Çinko alüminyum alaşımları, yüksek mukavemet, iyi dökülebilirlik ve üstün aşınma direnci gibi mekanik özellikleri nedeniyle endüstride önemli bir rol oynamaktadır. Özellikle ZA-27 alaşımı, bu özellikleriyle pirinç, dökme demir ve bronza kıyasla daha üstün performans göstermekte ve torna tezgâhı, su pompaları, çim biçme makineleri, tekstil ve otomotiv sanayisi gibi çeşitli alanlarda yaygın olarak kullanılmaktadır. Son yıllarda ZA-27 alaşımının aşınma direncini daha da artırmak amacıyla yapılan çalışmalar büyük bir ivme kazanmıştır. Bu çalışmada, ZA-27 alaşımına SiC ve Grafit partikülleri eklenerek üretilen hibrit kompozit malzemelerin aşınma direnci performanslarının tahmin edilmesi amaçlanmıştır. Aşınma direncinin iyileştirilmesi için yapılan deneyler maliyet, zaman ve iş gücü açısından yüksek gereksinimler doğurduğundan, bu çalışmada makine öğrenmesi yöntemleri kullanılarak deney sayısının minimize edilmesi ve ara değerler hakkında tahminlerde bulunulması hedeflenmiştir. Çalışmada, 240 adet veri seti %80 eğitim ve %20 test verisi olarak ayrılmış, Python ve MATLAB platformlarında çeşitli makine öğrenmesi modelleri uygulanmıştır. Bu modeller arasında Doğrusal Regresyon (LR), Karar Ağacı Regresyonu (DTR), Rasgele Orman Regresyonu (RFR), Destek Vektör Regresyonu (SVR), Yapay Sinir Ağları (ANN), Uyarlamalı Ağ Tabanlı Bulanık Çıkarım Sistemi (ANFIS) ve Gauss Süreç Regresyonu (GPR) yer almaktadır. Modellerin performansları, Ortalama Kare Hatası (MSE), Hata Kareler Ortalamasının Karekökü (RMSE), Ortalama Mutlak Hata (MAE) ve Mutlak Değişim Yüzdesi (R²) istatistikleri kullanılarak değerlendirilmiştir. Elde edilen sonuçlar, ANFIS modelinin diğer modellere kıyasla üstün performans gösterdiğini ve en düşük hata değerleri ile en yüksek doğruluk oranlarına sahip olduğunu net bir şekilde ortaya koymuştur. ANFIS modelinin ortalama kare hata (MSE) değeri 0.0428 olarak hesaplanmış olup, bu değer, değerlendirmeye alınan diğer modellerden anlamlı derecede düşüktür. Ayrıca, R² değeri açısından 0.9291 gibi yüksek bir başarıya ulaşarak, ANFIS modelinin verilerin açıklanabilirliğinde ve tahmin doğruluğunda ne kadar güçlü bir araç olduğunu göstermiştir. Bu modelin ardından Yapay Sinir Ağları (ANN) ve Gauss Süreç Regresyonu (GPR) modelleri de yüksek doğruluk oranları ile öne çıkmıştır. ANN modelinin MSE değeri 0.0516 ve R² değeri 0.9168 olarak bulunurken, GPR modelinin MSE değeri 0.0652 ve R² değeri 0.9024 olarak hesaplanmıştır. Bu sonuçlar, SiC ve Grafit partikülü takviyeli ZA-27 hibrit kompozit malzemelerinin aşınma direncini artırmadaki potansiyelini ve makine öğrenmesi yöntemlerinin bu tür malzeme performansı tahminlerinde etkin bir araç olduğunu ortaya koymaktadır. Sonuç olarak, bu çalışma, ZA-27 alaşımının aşınma direncini artırmada yapay zekâ ve makine öğrenmesi yöntemlerinin etkinliğini vurgulamakla kalmamış, aynı zamanda endüstriyel uygulamalarda deney maliyetlerini düşürmenin de mümkün olduğunu göstermiştir. Çalışma, gelecekte daha geniş veri setleri, farklı alaşım bileşimleri ve daha karmaşık modelleme yaklaşımları ile gerçekleştirilecek araştırmalar için önemli bir referans noktası oluşturmaktadır. Özellikle, makine öğrenmesi tabanlı yöntemlerin bu tür mühendislik problemlerinde yenilikçi ve uygulanabilir çözümler sunduğu bir kez daha kanıtlanmıştır.
