Explainable artificial intelligence framework for FRP composites design

Abstract

Fiber-reinforced polymer (FRP) materials are integral to various industries, from automotive and aerospace to infrastructure and construction. While FRP composite design guidelines have been established, the process of obtaining the desired strength of an FRP composite demands considerable time and resources. Despite recent advancements in Machine Learning (ML) models which are commonly used as predictive models, the inherent 'black box' nature of those models poses challenges in understanding the relationship between input design parameters and output strength of the composite. Moreover, these models do not provide tools to facilitate the designing process of the composite. The current study introduces an explainable Artificial Intelligence (XAI) framework that will provide understanding for the input–output relationships of the model through SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDPs). In addition, the framework provides for the first time a designing approach for adjusting the important design parameters to obtain the desired composite strength by the designer through utilizing an explainability technique called Counterfactual (CF). The framework is evaluated through the design of a 14-ply composite, successfully identifying critical design parameters, and specifying necessary adjustments to meet strength requirements.