ADVANCES IN MATERIALS, STRUCTURES, SENSING, AND DATA-DRIVEN ENGINEERING APPLICATIONS

EDITOR Prof. Dr. Raul D.S.G. CAMPILHO ADVANCES IN MATERIALS, STRUCTURES, SENSING, AND DATA-DRIVEN ENGINEERING APPLICATIONS Published by BZT TURAN PUBLISHING HOUSE Certificate Number: 202401 Delaware, United States www.bztturanpublishinghouse.com [email protected] EDITOR: Prof. Dr. Raul D.S.G. CAMPILHO ADVANCES IN MATERIALS, STRUCTURES, SENSING, AND DATA-DRIVEN ENGINEERING APPLICATIONS OPEN ACCESS Suggested Citation: Campilho, R. (2025). ADVANCES IN MATERIALS, STRUCTURES, SENSING, AND DATA-DRIVEN ENGINEERING APPLICATIONS. BZT TURAN PUBLISHING HOUSE. DOI: https://doi.org/10.30546/19023.978-9952-8610-75.2025.0057 Language: English Publication Date: October 2025 Cover Design By Mehmet ÇAKIR Print and digital versions typeset by BZT TURAN Media Co. Ltd. E-ISBN: 9 7 8 - 9 9 5 2 - 8 6 1 0 - 7 - 5 DOI: https://doi.org/10.30546/19023.978-9952-8610-7-5.2025.0057 iii PREFACE Engineering research today ranges from the design of high-performance materials and structures to the development of intelligent systems and humancentred solutions. The challenges faced by industry, defence, sport, and society require specialized approaches and an integration of perspectives from diverse domains. This book reflects such diversity, by gathering contributions that combine advanced modelling, numerical analysis, sensing technologies, and data-driven methods to address problems of both technical and social importance. The first part of the volume focuses on advanced materials and structural applications. It begins with a numerical study of ballistic protection using ultra-high-molecular-weight polyethylene (UHMWPE) and Kevlar plates, providing insights into hybrid armour systems. This is followed by a set of chapters dedicated to adhesive bonding in structural engineering, where cohesive zone modelling and extended finite element techniques are employed to evaluate impact performance, scarf geometries, and novel solutions in the canoeing boat industry. These works reinforce the importance of adhesives in modern structural design and propose new guidelines and improvements for their application in demanding environments. The second part of the book transitions into sensing technologies with a study on drone-based ground penetrating radar for subterranean object detection. This contribution highlights the growing role of unmanned aerial systems in defence, archaeology, and environmental monitoring, showing how classical radar methods can be adapted to modern mobility platforms. The third and final part turns to data-driven and human-centred engineering. Here, one chapter introduces a sentiment filtering and recommendation system tailored for Turkish e-commerce, illustrating the impact of natural language processing and trust-weighted heuristics on consumer-oriented platforms. The final chapter addresses spatial quality in student dormitories, using a systematic literature review to compile and analyse problems affecting user well-being. Taken together, these seven chapters emphasize the innovation of contemporary engineering research, demonstrating how fundamental advances in materials, mechanics, sensing, and data science can converge iv toward practical applications. It is my hope that this volume will serve as a valuable resource for researchers, practitioners, and students, and inspire further exploration of these topics. v Raul Duarte Salgueiral Gomes Campilho has a PhD in Mechanical Engineering (2009) and a Habilitation in Mechanical Engineering, both from the Faculty of Engineering of the University of Porto (Portugal). He is an associate professor of the Mechanical Engineering department of ISEP – School of Engineering, Polytechnic University of Porto (Portugal), where he teaches several courses of the Bachelor and Master degree in Mechanical Engineering. Since 2023 he is the vice-director of CIDEM – Centre for Research & Development in Mechanical Engineering. He develops his research activities in ISEP and INEGI – Institute of Science and Innovation in Mechanical and Industrial Engineering, which is a Research and Technology Organization (RTO), and is affiliated to the LAETA – Associate Laboratory of Energy, Transports and Aerospace. His research mainly focuses on adhesive joints; structural adhesives; design of bonded joints; experimental testing; Finite Element Method; Extended Finite Element Method; Meshless Methods; Cohesive Zone Models; composite materials; numerical modelling of composite materials; micromechanics; macromechanics; design of mechanical structures; flexible production; automation; robotics; and actuator systems. He was the principal investigator of 1 national R&D project and participated/participates as a researcher in 6 national R&D projects, all funded through competitive calls. He also participated in 1 R&D project in collaboration with the Portuguese industry (funded through competitive calls). He is/was the (co)supervisor of 6 PhD (1 concluded and 5 ongoing) and 258 MSc thesis. His publication record contains 340 articles & reviews published in Web of Science-indexed journals and 154 book chapters. He also coedited 3 books and was the (co)author of over 400 communications that were presented in international conferences and 18 communications presented in national conferences. vii CONTENTS PREFACE .................................................................................................................... iii CHAPTER 1 Investigation of Ballistic Behavior of UHMWPE and Kevlar Plates ....... 1 Ahmet Murat Asan CHAPTER 2 Procedures for the impact analysis of adhesively-bonded structures ........................................................................................................... 9 P.D.A. Da Silva R.D.S.G. Campilho CHAPTER 3 Numerical evaluation of scarf geometry adhesively-bonded joints by XFEM modelling .............................................................................................21 I.R.S. Araújo R.D.S.G. Campilho CHAPTER 4 Numerical CZM evaluation of adhesively-bonding solutions for canoeing boat fabrication .................................................................................. 39 João C.M. Santos Raul D.S.G. Campilho CHAPTER 5 Feasibility of Drone-Based Ground Penetrating Radar for Subterranean Foreign Object Detection/Mapping ....................................59 Celile Nur Yalçın INVESTIGATION OF BALLISTIC BEHAVIOR OF UHMWPE AND KEVLAR PLATES 6 References Abtew, M. A., Boussu, F., Bruniaux, P., Loghin, C., & Cristian, I. (2019). Ballistic impact mechanisms–A review on textiles and fibre-reinforced composites impact responses. Composite Structures, 223, 110966. https://doi.org/10.1016/j.compstruct.2019.110966 ANSYS. (2025). GRANTA Materials Data for Simulation (Sample). https://www.ansys.com/ products/materials Barros, D., Mota, C., Bessa, J., Cunha, F., Rosa, P., & Fangueiro, R. (2023). Blast fragment impact of angle-ply composite structures for buildings wall protection. Buildings, 13(8). https://doi.org/10.3390/buildings13081959 Bajya, M., Majumdar, A., Butola, B. S., Arora, S., & Bhattacharjee, D. (2021). Ballistic performance and failure modes of woven and unidirectional fabric based soft armour panels. Composite Structures, 255, 112941. https://doi.org/10.1016/j.compstruct.2020.112941 Carr, D. (1999). Failure mechanisms of yarns subjected to ballistic impact. Journal of Materials Science Letters, 18(7), 585–588. Edidin, A. A., & Kurtz, S. M. (2000). Influence of mechanical behavior on the wear of 4 clinically relevant polymeric biomaterials in a hip simulator. The Journal of Arthroplasty, 15(3), 321–331. Hearle, J. W. S. (2001). High-performance fibres. Elsevier. Hu, P., Yang, H., Zhang, P., Wang, W., Liu, J., & Cheng, Y. (2022). Experimental and numerical investigations into the ballistic performance of ultra-high molecular weight polyethylene fiber-reinforced laminates. Composite Structures, 290, 115499. https://doi. org/10.1016/j.compstruct.2022.115499 Karthikeyan, K., Russell, B. P., Fleck, N. A., Wadley, H. N. G., & Deshpande, V. S. (2013a). The effect of shear strength on the ballistic response of laminated composite plates. European Journal of Mechanics - A/Solids, 42, 35–53. https://doi.org/10.1016/j. euromechsol.2013.04.002 Karthikeyan, K., Russell, B. P., Fleck, N. A., O’Masta, M., Wadley, H. N. G., & Deshpande, V. S. (2013b). The soft impact response of composite laminate beams. International Journal of Impact Engineering, 60, 24–36. Li, X., Zhang, X., Guo, Y., Shim, V., Yang, J., & Chai, G. B. (2018). Influence of fiber type on the impact response of titanium-based fiber-metal laminates. International Journal of Impact Engineering, 114, 32–42. Ma, Y., Wang, J., Zhao, G., & Liu, Y. (2023). New insights into the damage assessment and energy dissipation weight mechanisms of ceramic/fiber laminated composites under ballistic impact. Ceramics International, 49(13), 21966–21977. https://doi.org/10.1016/j.ceramint.2023.04.021 Peng, L., Zhou, J., Wang, Q., Zhang, X., & Guan, Z. (2024). Numerical modelling of the ballistic impact response of hybrid composite structures. Composites Part C: Open Access, 14, 100474. https://doi.org/10.1016/j.jcomc.2024.100474 AHMET MURAT ASAN 7 Ramadhan, A., Talib, A. A., Rafie, A. S. M., & Zahari, R. (2012). Experimental and numerical simulation of energy absorption on composite Kevlar29/Polyester under high velocity impact. Journal of Advanced Science and Engineering Research, 2(1). http://psasir. upm.edu.my/id/eprint/23243 Reddy, T. S., Reddy, P. R. S., & Madhu, V. (2017). Response of E-glass/epoxy and Dyneema® composite laminates subjected to low and high velocity impact. Procedia Engineering, 173, 278–285. https://doi.org/10.1016/j.proeng.2016.12.014 Tan, V. B. C., & Khoo, K. J. L. (2005). Perforation of flexible laminates by projectiles of different geometry. International Journal of Impact Engineering, 31(7), 793–810. Virtue. (2024). UHMWPE. https://www.virtuetextile.com/ Wang, H., Weerasinghe, D., Mohotti, D., Hazell, P. J., Shim, V., Shankar, K., & Morozov, E. V. (2021). On the impact response of UHMWPE woven fabrics: Experiments and simulations. International Journal of Mechanical Sciences, 204, 106574. https://doi. org/10.1016/j.ijmecsci.2021.106574 Wikipedia. (2021). 9x19mm Parabellum. Retrieved December 2, 2021, from https://tr.wikipedia.org/wiki/9x19mm_Parabellum Zhu, W., Huang, G. Y., Feng, S. S., & Stronge, W. J. (2018). Conical nosed projectile perforation of polyethylene reinforced cross-ply laminates: Effect of fiber lateral displacement. International Journal of Impact Engineering, 118, 39–49. Contributors Ahmet Murat AŞAN graduated from Fırat University in mechanical engineering and electrical and electronics engineering. He has a master’s and doctorate degree in mechanical engineering. He is currently working as a researcher at Dicle University. His research interests include composite materials, fracture mechanics, the finite element method, vibration, ballistic, tensile, compression, shear, and fatigue tests. 9 CHAPTER 2 Procedures for the impact analysis of adhesively-bonded structures P.D.A. da Silva1 R.D.S.G. Campilho2 Abstract The application of structural adhesives has been increasing in the industry. Among the fields of application that most use this type of joints, the aeronautical and automotive industries stand out. Cohesive zone modelling (CZM) is widespread for static analysis of adhesive joints. However, in many applications, impact analyses are fundamental to assess the structural safety of adhesively bonded structures. This work addresses the numerical analysis of adhesive tubular joints subjected to impact loads, considering different adhesives. The numerical approach to model impact damage consisted of an adaption of the CZM technique using the Abaqus® software. A parametric numerical study carried out on the influence of the overlap length (LO) and adherends thickness (tp), on the strength of the adhesive joints. Loaddisplacement (P-δ) curves, absorbed energy (Ea) and maximum load (Pm) are presented for all joints tested. It was concluded that the strength of the 1 CIDEM, ISEP – School of Engineering, Polytechnic of Porto, R. Dr. António Bernardino de Almeida, 431, 4200-072 Porto, Portugal. 2 CIDEM, ISEP – School of Engineering, Polytechnic of Porto, R. Dr. António Bernardino de Almeida, 431, 4200-072 Porto, Portugal. Institute of Science and Innovation in Mechanical and Industrial Engineering, Rua Dr. Roberto Frias, 400, 4200-465 Porto, Portugal. PROCEDURES FOR THE IMPACT ANALYSIS OF ADHESIVELY-BONDED STRUCTURES 10 adhesive joints increases for stiff adhesives. Pm and E significantly improved for higher LO. Overall, clear design principles are proposed to maximize the tensile behavior of tubular adhesive joints under impact loads. Keywords: Adhesive joint, Structural adhesive, Impact analysis, Cohesive zone models. 