Machine Learning Application to Predict Production Performance of Carbon Fiber Reinforced Polymer Composites

Abstract

Carbon fiber reinforced polymer (CFRP) composites are becoming more widespread due to their excellent mechanical properties. Recently, many researchers in different disciplines have been studying machine learning implementation to efficiently assess composite materials' mechanical properties. In this sense, this research aims to apply machine learning approaches via building regression models that predict mechanical properties of CFRP composites, including flexural strength, flexural modulus, tensile strength, mode II energy-release rate of CFRP, and the fractal dimension-capacity slope. The input variables used to build the model are processing characteristics such as reinforcements volume fraction, interlayer volume fraction, glass transition temperature, and pressure during manufacturing. Different linear regression models have been used and compared: multiple linear, ridge, and lasso regression models. However, ridge regression provides better results than lasso regression. Among the five dependent variables above, flexural strength models result in a higher R2 value and lower MSE.

Date of Award	Dec 2022
Original language	American English
Supervisor	Maher Maalouf (Supervisor)