Discover failure criteria of composites from experimental data by sparse regression

Abstract

Over the past few years, a few experimental failure data of composites have been collected. It would be of interest to leverage the existing data to improve the prediction of failure criteria. In this paper, we developed a framework that combines sparse regression with compressed sensing to discover failure criteria of composites from experimental data, which leveraging advances in sparsity techniques and machine learning. This framework does not need Bigdata to train the model and is remarkably robust to the noised data, which satisfies the constraints of the current failure data. To test the performance of the proposed method, we collected the experimental data from the first Worldwide Failure Exercise (WWFE I) and fed it to the proposed method. To satisfy the engineering design needs, we proposed an optimization approach to enforce a constraint to the discovered failure criterion to yield a conservative model. Three examples were presented to demonstrate the proposed framework. The result shows that the proposed framework can identify the most important features and the discovered failure criterion match the experiment result well. Besides, with the enforced constraint, the proposed method can yield a conservative failure criterion.