ZA-27 alloy is highly regarded for its exceptional mechanical properties, impressive wear resistance, and excellent castability, making it a standout material in zinc-aluminum alloys. Compared to traditional materials such as brass, cast iron, and bronze, ZA-27 alloy provides excellent durability and reduced weight. This advantageous profile renders it an appealing alternative for numerous industrial applications. Notably, the alloy's outstanding wear resistance, particularly in comparison to brass alloys widely employed as sliding-bearing materials, highlights its relevance in various engineering fields. As a result, ZA-27 alloy has found widespread applications in diverse sectors such as lathe machines, water pumps, lawnmowers, textile machinery, and the automotive industry. Despite its many advantages, industries worldwide face significant economic challenges arising from wear-related issues. These problems lead to substantial maintenance, replacement, and downtime costs. Consequently, there is an urgent need to develop materials with enhanced wear resistance to address these challenges. This necessity drives ongoing efforts in materials engineering to design and produce advanced materials that can reduce wear-related losses and improve the longevity and performance of industrial components. These efforts often involve labor-intensive, costly, and time-consuming experimental studies aimed at optimizing the properties of materials. In response to these challenges, hybrid composite materials reinforced with silicon carbide (SiC) and graphite particles have been developed to improve the mechanical properties and wear resistance of ZA-27 alloy. SiC particles are renowned for their high hardness and ability to enhance wear resistance, while graphite particles contribute to improving the sliding properties of the material due to their intrinsic lubricating characteristics. The synergistic effects of these reinforcements result in hybrid composites with superior performance under demanding conditions. The squeeze casting method was employed to fabricate these composites, ensuring the uniform distribution of reinforcement particles and the creation of a homogeneous microstructure. This process involved mixing SiC and graphite particles into the molten ZA-27 alloy using a vortex technique, followed by solidification under a pressure of 50 MPa. This approach optimized the mechanical properties of the resulting material and ensured consistent performance across varying operating conditions. Although experimental methods have traditionally been the primary approach for evaluating the properties of materials, the high costs and extended timeframes associated with such studies have increasingly prompted researchers to explore alternative methodologies. Artificial intelligence (AI) and machine learning (ML) techniques have emerged as powerful tools for addressing these challenges. These computational methods have the potential to significantly streamline research processes by enabling accurate predictions of material behavior with reduced reliance on extensive experimental trials. Machine learning algorithms, in particular, are adept at analyzing complex datasets to uncover patterns and relationships that may not be readily apparent through traditional approaches. Their ability to handle multifaceted problems makes them invaluable in advancing materials science. This study employed AI and ML techniques to predict the volumetric wear rates of hybrid composite materials under various loading and sliding speed conditions. The primary objective was to determine the wear rate as a function of four independent variables: SiC ratio, graphite ratio, applied load, and sliding speed. A comprehensive dataset comprising 240 experimental results was compiled to achieve this goal. This dataset was divided into training and testing subsets, with 80% of the data allocated for training and 20% reserved for testing. The training dataset facilitated the development and optimization of predictive models, while the testing dataset provided an independent evaluation of the models' generalization capabilities, ensuring their reliability in predicting wear rates for unseen data. The study implemented various machine-learning algorithms using Python and MATLAB software environments. Python-based analyses focused on algorithms such as Linear Regression (LR), Decision Tree Regression (DTR), Random Forest Regression (RFR), and Support Vector Regression (SVR). These models were selected for their ability to handle simple and complex relationships between the variables. In parallel, MATLAB was employed to explore more advanced AI models, including Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), and Gaussian Process Regression (GPR). Each of these models offers unique advantages and limitations depending on the complexity of the data and the nature of the relationships among the input variables. To assess the performance of the models, several statistical metrics were employed, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the Coefficient of Determination (R²). These metrics comprehensively evaluated each model's accuracy and error rates, enabling a robust comparison of their predictive capabilities. Among the evaluated