1. Introduction There are currently many methods to estimate the strength of an adhesive joint. Conventional analytical methods provide the behaviour of adhesive joints at their linear elastic limit, either ruling out plastic behaviour or making the calculation quite complex. Thus, when the complexity of studies to determine the state of stress in an adhesive joint under certain conditions increases, the application of analytical methods of study becomes unfeasible, and numerical methods are adopted, in particular the finite element method (FEM). The FEM, introduced by Harris and Adams (1984) to adhesive joints, becomes the most widely used method. The related techniques incorporate factors such as joint rotation, the plasticity of the adherends, the plasticity of the adhesive, and the influence of the fillet. Opposed to static loads, dynamic loadings vary over the time. Within this scope, fatigue, modal analysis, and variable strain rate and impact loadings can be present in a structure. The variable strain rate/impact scenario in particular involves dynamic behaviour and requires explicit numerical integration. Currently, variable strain rate and impact studies are divided into continuum mechanics, damage mechanics and CZM models (Ramalho, Sánchez-Arce et al. 2022). Continuum mechanics is used to evaluate stresses and strains in adhesive joints. This approach is easy to programme. However, it should be considered that, under impact loading conditions, the stress propagates as a wave, which leads to several local stress peaks along the adhesive. Damage mechanics makes it possible to simulate the progressive degradation of the material until final failure along an arbitrary path. Different works are available in the literature under dynamic analysis of adhesive joints. Rao and Crocker (1990) used a theoretical model to study the bending vibration of a system of overlapping joints. First, the equations of motion in the joint region are derived mathematically using a differential calculus approach. The transverse displacements of the upper and lower beam are considered to be different. The adhesive were considered to be linearly viscoelastic and the Kelvin-Voigt model was used to represent this behaviour. The model can be used to predict the natural frequencies, modal damping ratios and mode shapes of the system for free vibration. Esmaeili, Zehsaz et al. (2015) carried out a fatigue study on the effect of the P.D.A. DA SILVA • R.D.S.G. CAMPILHO 11 tightening torque on bolted and hybrid (bonded/bolted) joints with different cyclic longitudinal loads. The fatigue of the specimens was predicted using six different multiaxial fatigue criteria by means of the local stress and strain distribution obtained from FEM analyses. The hybrid joints showed better fatigue life compared to the bolted joints. Increasing the tightening torque or clamping force on the joint leads to an increase in the fatigue resistance of double overlap bolted joints. This work addresses the numerical analysis of adhesive tubular joints subjected to impact loads, considering different adhesives. The numerical approach to model impact damage consisted of an adaption of the CZM technique using the Abaqus® software. A parametric numerical study carried out on the influence of LO and tp on the strength of the adhesive joints. P-δ curves, E and Pm are presented for all joints tested. 2. Experimental and numerical details 2.1. Materials To carry out the numerical analyses, it is important to define the mechanical properties of the materials employed. The material adopted for the adherends is DIN 55 Si7 (Silva, Peres et al. 2022). This material was chosen for two reasons: to avoid plastic deformation of the adherends during the numerical study and to ensure that the failure is cohesive in adhesive. Table 1 illustrates the mechanical properties of the material chosen for the adherends. Table 1 – Mechanical properties of DIN 55 Si7 (Silva, Peres et al. 2022). Properties Value Young’s modulus, E [GPa] 210 Tensile yield stress, σy [MPa] 1078 Tensile strength, σf [MPa] 1600 Tensile failure strain, εy [%] 6 Poisson’s ratio, v0.3 Density, ρ [g/cm3]7.8 A brittle adhesive (AV138) and two adhesives with high ductility (DP 8005 and XNR6852 E-2) were used. The mechanical properties were obtained by carrying out different types of tests. Tensile tests provided the Young Modulus (E) and tensile cohesive stress (tn 0), while the shear modulus (G) and the shear cohesive stress (ts 0) were obtained through thick-adherend PROCEDURES FOR THE IMPACT ANALYSIS OF ADHESIVELY-BONDED STRUCTURES 12 shear tests (TAST). Regarding the fracture properties, the double-cantilever beam (DCB) tests provided the tensile fracture toughness (GIC), while the end-notched flexure (ENF) tests provided the shear fracture toughness (GIIC). The density (ρ) and Poisson’s ratio (ν) are also required for processing the results. The final properties adjusted for this study can be found in Table 2 (Silva, Peres et al. 2022). Table 2 – Adhesives’ mechanical properties (Silva, Peres et al. 2022). Adhesive DP 8005 AV138 XNR6852 E-2 Young’s modulus, E [MPa] 590 4890 1742 Shear modulus, G [MPa] 159 1560 645.2 Tensile cohesive strength, tn 0 [MPa] 27.5 70.2 53.7 Shear cohesive strength, ts 0 [MPa] 36.7 51.7 45.8 Tensile toughness, GIC [N/mm] 1.1 0.35 1.68 Shear toughness, GIIC [N/mm] 60.6 18 Density, ρ [g/cm3]1.06 1.7 1.5 Poisson’s coefficient, v0.3a0.35b0.4c *a - estimated value; b - value supplied by the manufacturer; c - typical value for epoxy adhesives. 2.2. Geometries The Single Lap Joint (SLJ) that validates the numerical model with experiments under impact loads is initially defined. SLJs were manufactured to carry out the experimental tests, in accordance with ASTM D1002 and ISO 4587. Valente, Campilho et al. (2019) used these geometries to validate a CZM-based numerical model that could accurately predict the strength of adhesive joints submitted to impact loads, by comparing it with an experimental model and analysing the results. The SLJ is composed of two adherends with a length of 120 mm, LO of 25 mm and an adhesive thickness (ta) of 0.2 mm. The tp employed is 2 mm. Figure 1 presents the geometric parameters that define the adhesive joints used in the experimental tests. P.D.A. DA SILVA • R.D.S.G. CAMPILHO 13 Figure 1 – Geometry of joint used in the experimental test. The tubular adhesive joints used in the numerical parametric analysis are composed by two overlapping adherends with dissimilar diameters, bonded by an adhesive. In the case of tubular adherends, the value of LO is 10 mm and the length of the adherends is 60 mm, for a total specimen length of 100 mm. The outer diameter of the inner tube is fixed at 20 mm, tp is 2 mm, and ta is 0.2 mm. Figure 2 illustrates the tubular adhesive joint used in the numerical analysis. Figure 2 – Geometry of the tubular adhesive joint used in the numerical analysis. 2.3. Numerical modelling A bidimensional numerical analysis of the impact tests was performed using Abaqus®. The representation of the tubular nature of the joint PROCEDURES FOR THE IMPACT ANALYSIS OF ADHESIVELY-BONDED STRUCTURES 14 was based on the axisymmetric parameterisation, simulating the axis of revolution of the joint around its neutral axis, which was defined as being of the deformable type and with the basic characteristic of the wire type. Axisymmetric elements were selected to model both adherends (CAX4) and adhesive (COHAX4R). For the adherends, the elements were displayed in a structured mesh, while the cohesive elements were assigned a sweep type mesh. Besides, to simulate the test’s impact Ea, a solid homogeneous mass was created. Preliminary calculations showed that Ea of 40 J was required to separate the adherends and, by combining this value with the impact velocity previously estimated, it was possible to define the total mass and its respective density. Considering both materials properties and nature, elasticplastic behaviour was assumed for the adherends, and, for the adhesive, a quadratic stress criterion was employed to predict failure initiation, while damage propagation was ruled by a linear energetic criterion. Lastly, a linear stress-strain behaviour was assigned to the mass, and its elastic properties were defined to have no interference with the joint’s strength. Regarding boundary conditions (Figure 3), they were applied as follows: 1) nil longitudinal displacement at the left specimen’s edge; 2) nil radial displacement and velocity type predefined field of 1.75 m/s applied to the mass at the other edge, to assure the required Ea (40 J). Figure 3 – Set of boundary conditions for the numerical simulation. 3. Results 3.1. CZM validation To validate the performed numerical analysis, experimental results and numerical reference values obtained by (Valente, Campilho et al. 2019) P.D.A. DA SILVA • R.D.S.G. CAMPILHO 15 are taken as reference. For the adhesives AV138, DP8005, and XNR6852 E-2, a comparison between experimental and reference Pm values with the numerical Pm obtained in this work using the CZM model is presented in Figure 4 and detailed in Table 3 by the Pm relative difference (∆Pm). 11.75 14.43 13.61 16.79 19.93 19.90 24.00 26.92 28.39 0 10 20 30 40 Experimental Numerical reference Current numerical P m [kN] AV138 DP8005 XNR6852 E-2 Figure 4 – Comparison between experimental and numerical Pm values. Table 3 – Comparison between experimental and numerical Pm values. Adhesive AV138 DP8005 XNR6852 E-2 Numerical Reference (1) 14.43 kN 19.93 kN 26.92 kN Experimental Value (2) 11.75 kN 16.79 kN 24.00 kN Numerical Value (3) 13.61 kN 19.90 kN 28.39 kN ΔPm (3) – (1) -5.68% -0.15% 5.46% ΔPm (3) – (2) 12.89% 15.60% 16.31% 3.2. Parametric analysis 3.2.1. Overlap length Figure 5 (a) shows the P-δ curves for the joints bonded with the AV138 and LO=20 mm. The observed oscillations in P with δ are typical of impact loadings due to the inertial effects, leading to a P increase in steps up to Pm being reached. Between adhesives, the AV138 has the highest Pm, followed by the XNR6852 E-2 and finally the DP8005, relation that was found valid for all LO. The maximum (or failure) δ was inversely proportional to the adhesives’ stiffness. The Pm evolution with LO is shown in Figure 5 (b) for the three adhesives. Considering the AV138 results, Pm increases with LO. NUMERICAL EVALUATION OF SCARF GEOMETRY ADHESIVELY-BONDED JOINTS BY XFEM MODELLING 22 study of scarf adhesive joints in tension with different adhesives (Araldite® AV138, Araldite® 2015 and Sikaforce® 7752) and different scarf angles or α (3.43°, 10 °, 15°, 20°, 30° and 45°) by the eXtended Finite Element Method (XFEM). Initially, the experimental results obtained in a previous work are described, for the purpose of validating the obtained numerical results. The developed numerical work includes the distribution of the damage variable and joint strength. With the work carried out, the coherence of the numerical results with the experimental ones was observed, with emphasis on the joint strength as a function of α. It was found that joints with α=3.43° present the best results in terms of tensile strength of the joints. The adhesive Araldite® AV138 presents the best tensile behaviour, regardless of the α value. Based on the results obtained, it was considered that the XFEM is a tool that can be accurately used to design scarf adhesive joints. Keywords: Adhesive joint, eXtended Finite Element Method, Crack propagation method. 