisms (loadtransfereffect,Hall–Petchstrengthening,Orowanstrengthening,coefficientofther‐ malexpansionandelasticmodulus)thatusuallyfollowtheadditionofnanoparticles[2]. Themicrostructureanditsinfluenceonthemechanicalcharacteristicofthetested nanocompositeswereanalysedindetailinourpreviousstudy[6].Thenanoparticles showedatendencyforagglomerationandformationofclustersduetothelowwettability betweenthemandthemetalmatrix.Itwasalsoshownthatthemicrostructureswere nondendriticanddidnotdiffertoomuchregardlessofthesizeandamountofreinforce‐ ment.However,thenanocompositesreinforcedwithbiggerAl 2 O 3 nanoparticleshadbig‐ gergrainsizesaswell.Thegrainsizewasexpressedthroughtheaverageferetdiameter whichwas65μmatnanocompositesreinforcedwith0.3wt.%ofbigger‐sizenanoparticles and55μmatnanocompositesreinforcedwith0.3wt.%ofsmaller‐sizenanoparticles.This Figure 3. Normal probability plot for the developed RSM model. The results of the wear rate prediction, obtained by using Equation (1) and presented in the form of wear maps, are shown in Figure 4. When creating the wear maps, two factors are set to the constant mean value (hold value), while the changes in the other two factors are shown on the diagram. Based on the presented wear maps, the lowest wear was Lubricants 2023,11, 24 7 of 13 obtained for the nanocomposite with 0.5 wt. % Al 2 O 3 of 100 nm size. The influences of the main influencing factors on wear rate are also presented as 3D wear curves in Figure 4g,h. Generally, all nanocomposites showed relatively low wear values. The average wear rates of the nanocomposites reinforced with smaller-size and bigger-size nanoparticles, at a sliding speed of 0.25 and a normal load of 100 N, are more than two times lower (approx. 56% and 66%, respectively) than the wear rate of thixocasted ZA-27 alloy [ 3 ], suggesting that the addition of nanoparticles can significantly improve wear resistance of the matrix. We obtained something similar in our previous research [ 17 ], where the addition of 1 wt. % Al 2 O 3 nanoparticles (with particle size 20–30 nm) resulted in the reduction in the wear rate of modified ZA-27 alloy by approximately 60 %. Testing conditions and contact geometry were similar, i.e., the normal load was the same, but the sliding speed was higher (0.5 m/s) and lubricant viscosity (mineral engine oil SAE 15W-40, ACEA E3) was more than two times lower. Rohatgi and Schultz [ 1 ] in their review paper showed that for some MMnCs, addition of a small percentage of nanoparticles can affect specific material properties and that in some cases, the dispersed nanoparticles lead to property changes in the matrix itself. As expected, testing under a higher normal load and lower sliding speed caused higher wear rates. This is according to the Stribeck curve in the area of mixed lubrication [ 18 ]. In that area, lubrication conditions become harsh with load increase and/or speed decrease. On the other hand, the increase in the Al 2 O 3 particle size, as well as the Al 2 O 3 particle amount, increased the wear resistance of nanocomposites. It was already shown in our previous study that the presence of nanoparticles led to the strengthening of the nanocomposites at room temperature and that the largest contribution was due to the enhanced dislocation density strengthening mechanism [ 6 ]. It was also shown that the strengthening effect was higher with the higher amount (wt. %) of nanoparticles. It is most probable that the contact temperature, together with contact pressure, induced other strengthening mechanisms (load transfer effect, Hall–Petch strengthening, Orowan strengthening, coefficient of thermal expansion and elastic modulus) that usually follow the addition of nanoparticles [2]. The microstructure and its influence on the mechanical characteristic of the tested nanocomposites were analysed in detail in our previous study [ 6 ]. The nanoparticles showed a tendency for agglomeration and formation of clusters due to the low wettability between them and the metal matrix. It was also shown that the microstructures were nondendritic and did not differ too much regardless of the size and amount of reinforcement. However, the nanocomposites reinforced with bigger Al 2 O 3 nanoparticles had bigger grain sizes as well. The grain size was expressed through the average feret diameter which was 65 µ m at nanocomposites reinforced with 0.3 wt. % of bigger-size nanoparticles and 55 µm at nanocomposites reinforced with 0.3 wt. % of smaller-size nanoparticles. This could be the reason why the increase in the Al 2 O 3 particle size increased the wear resistance of nanocomposites. Bigger grains were harder to wear, which makes sense if we know that the widths of abrasive groves were from 10 to 20 µm (see wear mechanism discussion). The experimental output values of the wear rates, shown in Table 2, are used for training, validation and testing