models, the ANFIS model demonstrated the highest predictive accuracy, achieving an MSE value of 0.0428 and an R² value of 0.9291. These results highlight ANFIS's exceptional ability to model the complex relationships between the independent variables and the volumetric wear rate, making it the most effective predictive tool in this study. Following ANFIS, the ANN and GPR models also delivered strong performances. The ANN model achieved an MSE value of 0.0516 and an R² value of 0.9168, indicating its robustness in handling nonlinear relationships within the dataset. Similarly, the GPR model yielded an MSE value of 0.0652 and an R² value of 0.9024, confirming its suitability for material property prediction tasks. By contrast, simpler algorithms such as LR and DTR exhibited comparatively lower accuracy, underscoring the limitations of these methods in capturing the intricate interactions between the variables. These results emphasize the importance of employing advanced models for complex and nonlinear datasets tasks. The findings of this study underscore the significant benefits of hybrid composites reinforced with SiC and graphite particles in enhancing the wear resistance of ZA-27 alloy. These reinforcements not only improved the mechanical and tribological properties of the alloy but also highlighted its potential for use in demanding industrial applications. Moreover, the study demonstrated the transformative potential of AI and ML techniques in materials engineering. These computational tools offer a means to address complex challenges with remarkable efficiency, providing accurate predictions that reduce the need for extensive experimental testing. Among the various models evaluated, the ANFIS model emerged as a potent tool, delivering unparalleled predictive accuracy and minimal error rates. Its performance highlights the value of hybrid approaches that combine the strengths of neural networks and fuzzy logic systems. By effectively capturing the nonlinear relationships and uncertainties inherent in the dataset, ANFIS proved an indispensable asset in this study. The success of ANFIS and other advanced models underscores the broader applicability of AI methodologies in optimizing industrial processes and advancing materials science research. The implications of this study extend beyond its immediate findings, offering valuable insights for future research and practical applications. One potential avenue for further investigation involves the incorporation of more extensive and more diverse datasets. Expanding the dataset could enhance the generalizability of the predictive models and provide a deeper understanding of the factors influencing material properties. Additionally, exploring alternative material compositions and introducing new independent variables could uncover further opportunities for optimization. For instance, parameters such as temperature, environmental conditions, or additional reinforcement materials could provide a more comprehensive evaluation of the material's performance. Another promising direction involves the adaptation and comparison of other AI techniques from the literature. Methods such as deep learning, ensemble learning, or evolutionary algorithms could be explored to identify their relative strengths and weaknesses in addressing similar problems. By comparing the performance of these methods, researchers can gain valuable insights into the most effective approaches for specific applications. Moreover, integrating AI techniques with traditional experimental methods could pave the way for hybrid methodologies that combine computational tools' precision with empirical data's reliability. This study highlights the critical role of interdisciplinary approaches in advancing materials engineering. By integrating AI and ML techniques with experimental research, this work demonstrates the potential to develop more efficient and sustainable engineering solutions. The findings emphasize the importance of continued investment in computational tools and their application to complex engineering problems. Such efforts are essential to meet the evolving demands of modern industries and drive technological innovation. In conclusion, this study highlights the effectiveness of hybrid composites and the potential of artificial intelligence (AI) and machine learning (ML) techniques in materials engineering. The findings validate the effectiveness of advanced computational methods in predicting material properties while demonstrating their broader applicability in optimizing industrial processes and enhancing overall efficiency. These advancements underscore the value of addressing complex engineering problems by reducing costs while promoting sustainability. Based on the findings of this research, future studies can further enhance the understanding and performance of advanced materials and contribute to developing innovative solutions to the challenges faced by modern industries. Furthermore, integrating these methods with larger datasets and real-time systems can support the development of more durable and adaptive material designs for various industrial applications.
Except where otherwise noted, this item's license is described as http://creativecommons.org/licenses/by/4.0/