1. Introduction Currently, adhesively bonded joints have wide applicability and are used in various industrial sectors. The aeronautical industry was the one behind the widespread usage this joining method, which at the beginning of the last century applied adhesives based on casein (a natural polymeric material) in aircraft structures. During the 1950s, adhesives used in aircraft structures were capable to offer good stiffness and strength [1]. The use of adhesive joints in various applications proves to be more advantageous compared to more traditional mechanical joining methods, such as mechanical fastening, welding, riveting, among others. Adhesively bonded joints, when designed and manufactured correctly, provide significant advantages [2], such as more uniform stress distribution throughout the adhesive layer, with reduced stress concentrations. This distribution allows for higher stiffness and load transmissions, promoting weight reduction and lower cost. This bonding methodology also provides better fatigue resistance and vibration damping, as stresses are partially absorbed. As any other manufacturing process, several drawbacks can be identified, e.g., the need of design joints in a way that minimizes peel and cleavage loads as much as possible, a limited resistance to extreme conditions of temperature and humidity due to the polymeric nature of the adhesives, the need for clamping jigs to hold parts in position during the curing process and also the necessity of careful surface preparation procedures. Several joint architectures are available, allowing the designer to select the most suitable one, considering the application and I.R.S. ARAÚJO • R.D.S.G. CAMPILHO 23 subjected loads. For example, the single-lap joint (SLJ) is a widely used and studied design due to the ease of manufacture and predominant shear loading. One of the drawbacks of this configuration is the non-collinear forces that are transmitted, which cause the adhesive to be subject to peel stresses at the overlapping ends. To reduce this effect, other configurations are used, such as double lap joints, step joints and scarf joints, among others. Double lap joints, when compared to SLJ, are more complex and time-consuming to manufacture, however, the effects of bending are substantially lower [2]. Scarf and step joints present high strength, since these geometries favour the reduction of stress gradients along the adhesive bondline. On the other hand, due to the need to machine the overlapping area, these joint demands higher costs. In scarf joints, the strength depends on the scarf angle [3]. The evolutionary process of adhesive joints is closely related to the development of reliable prediction methodologies that allow increasing efficiency in their use, thus making it possible to overcome the paradigm of over dimensioned adhesive joints that resulted in more expensive and heavier structures, related to the lack of precise material models and adequate failure criteria that were evident a few decades ago. The two methodologies that can be applied to the analysis of adhesive joints are analytical and numerical methods. The Finite Element Method (FEM) is the most commonly used technique for the analysis of adhesive joints, having been initially applied by Harris and Adams [4], who introduced factors such as joint rotation, adherends and adhesives’ plasticity and the influence of fillets. Continuum mechanics was then used to predict the strength of adhesive joints, which requires stress distribution and an appropriate failure criterion. FEM can also be combined with fracture mechanics techniques to predict strength, either by the stress intensity factor or by energetic approaches such as the virtual crack closure technique. However, these modelling techniques make the process of evaluating crack growth difficult due to the need to recreate the mesh in the path of crack propagation, which has repercussions in terms of computational effort [5]. Over the last few decades, numerical modelling has seen major advances, one of which is the implementation of damage models using cohesive zone models (CZM). This technique couples conventional FEM models for regions where damage is not expected with fracture mechanics, through the use of cohesive elements to promote crack propagation. The concept of CZM began with studies by Barenblatt [6] and Dugdale [7], who described the damage in the fracture process zone in front of the crack under the effect of static loads. CZM allows to capture the beginning of a crack and its propagation within or at the interface of materials, or even in the delamination of composites. The implementation of CZM can NUMERICAL EVALUATION OF SCARF GEOMETRY ADHESIVELY-BONDED JOINTS BY XFEM MODELLING 24 be done in spring elements or, more conventionally, in cohesive elements [8]. The extended Finite Element Method (XFEM) uses enriched shape functions to represent a continuous displacement field. XFEM is a recent evolution of CZM, which allows the analysis and modelling of damage growth to predict fracture in structures, based on the strength of materials for damage initiation and deformations for failure assessment, instead of values of tn 0/ts 0 or δn 0/δs 0 (peak tractions and displacements in tension and shear, respectively) used in the CZM. Comparing to CZM, in XFEM it is no longer necessary that the crack follows a pre-defined path, which is a significant advantage. Thus, the crack can propagate freely within the structure without the need for the mesh to coincide with the geometry of the discontinuities and without the need to apply the mesh close to the crack [9]. Belytschko and Black [10], in the late 90s, presented the fundamental characteristics of this method, based on the concept of partition of unity, and which can be implemented in the FEM by introducing local enrichment functions for the displacements near the edge of the crack, to allow growth and separation between the crack faces. Recent and relevant studies carried out to predict the strength of scarf joints are available in the literature. Alves et al. [11] presented an experimental and numerical study of hybrid scarf joints. Carbon fibre reinforced polymer (CFRP) and aluminium adherends were bonded with Araldite® AV138 and Araldite® 2015, brittle and ductile adhesive, respectively, considering different scarf angles (α). Using the FEM, the peel (σy) and shear stresses (τxy) were obtained, while CZM was used to predict joint strength. The numerical results showed that the magnitude of σy and τxy increases with α, although this increase is more significant for σy. The damage variable showed that failure of the adhesive layer starts at the adhesive ends and grows towards the inner adhesive until failure. The experimental maximum load (Pm) increases exponentially with the reduction of α for the two tested adhesives, due to the increase in adhesive area and a more uniform stress distribution. The Pm values obtained by CZM are very close to those obtained experimentally. In the work of Sun et al. [12], an experimental and numerical study was carried out on the tensile performance of scarf adhesive joints. A ductile adhesive was used and CFRP adherends were considered. Experimentally, scarf joints were tested with different values of α (3°, 5°, 10°, 15°, 20° and 30°). Numerically, to validate the prediction accuracy of a user-defined CZM, the numerical results were compared with the experimental data. A triangular damage law was also used. Experimentally, it was found that Pm increases exponentially with α decrease, except for joints with adherends with the stacking sequence [45/0/-45/90]3S. The stress distribution in the adhesive is not uniform and depends on α and the adhesive stacking sequence. Comparison of results I.R.S. ARAÚJO • R.D.S.G. CAMPILHO 25 shows that the user-defined CZM was able to predict joint strength and displacement to failure with higher accuracy than the triangular damage law, which underestimated the experimentally obtained values. The objective of this work is the parametric numerical study of scarf adhesive joints in tension with different adhesives (Araldite® AV138, Araldite® 2015 and Sikaforce® 7752) and different scarf angles or α (3.43°, 10°, 15°, 20°, 30° and 45°) by the XFEM. Initially, the experimental results obtained in a previous work are described, for the purpose of validating the obtained numerical results. The developed numerical work includes the distribution of the damage variable and joint strength. With the work carried out, the coherence of the numerical results with the experimental ones was observed, with emphasis on the joint strength as a function of α. 2. Materials and methods 2.1. Joint geometry Figure 1 illustrates the geometry of the scarf joint. Its dimensional specifications are as follows (in mm): length Lt=170, adherend thickness tP=3 and adhesive thickness tA=0.2. These parameters remain constant, with only the scarf angle α (3.43o, 10o, 15o, 20o, 30o and 45o) varying, to study its influence on the joint’s strength. Figure 1. Scarf joint geometry. 2.2. Materials The material used as adherend was the aluminium alloy AW 6082T651. The selection of this material is not only due to its good mechanical properties but also to its wide range of structural applications in extruded and rolled forms. This aluminium alloy was characterized in previous works [13], where the following properties were defined: tensile strength (𝜎f) of 324.00±0.16 MPa, Young’s modulus (E) of 70.07±0.83 GPa, tensile NUMERICAL EVALUATION OF SCARF GEOMETRY ADHESIVELY-BONDED JOINTS BY XFEM MODELLING 26 yield stress (𝜎y) of 261.67±7.65 MPa, and tensile fracture strain (εf) of 21.70%±4.24%. The stress-strain curves (σ-ε) of the aluminium adherends were obtained experimentally, in accordance with ASTM standard E8/E8M [14], to be introduced in the numerical models. The adhesives considered for this work are the Araldite® AV138, a brittle epoxy adhesive, the Araldite® 2015, a ductile epoxy adhesive, and the Sikaforce® 7752, a highly ductile polyurethane adhesive. Table 1 presents all the relevant properties of the adhesives, characterized in a previous work, along with their respective values. Table 1. Properties of the adhesives Araldite® AV138, Araldite® 2015, and Sikaforce® 7752 [15-17]. Property AV138 2015 7752 Young’s modulus, E [GPa] 4.89±0.81 1.85±0.81 0.493±0.0896 Poisson’s ratio, v0.35a0.33a0.33a Tensile yield stress, 𝜎y [MPa] 36.49±2.47 12.3±0.61 3.24±0.5 Tensile failure strength, 𝜎f [MPa] 39.45±3.18 21.63±1.61 11.49±0.3 Tensile failure strain, εf [%] 1.21±0.10 4.77±0.15 19.18±1.4 Shear modulus, G13 [GPa] 1.56±0.01 0.56±0.21 0.187±0.0164 Shear yield stress, τy [MPa] 25.1±0.33 14.6±1.3 5.16±1.1 Shear failure strength, τf [MPa] 30.2±0.40 17.9±1.8 10.17±0.6 Shear failure strain, γf [%] 7.8±0.7 43.9±3.4 58.42±6.4 Toughness in tension, GIc [N/mm] 0.2b0.43±0.02 2.36±0.2 Toughness in shear, GIIc [N/mm] 0.38b4.7±0.34 5.41±0.5 a Manufacturer’s data b Estimated in reference [18] 2.3. Numerical modelling The numerical analysis of the scarf joint was carried out using the FEM software Abaqus®, since it allows the use of the integrated XFEM module for predicting the strength of the scarf joint. The joints were configured in a two-dimensional format, employing solid plane strain elements (specifically CPE4 and CPE3 in ABAQUS®) to model the adherends. The triangular CPE3 elements were used in the scarf edge to enable the respective slope without inducing distortions in the mesh. Figure 2 illustrates a detail of mesh refinement for a model with α=45º. I.R.S. ARAÚJO • R.D.S.G. CAMPILHO 27 Figure 2. Representation of the mesh constituent elements. In the longitudinal direction of the adherends, a selective mesh refinement was employed using the bias ratio. The mesh has a higher level of refinement near the adhesive layer. The number of elements and the refinement ratio for each edge of the joint were chosen to ensure greater refinement in the critical regions of the joint. This variation aims to reduce computational effort and time in obtaining results without compromising their accuracy. Figure 3 illustrates the effect of the bias along the length of the scarf joint. Figure 3. Bias effect along the length of the joint. To simulate real experimental test conditions, boundary and loading conditions were applied to the models in the ABAQUS® software to emulate real testing conditions. The joint was fixed at one end and constrained in the transverse direction at the opposite end. 2.4. XFEM formulation The XFEM serves as an enhancement to the conventional FEM. The XFEM integrates enrichment functions into the FEM formulation, primarily designed to represent displacement jumps between crack faces during crack propagation [19]. When simulating damage within Abaqus®, damage initiation and propagation are triggered in regions where the stresses and/or NUMERICAL EVALUATION OF SCARF GEOMETRY ADHESIVELY-BONDED JOINTS BY XFEM MODELLING 28 strains exceed predetermined thresholds. Abaqus® offers a choice of six crack initiation criteria. Among these, the MAXPS (maximum principal stress) and MAXPE (maximum principal strain) criteria rely on specific functions, as described in the respective order max max oo max max orff σε σε   = =     (1) σmax and σo max represent the current and permissible maximum principal stress. The use of Macaulay brackets signifies that a purely compressive stress state does not lead to damage initiation. Similarly, εmax and εo max represent the current and allowable maximum principal strain. The MAXS (maximum nominal stress) and MAXE (maximum nominal strain) criteria are expressed using the following mathematical functions nn ss 00 0 0 ns n s max , or max , tt ff tt εε εε    = =       (2) tn and ts