of the ANN. It was trained using 70% of the data, while 15% was used for testing and 15% for validation. The performance of the modelled ANN is shown in Figure 5. Mean square error (MSE) for training initially has a high value which decreased to a very small value as the number of epochs increases, i.e., the ANN’s training process is being performed correctly. Even though training continues until epoch 6, the best validation performance is achieved at epoch 0 with a value of 0.00031244, and the final value of the gradient coefficient at epoch 6 is very close to zero, i.e., 0.0014. The regression coefficient for training, validation and testing, as well as the overall regression coefficient of the network, was obtained and shown in Figure 6. The overall regression coefficient of the network was very close to 1 (R= 0.98729), indicating a good fit and good agreement between the experimental results and the ANN model prediction. Lubricants 2023,11, 24 8 of 13 Lubricants2023,11,xFORPEERREVIEW8of14   couldbethereasonwhytheincreaseintheAl 2 O 3 particlesizeincreasedthewearre‐ sistanceofnanocomposites.Biggergrainswerehardertowear,whichmakessenseifwe knowthatthewidthsofabrasivegroveswerefrom10to20μm(seewearmechanism discussion).     (a)(b) (c)(d) (f) (e) (g)(h) Figure 4. Wear maps for the tested nanocomposites, i.e., dependence of wear rate on: ( a , g ) Al 2 O 3 amount and Al 2 O 3 size, ( b ) Al 2 O 3 amount and sliding speed and ( c , h ) Al 2 O 3 amount and normal load, ( d ) Al 2 O 3 size and sliding speed, ( e ) Al 2 O 3 size and normal load and ( f ) sliding speed and normal load. Lubricants 2023,11, 24 9 of 13 Lubricants2023,11,xFORPEERREVIEW9of14   Figure4.Wearmapsforthetestednanocomposites,i.e.,dependenceofwearrateon:(a,g)Al2O3 amountandAl2O3size,(b)Al2O3amountandslidingspeedand(c,h)Al2O3amountandnormal load,(d)Al2O3sizeandslidingspeed,(e)Al2O3sizeandnormalloadand(f)slidingspeedand normalload. Theexperimentaloutputvaluesofthewearrates,showninTable2,areusedfor training,validationandtestingoftheANN.Itwastrainedusing70%ofthedata,while 15%wasusedfortestingand15%forvalidation.TheperformanceofthemodelledANN isshowninFigure5.Meansquareerror(MSE)fortraininginitiallyhasahighvaluewhich decreasedtoaverysmallvalueasthenumberofepochsincreases,i.e.,theANN’straining processisbeingperformedcorrectly.Eventhoughtrainingcontinuesuntilepoch6,the bestvalidationperformanceisachievedatepoch0withavalueof0.00031244,andthe finalvalueofthegradientcoefficientatepoch6isveryclosetozero,i.e.,0.0014.There‐ gressioncoefficientfortraining,validationandtesting,aswellastheoverallregression coefficientofthenetwork,wasobtainedandshowninFigure6.Theoverallregression coefficientofthenetworkwasverycloseto1(R=0.98729),indicatingagoodfitandgood agreementbetweentheexperimentalresultsandtheANNmodelprediction.  Figure5.PerformanceofthemodelledANN—meansquareerrorandtrainingstateplots. Figure 5. Performance of the modelled ANN—mean square error and training state plots. Lubricants2023,11,xFORPEERREVIEW10of14    Figure6.AccuracyofthemodelledANN—regressionanalysisofdifferentphases. ThevaluesofthewearratepredictedwiththemodelledRSMandANN,withinthe limitsoftheexperiment,arepresentedinTable5andcomparedwiththeexperimental data.ItsgraphicalinterpretationisshowninFigure7.Inbothcases(RSMandANNpre‐ diction),thereisagoodcorrelationbetweenexperimentalandpredictedvaluesandboth predictionmethodscanbeusedwithhighreliability.However,valuesobtainedbythe modelledANNareclosertoexperimentalvalues;therefore,itcanbeconcludedthatin thiscase,theANNismoreefficientinpredictingwearrate. Table5.PredictedvaluesofthewearrateobtainedbyusingtheRSMandANN. Testno.WearRate×10–4,mm3/m RSMPredictedRSMErrorANNPredictedANNError 11.412550.2593091.666860.004998 22.46300−0.1253212.331260.006419 30.82170−0.1202660.70289−0.001455 41.36140−0.0455251.31694−0.001060 50.922350.0184930.932340.008505 61.887300.0243621.876690.034968 70.26400−0.1134040.15359−0.002993 80.718200.1075480.821100.004648 91.00965−0.1583560.99211−0.140817 101.97670−0.1763081.81776−0.017371 110.517500.1131330.628180.002449 120.973800.2754611.049520.199744 130.60285−0.0292520.566600.006995 141.484400.1068351.65411−0.062870 150.043200.0122260.051070.004356 160.41400−0.1365880.28120−0.003788 Figure 6. Accuracy of the modelled ANN—regression analysis of different phases. The values of the wear rate predicted with the modelled RSM and ANN, within the limits of the experiment, are presented in Table 5and compared with the experimental data. Its graphical interpretation is shown in Figure 7. In both cases (RSM and ANN prediction), there is a good correlation between experimental and predicted values and both prediction methods can be used with high reliability. However, values obtained by the modelled ANN are closer to experimental values; therefore, it can be concluded that in this case, the ANN is more efficient in predicting wear rate.