Prediction of mechanical properties of glass-epoxy composites using supervised learning

Traditional testing methods for evaluating the mechanical properties of composite laminates, such as Interlaminar Shear Strength (ILSS) and In-Plane Shear Strength, are known to be resource intensive, time consuming, and expensive. This often leads to setbacks and failures in the development process. In this study, the development of a predictive model was proposed to estimate these key mechanical properties based on specific test measurements and dimensions. The glass-epoxy family of composites has been mainly focused on because it is widely used in various industries. To explore the feasibility of this approach, supervised machine learning techniques were employed, which offer an efficient means of creating predictive models based on test characteristics. The selected tests for consideration include the Short Beam Strength Test (SBS Test) and Iosipescu or V-notch test for ILSS and in-plane shear strength, respectively. The performance of different machine learning algorithms such as decision tree, multiple linear regression, ridge regression, and artificial neural networks was evaluated to identify the most suitable model for the dataset. Given the limited availability of data, the study emphasizes the importance of achieving good performance even with small datasets. The findings from this research hold promise for streamlining the testing process and improving the efficiency of composite material development. The study highlights the effectiveness of the decision tree model in predicting mechanical properties of glass-epoxy composites, particularly for small datasets with nonlinear relationships. It shows that the decision tree outperforms other models, including multiple linear regression and ridge regression, in capturing complex patterns. The analysis also identifies key influential factors, such as the span-to-depth ratio for interlaminar shear strength and specimen thickness for in-plane shear strength, providing valuable insights for future composite material design and testing. The novelty of this study lies in its emphasis on feature importance analysis to predict the mechanical properties of glass-epoxy composites. By examining how specimen dimensions influence Interlaminar Shear Strength and In-Plane Shear Strength, the work provides valuable insights into the key factors affecting composite behavior. This approach not only highlights the most significant dimensions but also offers a practical method for predicting composite properties using limited datasets, paving the way for optimized material design and future research in composite material prediction.