are the current normal and shear traction components to the cracked surface. The strain parameters have identical significance. The quadratic nominal stress (QUADS) and quadratic nominal strain (QUADE) criteria are based on the introduction of the following functions, respectively 22 22 nn ss 00 0 0 ns n s or tt ff tt εε εε       =+=+           (3) All criteria are fulfilled, and damage initiates, when f reaches unity. For damage growth, the fundamental expression of the displacement vector u, including the displacements enrichment, is written as [20] ( ) ( ) 1 N i i N x Hx = = +   ∑ u ua ii . (4) Ni(x) and ui relate to the conventional Finite Element formulation. Ni(x) represents the nodal shape functions, while ui stands for the nodal displacement vector associated with the continuous part of the formulation. I.R.S. ARAÚJO • R.D.S.G. CAMPILHO 29 The second term enclosed in brackets, H(x)ai, is only active in the nodes for which any relating shape function is cut by the crack and can be expressed by the product of the nodal enriched degree of freedom vector including the mentioned nodes, ai, with the associated discontinuous shape function, H(x), across the crack surfaces. This method is built upon the concept of introducing phantom nodes that subdivide elements intersected by a crack, effectively simulating the separation between newly created sub-elements. The ability to propagate a crack along an arbitrary path is facilitated by these phantom nodes, which initially share the same coordinates as the real nodes. These phantom nodes remain entirely constrained to the real nodes until damage initiation occurs. Once crossed by a crack, the element gets divided into two sub-domains. The discontinuity in displacements is achieved by adding phantom nodes atop the original nodes. When an element undergoes cracking, each of the two resulting sub-elements consists of real nodes (those corresponding to the cracked portion) and phantom nodes (those no longer belonging to the respective part of the original element). These two elements exhibit fully independent displacement fields and replace the original one. From this point onward, each pair of real/phantom nodes in the cracked element can separate following an appropriate cohesive law until failure. At this stage, the real and phantom nodes are free to move without constraints, effectively simulating crack growth. A softening XFEM law is considered, employing an energetic failure power law criterion of a specific type I II IC IIC 1. GG GG αα    +=       (5) where GI and GII represent the current fracture energies in tension and shear, respectively. The analysis in this work focused on the QUADS initiation criterion and a linear damage law with a power parameter of α=1. 3. Results 3.1. Experimental results Figure 4 presents the experimental Pm average values and their respective standard deviation (STDV) as a function of α, for the three evaluated adhesives. From data analysis, a similar strength evolution with α is noticeable among them. Pm of all adhesives is found for α=3.43o, while for the other angles Pm diminish and the lowest reduction was attained between α=3.43o and α=10o. NUMERICAL EVALUATION OF SCARF GEOMETRY ADHESIVELY-BONDED JOINTS BY XFEM MODELLING 30 Comparing the performance of all adhesives, the Araldite® AV138 stands out, with higher Pm for α=3.43o by 35.6 and 120.0% when compared with the Araldite® 2015 and the SikaForce® 7752, respectively. 0 10 20 30 40 010 20 30 40 50 P m [kN] α[º] AV138 2015 7752 Figure 4. Experimental Pm average results and STDV as function of α for the three adhesives. Table 2 shows the average experimental Pm values together with STDV and coefficients of variation (CoV). A higher CoV dispersion was found for the joints with α=45o bonded with the Araldite® 2015 and Araldite® AV138, of 17.43 and 11.03%, respectively, which may result from specimens’ manufacture or issues associated with testing procedures. The SikaForce® 7752 presented higher deviations with the scarf architectures of α=15o, α=20o and α=30o. All remaining values are acceptable. Table 2. Pm average values, STDV, and CoV for scarf joints bonded with the different adhesives. α [o] 3.43 10 15 20 30 45 Araldite® AV138 STDV [N] 29700 1500 10648 346.63 7926 642.49 5715 310.02 4480 150.89 3325 366.66 CoV [%] 5.05 3.26 8.11 5.42 3.37 11.03 Araldite® 2015 STDV [N] 21903 1600 7509 772.04 4858 141.92 3567 205.56 2832 101.03 1868 325.56 CoV [%] 7.30 10.28 2.92 5.76 3.57 17.43 Sikaforce® 7752 STDV [N] 13500 510.90 4677 195.40 3132 286.25 2378 229.38 1543 167.86 1142 63.51 CoV [%] 3.78 4.18 9.14 9.65 10.88 5.56 I.R.S. ARAÚJO • R.D.S.G. CAMPILHO 31 Data from Table 2 shows that, while α diminishes, Pm increases for the different evaluated scarf angles and adhesives. Pm variations progressively increase for joint configurations of α=45o, 30o, 20o, 15o and 10o, whereas between α=10o and α=3.43o a more pronounced variation occurs. Comparing all adhesives, the Araldite® AV138 was the best performing adhesive since it attains the highest Pm for the different tested α. Contrarily, the SikaForce® 7752 presented the worst behaviour. Taking the α=3.43o configuration as baseline, the brittle Araldite® AV138 presented a Pm of 29.7 kN, whereas the ductile Araldite® 2015 attained 21.9 kN. As mentioned above, the SikaForce® 7752, with Pm of 13.5 kN, was the worst performing adhesive. Thus, considering the Pm evaluation for the three adhesives, the Araldite® AV138 is the best choice for the scarf joints, followed by the Araldite® 2015, which presented quite reasonable results. Moreover, heterogeneity was found in the obtained CoV. Nonetheless, both Araldite® AV138 and SikaForce® 7752 presented a lower dispersion of the results obtained in the tests. 3.2. Damage analysis This section presents the study of the stiffness degradation (SDEG) damage variable for the various joint configurations. This variable represents the percentile degradation of the cracked elements compared to a purely linear behaviour. The study of the SDEG damage variable of the elements of the adhesive layer, along x/LO, is one of the tools that helps to compare the various joint configurations for the different adhesives. This variable has values between 0 (undamaged material, up to peak stress) and 1 (failure), providing the degradation of the stiffness of the damage law under mixedmode conditions. Figure 5 shows the extent of damage (SDEG) at the instant of Pm with the normalised length of the adhesive layer (x/L), in scarf joints with different α bonded with the adhesives Araldite® AV138 (a), Araldite® 2015 (b) and Sikaforce® 7752 (c). Plot analysis shows that the greatest incidence of damage in this type of joint takes place at the adhesive layer ends, thus in agreement with the distribution of stresses presented in previous works [11]. At the middle of the adhesive layer, damage is typically zero at the moment Pm is reached. The comparison between different α shows that, as this geometrical parameter increases, there is less localised damage at the ends and more uniform damage throughout the adhesive layer, which also agrees with the documented variation of stress distributions as a function of α [21], since stresses also become more uniform for higher values of α. Between adhesives, it can be seen that, as their stiffness increases, the magnitude of the damage in the various α increases and is concentrated in smaller areas at the ends of the bond, showing more oscillations. On the other hand, the 39 CHAPTER 4 Numerical CZM evaluation of adhesivelybonding solutions for canoeing boat fabrication João C.M. Santos1 Raul D.S.G. Campilho2 Abstract Canoeing is a nautical sport that appeared in history thousands of years ago simply as a means of survival. Nowadays, this sport is practiced all over the world as a hobby or as a means of competition. Given the desire to improve the quality of construction and performance of boats, currently their manufacture is focused on the use of composite materials, which can be conveniently bonded with adhesives. On the other hand, for a manufacturing company of these boats to remain competitive in the market, it is required the continuous improvement of these joints in terms of strength and manufacturing cost. In this work, an adhesive joint existing in a canoeing boat is numerically studied by cohesive zone modelling (CZM), more specifically the joint between the hull and the deck of a kayak. To achieve the goal of this work, a numerical 1 Departamento de Engenharia Mecânica, Instituto Superior de Engenharia do Porto, Instituto Politécnico do Porto, R. Dr. António Bernardino de Almeida, 431, 4200-072 Porto, Portugal. 2 Departamento de Engenharia Mecânica, Instituto Superior de Engenharia do Porto, Instituto Politécnico do Porto, R. Dr. António Bernardino de Almeida, 431, 4200-072 Porto, Portugal. INEGI – Pólo FEUP, Rua Dr. Roberto Frias, 400, 4200-465 Porto, Portugal. NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 40 analysis is performed, in which the existing joint configuration was tested, different geometric changes were analysed, and different types of adhesives were considered. A prior validation of the CZM technique was performed with experimental data obtained in a previous work. Initially, CZM was positively validated, followed by the best geometry-adhesive combination, which significantly improved on the current used joint. Keywords: Canoeing, Adhesive joints, Structural adhesive, Composites, Numerical modelling, Finite element method, Cohesive zone model. 1. Introduction Since ancient times, man has been trying to explore the aquatic environment. Canoeing was one of the first means of travelling on water. Because of the need to survive, human beings seek out rivers and seas to fish, hunt or even navigate them. The oldest evidence of this sport was found in the tomb of a Sumerian king during archaeological excavations near the river Euphrates (Western Asia) and dates back around six thousand years [1]. Throughout this long history, two types of vessels have emerged that generally characterise the sport throughout the world. One in a “closed” style, propelled by a double-bladed paddle known as a kayak; and the other in an “open” style, propelled by a single-bladed paddle known as a canoe. The Kayak originated in Greenland, where the Eskimos used it as a means of individual transport and for hunting and fishing activities [2]. Adhesive bonding is increasingly present in society and is seen as an alternative to traditional methods. Considering that traditional methods can only be applied to materials above certain minimum thickness values to avoid damaging the material during joining (e.g., tearing, bending, burning), adhesives are advantageous because they can be applied to materials of any thickness, and eliminate the need for drilling, which results in no stress concentrations at these points. With the most innovative adhesives, it is now possible to join most dissimilar materials, such as metals, plastics, or glass. In addition, adhesive connections are associated with lower manufacturing costs, higher fatigue resistance and high vibration damping capability [3]. However, adhesive joints also present numerous disadvantages, such as the need for careful preparation of the adherends, a very long curing time, repair difficulties, among others [4]. Various joint configurations have been proposed and used by designers. One of the most widely used geometries is the single-lap joint, due to the simplicity of its manufacturing process [5, 6]. This configuration has a major drawback: when subjected to tensile loads, the asymmetry of the stress transfer lines causes deflection of the joint, leading JOÃO C.M. SANTOS • RAUL D.S.G. CAMPILHO 41 to peel stresses (σy) at the edges of the overlap, which has a detrimental impact on its performance [7]. Aiming to overcome the aforementioned limitations, several adhesive joint configurations have been suggested, as is the case of the double-lap, scarf, stepped, among others [8, 9]. Over the years, various approaches have been proposed to predict the strength of adhesive joints, such as analytical methods like Volkersen’s pioneering formulations [10, 11] or numerical methods like the ones based on the Finite Element Method (FEM) [12, 13]. Currently, one of the most recognised and widely adopted methods is Cohesive Zone Modelling (CZM). In combination with FEM and employing fracture mechanics concepts, this method allows for an accurate prediction of the adhesive joints’ strength. Using the CZM method it is possible to simulate the onset and propagation of cracks or delamination in composite materials, or to analyse cohesive and interfacial defects [14]. Nowadays, new