Prediction and optimization of tensile strength and damages in fiber reinforced epoxy composites by machine learning models

The constant prediction of the tensile strength which allows predicting the modes of failure in the fiber-reinforced epoxy composites is of fundamental importance in the engineering. But it is very hard to realize due to the complexity of the mutual interactions between material and architectural aspects. This paper attempts to hone a gainable machine learning model to simultaneously foresee the tensile strength and categorize the main type of failure of hybrid composites like carbon, Kevlar and glass fibres in different layered forms. Several regression and classification models were optimized and built using an experimental dataset. The findings reveal Random Forest algorithm as a powerful predictive model having sound coefficient of determination value and small root mean squared error value for predicting strength while ideal accuracy found for predicting the type of failure. The feature importance analysis indicated thickness and lay-up configuration as influential input parameters for determining the strength and failure type. The study provides an insightful, predictable, interpretable predictive modelling to unravel the complexity of structure-property relationships of hybrid composites that has a potential of allowing such patterns to advance the design and analysis of material without necessarily doing costly, expansive experimental campaigns in aerospace and automobile industry.