Discover Composites Failure Criterion by Machine Learning

A reliable design of a composite structure needs to consider the failure of the composites. Hashin failure criterion is one of the most popular phenomenological models in engineering practice due to its simplicity of application. Although remarkable success has been achieved from the Hashin failure criterion, it does not always fit the experimental results very well. Over the past few years, a few experimental failure data have been collected. It would be of interest to leverage the existing data to improve the prediction of failure criteria. In this paper, we proposed to apply a framework that combines sparse regression with compressed sensing to discover failure criteria from data. Following the phenomenological failure models, we divided the failure of composites into tensile and compressive fiber modes, tensile and compressive matrix modes. Two examples were studied with the proposed framework. The first example was presented to demonstrate the capability of the framework. The data was generated by the Hashin failure criterion and added various magnitudes of noise. The proposed framework was implemented to discover the failure criterion from the noised data. For the second example, the proposed method was used to discover failure criteria from the experimental data which are collected from the first world wide failure exercise (WWFE I). Both examples show that the proposed method can discover the failure criteria accurately. References: Tao, F.; Liu, X.; Du, H.; and Yu, W.: “Discovering Failure Criteria of Composites by Sparse Regression and Compressed Sensing,” Proceedings of the American Society for Composites 36th Technical Conference, Virtual Conference, Sept. 19-23, 2021.