numerical methods have been developed that can be used to predict the strength of adhesive joints, as is the case of the eXtended finite element method (XFEM) [15, 16] and meshless methods [17, 18]. Several researchers have studied adhesive joints applied in the context of naval structures. Alderucci et al. [19] experimentally analysed the effects of different surface patterns produced by mechanical engraving on composite or aluminium adhesive joints. Four types of patterns were considered for the overlap joints (no engraving, pattern 0°, pattern 45° and pattern 45°X), which were subjected to tensile tests. In view of the results found, the Al/ Al joints showed higher strength in the 45° and 45°X patterns, while the specimen with the 0° pattern showed less strength than the “no pattern” specimen, i.e., when comparing these two the production of a pattern does not bring any advantages. The Al/composite joints showed higher strength when manufactured with some kind of pattern. In conclusion, in naval applications where lightweight materials need to be joined (e.g., aluminium and/or composites) the existence of grooves can help improve the mechanical strength of the adhesive joint. Ulus et al. [20] performed a study of the fracture and dynamic mechanical behaviour of hybrid adhesive joints after long-term ageing in seawater. The aim was to develop mode I and mode II delamination resistance and glass transition temperature (Tg) data as a comprehensive design guideline for modified basalt fibre reinforced polymers (BFRP)-aluminium hybrid joints subjected to seawater ageing (six months’ exposure). In addition, the adhesive was reinforced with Halloysite nanotubes to enhance fracture toughness and to slow down water absorption. After exposure, the adhesively bonded reinforced joints exhibited ∼36% higher fracture resistance than the non-exposed bonded joints. Darla et al. [21] presented a study evaluating the strength of the connection between NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 42 an aluminium hub and a carbon fibre reinforced polymer (CFRP) propeller blade profile for marine applications. The Araldite® 2011 adhesive was used to bond the components, and the assembly was subjected to axial thrust conditions. The authors also carried out a numerical study given the difficulty in assessing the strength of the experimental joint due to the complex geometry, and concluded that the bonded joint specimens between the aluminium hub and the CFRP propeller blade profile were influenced by the prolonged duration of ageing at the interface between the aluminium hub and the adhesive, resulting in joint failure at lower load. In the work presented in reference [22], the authors proposed a global approach to assess the integrity of a full-scale adhesively bonded bi-material joint for marine applications. The joint represents a cross-section of the adhesive bond between a steel hull and a sandwich composite superstructure used in the maritime industry. The joint was subjected to a tensile quasi-static load profile including six load cycles till the final collapse of the specimen. During the tests, the authors used three structural integrity monitoring techniques, namely acoustic emission, fibre optic sensor and digital image correlation, to establish the state of damage. In addition, a FEM model was developed to reproduce the damage mechanisms and compare the numerical results with the experimental ones. The combination of all used techniques could detect the onset of damage, evaluate the extent of damage, identify the critical regions and differentiate the different damage mechanisms. In this work, an adhesive joint existing in a canoeing boat is numerically studied by CZM, more specifically the joint between the hull and the deck of a kayak. To achieve the goal of this work, a numerical analysis is performed, in which the existing joint configuration was tested, different geometric changes were analysed, and different types of adhesives were considered. A prior validation of the CZM technique was performed with experimental data obtained in a previous work. 2. Materials and methods 2.1. Joint geometries In this work the goal is to optimize a bonded joint employed in kayak manufacturing. Given the complexity of the structure, it was decided to reduce the study of the adhesive joint to just one section of the vessel. For this purpose, the front area of a single-seater kayak’s (K1) was considered, which externally is subjected to small impacts caused by paddling or other impacts due to a collision between boats and, internally, is an attachment JOÃO C.M. SANTOS • RAUL D.S.G. CAMPILHO 43 point for the footrest. In fact, this section is an essential point for supporting the user’s feet and, consequently, a large part of the force applied to the water for the movement of the vessel during paddling will be unloaded in this area (Fig. 1 a). Fig. 1 (b) shows the sectional view of the section that will be analysed. The red line refers to the hull while the blue line refers to the deck of the kayak. Currently the two parts are bonded considering a butt joint configuration (Fig. 1 c). a) b) c) Fig. 1. Zone of the Kayak under analysis a), sectional view of the kayak b) and detail of the sectional view c). In addition to the current butt joint geometry (Fig. 2 a), two more different geometries (Fig. 2 b and c) are proposed and analysed. Both solutions are chamfer-type joints considering a chamfer or scarf angle of 45°, but symmetrically arranged and an adhesive thickness of 0.2 mm. Although these geometries have similar characteristics, the curvature of the kayak and the different thickness of the parts to be joined will produce different behaviours in the adhesive layer. Fig. 2 (b) shows the first type of the chamfered geometry (chamfer 1). In this case, seen from the outside, the deck will overlap the hull. The second type (chamfer 2) is presented in Fig. 2 (c), in which the hull overlaps the deck. These configurations will be subjected to four different types of loading, namely traction, compression, bending and shear, to determine the best configuration and improve the current assembly. NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 44 a) b) c) Fig. 2. Schematic representation of the butt joint (a), chamfer 1 (b) and chamfer 2 (c). For numerical model validation, the single-lap joint geometry was considered (Fig. 3). As boundary conditions, the joint was clamped at one of the edges, and the opposite edge was subjected to a horizontal displacement with vertical movement restriction. The main dimensions of the analysed specimen are the overlap length (LO)=12.5, 25, 37.5, and 50 mm, adherends’ thickness (tP)=3 mm, adhesive thickness (tA)=0.2 mm, joint total length (LT)=170 mm, and width (B)=25 mm. Fig. 3. Schematic representation of the single-lap joint geometry and boundary conditions. 2.2. Materials To provide a wider range of results, in addition to the main adhesive, three other adhesives were selected to evaluate the different adhesive joint geometries to bond the hull and deck of a kayak. The Sika Adekit® A 140-1 JOÃO C.M. SANTOS • RAUL D.S.G. CAMPILHO 45 [23] has excellent mechanical performance and resistance to dynamic loads (vibrations and impacts). In addition, it is versatile enough to be applied both vertically and horizontally, being suitable for filling irregular joints and being used in aggressive environments. To ensure a wide range of ductilities, the brittle epoxy Araldite® AV138 [24], the moderately ductile epoxy Araldite® 2015 [25, 26], and the ductile polyurethane Sikaforce® 7752 [27] were also considered. Table 1. Mechanical and fracture properties of the four adhesives. Property Adekit® A 140-1 Araldite® AV138 Araldite® 2015 Sikaforce® 7752 Young’s modulus, E [GPa] 2.57 4.89±0.81 1.85±0.81 0.493±0.0896 Poisson coefficient, v0.35a0.33a0.33a Tensile yield stress, σy [MPa] - 36.49±2.47 12.3±0.61 3.24±0.5 Tensile strength, σf [MPa] 30 39.45±3.18 21.63±1.61 11.49±0.3 Tensile failure strain, εf [%] 4 1.21±0.10 4.77±0.15 19.18±1.4 Shear modulus, G [GPa] 0.99 1.56±0.01 0.56±0.21 0.187±0.0164 Shear yield stress, τy [MPa] - 25.1±0.33 14.6±1.3 5.16±1.1 Shear strength, τf [MPa] 20 30.2±0.40 17.9±1.8 10.17±0.6 Shear failure strain, γf [%] - 7.8±0.7 43.9±3.4 58.42±6.4 Tensile toughness, GIc [N/ mm] 0.5 0.2b0.43±0.02 2.36±0.2 Shear toughness, GIIc [N/ mm] 5.56 0.38b4.7±0.34 5.41±0.5 a Manufacturer’s value b Obtained in reference [28] The geometry of the hull and deck have different configurations. As the hull is more prone to impacts and other stresses, more layers of reinforcements are applied to make the hull more resistant. In the area of higher stress (the central part of the boat, where the user will be), the hull is made up of five layers, while the deck is only made up of three layers in all. Fig. 4 shows the correct stacking order for the deck adherend (left) and the hull adherend (right). NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 46 Fig. 4. Stacking order for the deck adherend (left) and the hull adherend (right). Table 2 shows the properties of the materials that constitute the sandwich structure of a boat. The (A) carbon fibre [29] and (B) glass fibre [30] come in the form of fabric and the core of the kayak is (C) PVC foam [31]. To connect these materials, the matrix Resolcoat 1400-1407 epoxy resin was used [32]. Table 2. Properties of the boat materials. Property (A) Carbon fibre (B) Glass fibre (C) PVC foam Resolcoat Young’s modulus, E [MPa] 240000 33000 95 3100-3300 Yield stress, σf [MPa] 4200 626 2.5 - Fibre density, ρ [g/cm3]1.78 2.6 0.08 1.15 Area weight, Gr [g/m2] 160 220 - - Thickness, t [mm] 0.30 0.25 3 - Shear modulus, G [MPa] - - 27 - Poisson coefficient, ν- - 0.4 0.35 Shear yield stress, τf [MPa] - - 1.15 - 2.3. Numerical modelling The numerical models were developed in software Abaqus®. To simplify the procedure while guaranteeing accuracy of the results, a 2D representation of the geometry was chosen. The joint to be modelled is a butt joint with curved adherends, to replicate the joint between the hull and deck of a kayak. A triangular CZM is used to simulate failure in the adhesive layer connecting the hull to the deck. The materials, both the hull and the deck, are linear elastic and isotropic. A thickness of 0.2 mm was considered for the adhesive. The adherends were dimensioned for a length of 50 mm. In terms of thickness, the deck adherend is 3.6 mm thick, while the hull adherend JOÃO C.M. SANTOS • RAUL D.S.G. CAMPILHO 47 is 4.2 mm thick, centred on each other, i.e., with 0.3 mm offset on each side. The joint was subdivided into sections to allow the differentiation of each material and the subsequent attribution of their distinct properties. For the PVC core, a thickness of 3 mm was assumed, while for the GFRP and CFRP, a thickness of 0.3 mm was considered for each layer. For each section, the mechanical properties and their respective behaviour were defined from the data of Table 1 and Table 2. PVC, CFRP and GFRP were defined as homogeneous solid materials, while the adhesive was defined as a cohesive material. The software Abaqus® performs the simulation on an incremental basis. To facilitate convergence during the damage propagation phase, the minimum allowed increment size (in % of the applied loading) is 1×10-20, and the maximum increment size is equal to the initial value (0.5%). Boundary conditions were established to emulate different possible loadings applied to the joint. To this end, a geometric restriction was applied to the lower adherend, while a displacement condition was applied to the upper adherend. For the adherend layers, a structured element definition was selected, while a sweep element definition was used for the adhesive. Nonetheless, to analyse stress distributions, a structured element definition was considered for the adhesive layer as well, which makes it possible to extract stresses with precision at the adhesive mid-thickness. CPE4 elements were used for the adherend layers, while COH2D4 elements were used for the adhesive. To obtain the stress distributions in the adhesive, CPE4 elements were also used in the corresponding partition. In all cases, a viscosity of 1×10-5 Pa.s. for the elements were considered. Bias (size grading effects) in the constructed mesh ensured good refinement at the most critical locations to extract the stresses at these locations more accurately. In the CZM damage propagation models, in which the strength of the joint is studied, the tests are carried out until attaining the initially imposed displacement, so that the joint failure is achieved. 2.4. CZM description Cohesive Zone Models (CZM) establish a damage law between paired nodes within cohesive elements that relates stresses and displacements, enabling modelling the structures’ behaviour up to their mechanical limits [33]. Over time, numerous models have been developed, each tailored to different material and behaviours, offering enhanced accuracy, as demonstrated by [34]. In this study, the triangular shaped law was chosen for both adhesive materials. Its simplicity in configuration and usability within numerical analysis software makes it an attractive choice. Notably, it delivers precise results in a remarkably efficient manner. This model’s NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 54 • For the adhesive currently used in the manufacture of boats, the Sika Adekit® A140-1 adhesive, the most suitable geometry is chamfer 2, followed by the butt geometry and finally the chamfer 1 geometry. The chamfer 2 geometry obtains the highest Pm for tensile and shear stresses, and the second highest value for the other two loadings. The best geometry under compression is butt, and under bending is chamfer 1. However, by comparing the average Pm for the chamfered and butt geometries and for the four loadings, the chamfer 1 geometry is ultimately the best geometry for this adhesive; • For the Araldite® AV138, the geometry that gives it the best properties compared to the other adhesives is the butt geometry, followed by the chamfer 2 geometry and finally the chamfer 1 geometry. By comparing the Pm values for this adhesive, the chamfer 2 geometry provides the best strength under bending and shear, and the second-best strength under axial tension. The butt geometry once again gives the best Pm for this adhesive in compression, and the best tensile strength is given by the chamfer 1 geometry. In short, the average values obtained for the different loadings show that the chamfer 2 geometry provides the best characteristics overall; • Analysing the Araldite® 2015 shows that the geometry that gives the best characteristics, compared to the other adhesives, is the chamfer 1 geometry, followed by the chamfer 2 geometry and finally the butt geometry. The chamfer 1 geometry has the best resistance to tensile, bending and shear loadings. The butt geometry has the best performance under compressive loadings in this adhesive, but the worst strength for other loadings. As shown in Table 3, it can be concluded that the chamfer 1 geometry is the best of the three studied geometries if this adhesive is used. Comparing the average Pm obtained for each geometry, the chamfer 1 geometry once again stands out; • The Sikaforce® 7752 gives the best behaviour for the chamfer 1 geometry, followed by the butt geometry and, finally, the chamfer 2 geometry. According to the Pm values, the chamfer 1 geometry provides the best strength for this adhesive under tensile, bending and shear loadings. On the other hand, the butt geometry once again provides the best strength under a compressive loading. BY comparing the general behaviour of the chamfer geometries in relation to the butt geometry, it can be concluded that the chamfer 1 geometry guarantees the best characteristics for this adhesive. JOÃO C.M. SANTOS • RAUL D.S.G. CAMPILHO 55 Analysing the fabrication procedure for the different geometries shows that the butt geometry has the shortest production time and consequently the lowest production costs, since no surface preparation is carried out and the adhesive is placed directly on the joint. For the chamfered geometry, whether type 1 or type 2, it is necessary to prepare the joint and fabricate the chamfered surfaces, which will increase manufacturing times and consequently production costs. Thus, this feature should be considered in the design process. 4. Conclusions This work aimed to optimize the strength of an adhesive joint formed between the hull and deck of a boat used in canoeing, and subsequently propose geometric solutions and types of adhesive that could improve the current configuration of the adhesive joint. The CZM technique was initially validated with standard SLJ geometries and three adhesives. The observed predictive capacity was quite satisfactory for the two Araldite® adhesives, while moderate underpredictions were found for the 7752, although this does not invalidate the method. τxy/τavg and σy/τavg stresses showed stress concentrations at the ends of the adhesive for all loading types. By comparing the different geometries, the butt geometry has smallest peak stresses under a shear loading, the chamfer 1 geometry under tensile and compressive loadings, and the chamfer 2 geometry under bending loadings. Between adhesives, the AV138 had the highest peaks on account of its stiffness. The strength analysis showed that, for the butt geometry, the AV138 maximizes Pm. For the chamfer 1 geometry, the 7752 is the best solution, and for the chamfer 2 geometry, the AV138 performs best. In view of all obtained results, increasing the boat strength without significantly increasing production time is accomplished by applying the AV138 in a butt geometry. For the best possible configuration disregarding costs and fabrication time, there are two possible configurations: the 7752 applied to the chamfer 1 geometry, and the AV138 applied to the chamfer 2 geometry. About the latter two configurations, the first excels in smaller τxy and σy stresses. Furthermore, since a more ductile adhesive is used in the first configuration, it will better resist the stresses experienced. NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 56 References 1. Starr, C.G., A history of the ancient world. 1991: Oxford University Press, USA. 2. Vernon, J., The Lincoln Kayak. The Mariner’s Mirror, 1984. 70(4): p. 415-426. 3. Petrie, E.M., Handbook of adhesives and sealants. 2007: McGraw-Hill Education. 4. Romano, M.G., M. Guida, F. Marulo, M. Giugliano Auricchio, and S. Russo, Characterization of Adhesives Bonding in Aircraft Structures. Materials, 2020. 13(21): p. 4816. 5. Tsai, M.Y. and J. Morton, An evaluation of analytical and numerical solutions to the single-lap joint. International Journal of Solids and Structures, 1994. 31(18): p. 2537-2563. 6. Öztoprak, N. and G.M. Gençer, Load-bearing capacity of polyamide 6 (PA6) composite to 7075-O aerospace Al-alloy single-lap joints: influence of various laser textured patterns on hot press bonding. Journal of Adhesion Science and Technology, 2023: p. 1-19. 7. Hart-Smith, L.J., Adhesive-bonded single-lap joints. 1973, NASA Contract Report, NASA CR-112236. 8. Petrie, E.M., The fundamentals of adhesive joint design and construction: Function-specific construction is the key to proper adhesion and load-bearing capabilities. Metal Finishing, 2008. 106(11): p. 55-57. 9. Adams, R.D., J. Comyn, and W.C. Wake, Structural adhesive joints in engineering. 2nd ed. 1997, London, United Kingdom: Chapman & Hall. 10. Volkersen, O., Die nietkraftoerteilung in zubeanspruchten nietverbindungen mit konstanten loschonquerschnitten. Luftfahrtforschung, 1938. 15 p. 41-47. 11. Tserpes, K., A. Barroso-Caro, P.A. Carraro, V.C. Beber, I. Floros, W. Gamon, M. Kozłowski, F. Santandrea, M. Shahverdi, D. Skejić, C. Bedon, and V. Rajčić, A review on failure theories and simulation models for adhesive joints. The Journal of Adhesion, 2021: p. 1-61. 12. He, X., A review of finite element analysis of adhesively bonded joints. International Journal of Adhesion & Adhesives, 2011. 31(4): p. 248-264. 13. de Sousa, C.C.R.G., R.D.S.G. Campilho, E.A.S. Marques, M. Costa, and L.F.M. da Silva, Overview of different strength prediction techniques for single-lap bonded joints. Journal of Materials: Design and Application - Part L, 2017. 231: p. 210-223. 14. Saeedifar, M., M. Ahmadi Najafabadi, J. Yousefi, R. Mohammadi, H. Hosseini Toudeshky, and G. Minak, Delamination analysis in composite laminates by means of Acoustic Emission and bi-linear/tri-linear Cohesive Zone Modeling. Composite Structures, 2017. 161: p. 505-512. 15. Belytschko, T. and T. Black, Elastic crack growth in finite elements with minimal remeshing. International Journal for Numerical Methods in Engineering, 1999. 45(5): p. 601-620. JOÃO C.M. SANTOS • RAUL D.S.G. CAMPILHO 57 16. Xará, J.T.S. and R.D.S.G. Campilho, Strength estimation of hybrid single-L bonded joints by the eXtended Finite Element Method. Composite Structures, 2018. 183: p. 397-406. 17. Chen, J.-S., C. Pan, C.-T. Wu, and W.K. Liu, Reproducing Kernel Particle Methods for large deformation analysis of non-linear structures. Computer Methods in Applied Mechanics and Engineering, 1996. 139(1): p. 195-227. 18. Ramalho, L.D.C., R.D.S.G. Campilho, and J. Belinha, Predicting single-lap joint strength using the natural neighbour radial point interpolation method. Journal of the Brazilian Society of Mechanical Sciences and Engineering, 2019. 41(9): p. 362. 19. Alderucci, T., C. Borsellino, and G. Di Bella, Effect of surface pattern on strength of structural lightweight bonded joints for marine applications. International Journal of Adhesion and Adhesives, 2022. 117: p. 103005. 20. Ulus, H., H.B. Kaybal, F. Cacık, V. Eskizeybek, and A. Avcı, Fracture and dynamic mechanical analysis of seawater aged aluminum-BFRP hybrid adhesive joints. Engineering Fracture Mechanics, 2022. 268: p. 108507. 21. Darla, V., B. Satish Ben, and K.V. Sai Srinadh, Investigation on interfacial bonding strength between aluminum hub and CFRP bonded joints exposed to accelerated aging conditions. Journal of Adhesion Science and Technology: p. 1-13. 22. Saeedifar, M., M.N. Saleh, A. Krairi, S.T. de Freitas, and D. Zarouchas, Structural integrity assessment of a full-scale adhesively-bonded bi-material joint for maritime applications. Thin-Walled Structures, 2023. 184: p. 110487. 23. Sika, Product data sheet ADEKIT A140-1 / H9940-1. 2020. 24. Neto, J.A.B.P., R.D.S.G. Campilho, and L.F.M. da Silva, Parametric study of adhesive joints with composites. International Journal of Adhesion and Adhesives, 2012. 37: p. 96-101. 25. Campilho, R.D.S.G., M.D. Banea, J.A.B.P. Neto, and L.F.M. da Silva, Modelling adhesive joints with cohesive zone models: effect of the cohesive law shape of the adhesive layer. International Journal of Adhesion & Adhesives, 2013. 44: p. 48-56. 26. Campilho, R.D.S.G., A.M.G. Pinto, M.D. Banea, and L.F.M. da Silva, Optimization study of hybrid spot-welded/bonded single-lap joints. International Journal of Adhesion and Adhesives, 2012. 37: p. 86-95. 27. Faneco, T.M.S., Caracterização das propriedades mecânicas de um adesivo estrutural de alta ductilidade. 2014, Tese de Mestrado em Engenharia Mecânica – Ramo de Materiais e Tecnologias de Fabrico. Instituto Superior de Engenharia do Porto: Porto. 28. Campilho, R., Adhesive, welded and weld-bonded single-lap joints: Numerical technique for strength prediction. Vol. 24. 2012. 35-42. 29. HAUFLER COMPOSITES Techinacal data sheet Woven Carbon Fibre Fabric 160 g/m², Plain 2017. 30. Interglas Porcher Industries Glass Filament Fabrics for Plastics Reinforcement - Product Specification. 2019. NUMERICAL CZM EVALUATION OF ADHESIVELY-BONDING SOLUTIONS FOR CANOEING BOAT 58 31. Diab Group Techincal data Divinycell H. 2021. 32. Résoltech, Data sheet RESOLCOAT 1400-1407 and Accelerator AC140. 33. Alfano, M., F. Furgiuele, A. Leonardi, C. Maletta, and G.H. Paulino, Cohesive Zone Modeling of Mode I Fracture in Adhesive Bonded Joints. Key Engineering Materials, 2007. 348-349: p. 13-16. 34. Campilho, R.D., M.D. Banea, J. Neto, and L.F. da Silva, Modelling adhesive joints with cohesive zone models: effect of the cohesive law shape of the adhesive layer. International Journal of Adhesion & Adhesives, 2013. 44: p. 48-56. 35. Valente, J.P.A., R.D.S.G. Campilho, E.A.S. Marques, J.J.M. Machado, and L.F.M. da Silva, Geometrical optimization of adhesive joints under tensile impact loads using cohesive zone modelling. International Journal of Adhesion & Adhesives, 2020. 97: p. 102492. 36. Sane, A.U., P.M. Padole, C.M. Manjunatha, R.V. Uddanwadiker, and P. Jhunjhunwala, Mixed mode cohesive zone modelling and analysis of adhesively bonded composite T-joint under pull-out load. Journal of the Brazilian Society of Mechanical Sciences and Engineering, 2018. 40(3): p. 167. 37. Rocha, R.J.B. and R.D.S.G. Campilho, Evaluation of different modelling conditions in the cohesive zone analysis of single-lap bonded joints. The Journal of Adhesion, 2018. 94(7): p. 562-582. 38. Dimitri, R., M. Trullo, L. De Lorenzis, and G. Zavarise, Coupled cohesive zone models for mixed-mode fracture: A comparative study. Engineering Fracture Mechanics, 2015. 148: p. 145-179. 59 CHAPTER 5 Feasibility of Drone-Based Ground Penetrating Radar for Subterranean Foreign Object Detection/Mapping Celile Nur Yalçın1 Abstract This study evaluates the feasibility of using drone-based Ground Penetrating Radar (GPR) to detect and map underground foreign objects at high altitudes. Integration of GPR technology with drones enables efficient scanning of large areas. Various parameters such as operating altitude, drone speed, frequency, antenna gain, soil attenuation, field of view, target size and material type were taken into consideration. The research parameters were kept wide for this radar system, which can be customized to be integrated with different unmanned aerial vehicles. Simulations were run in MATLAB with a custom Graphical User Interface (GUI) using both pulsed and Frequency Modulated Continuous Wave (FMCW) radar modulation techniques. The results show that drone-based GPR systems can achieve sufficient Signal-toNoise Ratio (SNR) and range resolution for effective underground detection and show significant potential for archaeology, defense industry, utility mapping and environmental monitoring applications. 1 Ankara Yıldırım Beyazıt University, Institute of Science, Electrical and Electronics Engineering Master’s Degree Student, ORCID Code: https://orcid.org/0009-0009-64025083, [email protected] FEASIBILITY OF DRONE-BASED GROUND PENETRATING RADAR FOR SUBTERRANEAN FOREIGN OBJECT 60 Keywords — Drone, GPR, Underground Detection, Foreign Object, Mapping. 1.Introduction Subsurface detection holds critical significance across various disciplines such as civil engineering, archaeology, environmental studies, and security. Ground Penetrating Radar (GPR) is a non-invasive technique that employs electromagnetic waves to image subsurface features. Conventional groundbased GPR methods are often time-intensive and limited in terms of scanning area coverage. The advancement of drone technology offers a solution by enabling rapid and extensive aerial surveys. The integration of drone and GPR technologies facilitates efficient and high-resolution mapping of large areas. This study evaluates the feasibility of drone-based GPR systems operating at high altitudes and investigates key system parameters by analyzing performance metrics such as signal-to-noise ratio (SNR) and range resolution through simulation. 2. Sensing Techniques Utilized in GPR Technology Ground Penetrating Radar (GPR) is a geophysical imaging method that uses electromagnetic waves to detect and map subsurface structures. GPR systems can be optimized for different applications based on frequency and modulation techniques. The choice of these techniques is primarily influenced by the targeted depth, resolution requirements, and environmental conditions. The following sections provide an overview of the principal techniques commonly employed in GPR systems. 2.1 Pulsed GPR Pulsed GPR systems emit high-frequency electromagnetic waves in short pulses and analyze the signals reflected from subsurface structures. The depth of objects is calculated using the time delay of the return signal, the propagation velocity of electromagnetic waves in the medium, and the traveled distance eq (1)-(2). 2d t v = (1) where: ( t ) = time delay (s) CELILE NUR YALÇIN 61 ( d ) = object depth (m) ( v ) = electromagnetic wave velocity (m/s) c v r ε = (2) where: ( c ) = speed of light 𝜀𝑟 = relative permittivity of the medium Advantages: • Enables rapid data acquisition. • Provides high penetration depth. • Simple signal analysis due to direct time-domain measurement. Disadvantages: • Signal attenuation increases at higher frequencies. • Higher noise levels can occur, leading to increased data processing requirements. 2.2 Continuous Wave (CW) GPR The Continuous Wave (CW) technique transmits a continuous electromagnetic signal and evaluates the signals reflected from beneath the surface. Unlike pulsed GPR, the signal duration is constant, and distance information is obtained through modulation techniques. Advantages: • Delivers a strong and continuous signal, resulting in a high signal-to-noise ratio (SNR). • More compatible with moving platforms such as drone-based GPR systems. Disadvantages: • Depth measurement accuracy varies depending on the modulation approach. • Requires more complex data processing compared to pulsed systems. FEASIBILITY OF DRONE-BASED GROUND PENETRATING RADAR FOR SUBTERRANEAN FOREIGN OBJECT 62 2.3 Frequency Modulated Continuous Wave (FMCW) GPR The FMCW technique obtains range information by continuously varying the frequency of the transmitted signal over a specified time interval. The position and physical properties of the objects are determined by analyzing the frequency difference between the transmitted and received signals eq (3). 2 res c R B = (3) where: res R= range resolution (m) ( B ) = bandwidth (Hz) ( c ) = speed of light (m/s) SNR calculation eq (4): SNR = r n P P (4) where: r P = received power (W) n P = noise power (W) Advantages: • Provides higher resolution. • Energy-efficient due to operation with low power signals. • Enables more precise depth estimation. Disadvantages: • Higher system cost due to increased complexity of electronic components. • Data processing requires more computational resources compared to pulsed GPR. CELILE NUR YALÇIN 63 2.4 Time-Domain GPR Time-domain GPR systems detect subsurface objects by analyzing the propagation time of electromagnetic waves. This technique is capable of distinguishing media with different dielectric properties. Advantages: • Capable of high-speed data acquisition over large-scale areas. • Offers high depth accuracy and is suitable for multilayered environments. Disadvantages: • Requires larger storage capacity due to high data volume. • Analysis time may be longer compared to other techniques. 2.5 Multi-Frequency GPR Multi-frequency GPR systems transmit electromagnetic waves at multiple frequencies simultaneously, allowing for precise detection of objects at various depths. Advantages: High frequencies enable detection of surface details, while low frequencies allow penetration to greater depths. Offers a broader range of applications. Disadvantages: Complexity in data processing due to multi-frequency inputs. Hardware costs are higher relative to single-frequency systems. Conclusion and Evaluation The techniques employed in GPR technology vary based on operational principles and application domains. Pulsed GPR offers advantages such as deep penetration and straightforward data processing, whereas FMCW systems provide superior resolution. Time-domain and multi-frequency GPR systems are more suitable for extensive surveys. Selection of the appropriate technique should consider environmental conditions, required depth, desired resolution, and operational costs. This summary elaborates on fundamental GPR sensing techniques, thereby contributing to informed decision-making regarding method selection for diverse applications. TRUST-WEIGHTED SENTIMENT FILTERING AND QUERY-BASED RECOMMENDATION FOR TURKISH 70 Yildiz et al. addressed these challenges by introducing a BERT-based sentiment classification model tailored for Turkish e-commerce reviews, outperforming standard classifiers and confirming the value of languagespecific optimization17. Their findings reinforce the importance of tokenlevel handling, synonym expansion, and sentiment consistency in Turkishlanguage recommendation systems. Finally, domain-specific research has begun to emphasize aspect-based sentiment analysis (ABSA) as a pivotal technique for aligning user queries with relevant product features. Liu et al. proposed an attention-based ABSA framework for e-commerce platforms18, allowing recommendations to respond not only to general sentiment polarity but also to context-specific features like “battery life” or “screen quality,” which are especially pertinent in product review filtering . In light of these developments, the present study contributes by integrating high-precision regex-based retrieval with sentiment filtering and review reliability scoring to produce a lightweight, transparent, and user-intentaligned recommendation pipeline. This approach is not only explainable and language-sensitive but also computationally efficient—addressing the dual challenge of transparency and scalability. METHODOLOGY This study adopts a design-oriented methodological framework to develop a sentiment-aware and reliability-weighted recommendation system tailored to Turkish e-commerce data. The approach integrates rule-based sentiment labeling using star ratings, linguistic preprocessing (tokenization, lowercasing, and stopword elimination), and visual exploration via word clouds. Reliability is calculated through user engagement metrics, specifically a normalized ratio of positive (clap) to total feedback (clap + thumbsdown), reflecting recent findings that user trust signals enhance recommendation relevance and credibility⁷. Instead of using pre-trained transformer-based architectures, this work emphasizes lightweight, interpretable, and language-specific techniques, which have shown effectiveness in morphologically complex languages like Turkish⁸. A query-driven retrieval mechanism is also implemented, allowing users to search product feedback with natural phrases such as “telefon ekranı güzel”. The matching logic incorporates pattern-based keyword detection and filters the results according to sentiment class and a visual reliability score, in line with recent research emphasizing human-centered and transparent recommendation pipelines⁹⁻¹¹. MUSTAFA ERŞAHİN • UTKU GEZENSOY 71 Dataset Description The dataset utilized in this study consists of Turkish-language e-commerce product reviews, including structured metadata such as product identifiers, star ratings, review titles, full review texts, and user engagement metrics (clap and thumbsdown counts). These reviews represent real-world customer opinions from a diverse range of product categories, allowing for sentiment analysis grounded in authentic linguistic patterns and user preferences. (Table 1: Each review entry contains the following key attributes). Table 1: Each review entry contains the following key attributes To enhance interpretability and ensure consistency in sentiment annotation, a sentiment polarity label was created based on star ratings: • Reviews rated 1 or 2 were labeled as negative (-1) • Ratings of 3 were considered neutral (0) • Ratings of 4 or 5 were labeled as positive (+1) In addition to sentiment labels, a reliability score was calculated for each review. This score, normalized and converted into percentage format, serves as a proxy for user trustworthiness and plays a vital role in filtering and prioritizing reviews within the recommendation system. The dataset includes tens of thousands of reviews, making it a suitable corpus for sentiment modeling, reliability analysis, and user-query matching in Turkish-language contexts. Furthermore, the presence of both quantitative and qualitative feedback enables a hybrid approach that leverages rule-based classification alongside linguistic feature exploration. Data Preprocessing Prior to model training and recommendation generation, the raw dataset underwent a series of preprocessing steps designed to enhance data quality, reduce noise, and standardize textual features for natural language analysis. TRUST-WEIGHTED SENTIMENT FILTERING AND QUERY-BASED RECOMMENDATION FOR TURKISH 72 First, missing values were addressed. All missing entries in the title column were replaced with the placeholder “Basliksiz”, while reviews with null values in the review field were removed to prevent downstream inconsistencies. Additionally, the display settings of the data frame were adjusted to reveal complete textual content and ensure visual clarity during analysis. Next, a sentiment label was assigned to each review based on its corresponding star rating. Following established conventions in the literature, star ratings of 1 or 2 were categorized as negative (-1), 3 as neutral (0), and 4 or 5 as positive (1). This transformation enabled supervised learning models to utilize a structured target variable for classification tasks. To further prepare the text for analysis, a custom text cleaning function was implemented. This function involved lowercasing all letters, removing punctuation and special characters, and eliminating stopwords specific to the Turkish language. The stopword list was sourced from NLTK and manually extended to better suit colloquial usage patterns. The cleaned text was stored in a new column named cleaned_review. To visualize the distribution of high-frequency terms after cleaning, a word cloud was generated, offering a compact summary of dominant expressions used across product reviews (Figure 1: Most Frequently Occurring Words). Figure 1: Most Frequently Occurring Words Review Reliability Scoring To enhance the interpretability and trustworthiness of user-generated content, a reliability score was computed for each review based on MUSTAFA ERŞAHİN • UTKU GEZENSOY 73 community engagement metrics. Specifically, the clap and thumbsdown counts—representing positive and negative user reactions, respectively— were combined using a ratio-based formula: Figure 2: Reliability Score Formula The reliability score RRR, as shown in Figure 2, was designed to reflect the social trust of each review based on positive versus negative user interactions. This ratio quantifies the proportion of users who found a particular review helpful or agreeable. To ensure consistent readability, the resulting score was rounded to the nearest multiple of 5 (e.g., 73% becomes 75%), producing reliability values in increments such as 60%, 70%, or 95%. Unlike traditional sentiment analysis techniques that rely solely on reviewer input, this additional metric introduces a community-validated perspective, enriching the recommendation logic with a measure of review credibility. This layer of validation is particularly beneficial in filtering out potentially misleading or low-quality reviews, as it accounts for collective agreement or disagreement among platform users. The final reliability value is included as a separate attribute in the dataset and is displayed alongside each matched review during keyword-based query results, providing users with contextual insight into the social trust of each opinion. Search Query Matching Mechanism To enable effective retrieval of relevant product reviews, the system incorporates a lightweight, pattern-based query matching engine. Instead of relying on embedding-based or transformer-based semantic similarity models, which are computationally intensive and often language-resource dependent, this study adopts a regular expression (regex)-driven approach optimized for Turkish textual input. User queries—phrases such as “ekranı güzel” or “kaliteli kumaş”—are first preprocessed by converting all characters to lowercase and removing non-alphanumeric symbols. These cleaned queries are then tokenized into keywords. A basic keyword extractor isolates meaningful terms, which are TRUST-WEIGHTED SENTIMENT FILTERING AND QUERY-BASED RECOMMENDATION FOR TURKISH 74 matched against preprocessed review texts stored in the cleaned_review column. The review corpus is filtered using the following matching rule: a review is considered relevant only if it contains all the tokens from the user query. This Boolean AND logic ensures high-precision retrieval, emphasizing explicit semantic overlap. Additionally, a sentiment constraint can be applied (e.g., restricting to only positive sentiment reviews), as defined by the rank output label. To account for lexical variation in Turkish, regular expressions are extended to include common synonyms and morphological variants. For instance, the search token “ekran” may also match “görüntü” or “panel” based on a manually defined synonym map. This mechanism not only allows for interpretable and transparent result generation, but also provides a foundation for aspect-based search, where users can retrieve reviews that express sentiment toward specific product features rather than general satisfaction. An example of successful query execution is illustrated in Figure 3, where the search for “kaliteli kumaş” returns 152 relevant reviews ranked by reliability. In contrast, Figure 4 demonstrates the system’s graceful handling of unmatched queries—such as “uzay teknolojisi”—by providing clear feedback and returning an empty results table instead of failing silently or producing unrelated output. Figure 3: Example Query 1 MUSTAFA ERŞAHİN • UTKU GEZENSOY 75 Figure 4: Example Query 2 Recommendation Output and Ranking Logic Following the query-based retrieval of relevant reviews, the system performs an additional ranking step to guide users toward the most trustworthy and representative content. This ranking is primarily governed by the reliability score, which quantifies the collective user agreement on the helpfulness of a review based on positive (clap) and negative (thumbsdown) feedback. Reviews with higher reliability are prioritized to appear earlier in the output list, enhancing the transparency and quality of recommendations. Beyond mere presentation, a secondary optional product recommendation mechanism is triggered when the user’s query includes at least one term from a predefined product vocabulary (e.g., “telefon”, “pantolon”, “tablet”). If such a term is detected, the system interprets the query not only as a request for feature-aligned reviews, but also as a preference for discovering a relevant product. In such cases, the system identifies products associated with high-reliability reviews (≥60%) and returns one top product per matching item, using groupby logic to prevent repetition. This process mirrors recent developments in explainable recommender systems, where user queries are mapped to opinion-enriched items through transparent rules instead of latent factor embeddings. The final output includes: • Reviews filtered by query keyword(s) and sentiment class • Ordered results based on review reliability • A distinct recommendation list (if applicable), where one high-confidence review per product is displayed as a suggestion This dual-structured architecture—combining exact-match review retrieval with aggregated, trust-based recommendation output—enables users to receive contextual insights while promoting high-quality user feedback. Figure 5 shows the system response for the query “ekranı iyi”, where the input expresses a feature preference but lacks a specific product class. As a TRUST-WEIGHTED SENTIMENT FILTERING AND QUERY-BASED RECOMMENDATION FOR TURKISH 76 result, only relevant reviews are displayed, and no product recommendation is made. In contrast, Figure 6 presents the query “ekranı iyi telefon”, where the inclusion of a known product term (“telefon”) triggers the recommendation logic. The system not only retrieves relevant reviews but also suggests specific phones that meet both the sentiment and reliability thresholds, demonstrating the adaptability of the hybrid architecture. Figure 5: Example Query 3 MUSTAFA ERŞAHİN • UTKU GEZENSOY 77 Figure 6: Example Query 4 CONCLUSION AND DISCUSSION This study successfully developed a sentiment analysis–based recommendation approach tailored to Turkish-language e-commerce reviews. The preprocessing pipeline included stopword removal, normalization, and token-level cleaning to ensure linguistic consistency. A sentiment label was then assigned to each review based on star ratings, enabling sentiment classification into negative, neutral, and positive classes. In addition to textual sentiment indicators, a reliability score was introduced, calculated from the ratio of positive to total user feedback (clap vs. thumbsdown), thereby incorporating community consensus into the recommendation logic. This trust metric provided an additional dimension for filtering and interpreting user reviews, particularly when presenting suggestions for user-entered queries. The core retrieval system allowed users to enter aspect-based search terms such as “ekranı iyi telefon” (“a phone with a good screen”), upon which the algorithm filtered and listed reviews that contained both the relevant feature (e.g., “ekran”) and a valid product class (e.g., “telefon”), ranking them based on sentiment and reliability. TRUST-WEIGHTED SENTIMENT FILTERING AND QUERY-BASED RECOMMENDATION FOR TURKISH 78 The hybrid use of regular expression–based pattern matching and reliability-weighted display logic enabled the system to mimic a lightweight recommendation engine without relying on traditional collaborative filtering or matrix factorization methods. Despite the lack of a machine learning classifier at this stage, the proposed model effectively handled aspect-based sentiment retrieval in Turkish, which is particularly valuable given the scarcity of high-quality, labeled NLP resources for the language. In conclusion, the proposed architecture demonstrates a functional and scalable framework for extracting relevant, sentiment-rich, and trustworthy product suggestions from free-form Turkish reviews. It lays foundational groundwork for future integration of more advanced NLP models, such as transformers or supervised classifiers, while offering immediate value in low-resource recommendation scenarios. he findings of this study provide meaningful insights into the practical deployment of sentiment-based recommendation systems in low-resource languages such as Turkish. By combining pattern-based query logic with sentiment classification and user-based reliability scoring, the proposed system bridges the gap between computational efficiency and interpretability— two often conflicting goals in modern natural language processing (NLP) applications. Unlike neural network-based models that demand large annotated datasets and high computational resources, our approach relies on simple rule-based techniques—such as Boolean keyword matching and sentiment labeling via star ratings—that are both scalable and transparent. This strategy aligns with recent calls in the literature advocating for low-complexity, explainable AI methods in real-world deployment scenarios, particularly in under-resourced language contexts . The use of a derived reliability score based on user engagement (claps vs. thumbsdown) introduces a novel metric for quality control in recommendation systems. Traditional collaborative filtering and matrix factorization approaches often lack such direct quality signals and depend heavily on latent user-item interactions . In contrast, our framework directly incorporates perceived usefulness into ranking logic, enhancing user trust and content relevance. Moreover, the integration of aspect-based filtering— allowing users to search for feature-specific expressions like “ekranı iyi” (good screen)—demonstrates the feasibility of deploying fine-grained sentiment analysis without the need for dependency parsing or pretrained transformers. This is particularly relevant in languages like Turkish, where morphological richness can hinder generalization in black-box models . MUSTAFA ERŞAHİN • UTKU GEZENSOY 79 Despite its simplicity, the system also supports product-level recommendations by leveraging a hybrid of content-based filtering and ruledriven heuristics. This layered recommendation architecture is inspired by recent developments in interpretable recommendation frameworks . Notably, the dual-output structure—retrieved reviews and aggregated product suggestions—enhances the system’s practicality in e-commerce scenarios. However, the approach is not without limitations. The reliance on keyword overlap may restrict recall in cases of paraphrased or implicit sentiment expressions. Furthermore, the handcrafted synonym maps, while effective, are not exhaustive and may introduce bias or omission errors. Future iterations of the system could explore lightweight word embedding models (e.g., FastText) to complement the rule-based engine without sacrificing interpretability . In conclusion, this study demonstrates that effective sentiment-based recommendation is achievable in Turkish using rule-based NLP pipelines augmented by user feedback mechanisms. It contributes to the ongoing discourse on building lightweight, transparent, and culturally adaptive AI systems for recommendation in low-resource environments. FUTURE WORK While the current study demonstrates the viability of rule-based sentimentdriven recommendation systems in Turkish-language e-commerce contexts, several avenues remain open for further enhancement. First and foremost, the inclusion of embedding-based semantic search methods—such as those utilizing FastText or BERTurk—could mitigate the rigidity of keyword-based query matching. These models are capable of capturing lexical and semantic similarity even when user input does not exactly match review phrasing, which could significantly improve both recall and user experience in free-text queries . Second, expanding the synonym mapping strategy to a data-driven, corpus-based synonym discovery method (e.g., PMI or co-occurrence statistics) would allow for greater adaptability across product domains. Given Turkish’s agglutinative morphology, statistical expansion techniques or morphological analyzers like Zemberek could be employed to more accurately detect stemmed and derived forms . Furthermore, the reliability score currently uses only binary feedback (clap/thumbsdown). Future systems could integrate temporal data (e.g., time-weighted feedback), reviewer metadata (e.g., verified purchaser status), or behavioral engagement metrics