Research on Three-Dimensional Failure Envelope Surface of Composite Materials Based on Data-driven

Data-driven methods based on machine learning have been successfully used in composite material design, structural analysis, data processing, material performance characterization, prediction and testing, and other fields. There is no theory in the field of composite material applications that can provide high-precision failure analysis and prediction results in all situations. This paper designed a biaxial compression-compression test of composite materials to obtain multi-variable data such as DIC digital speckle data, testing machine load-displacement data, and strain data. For the multidimensional data generated by the biaxial test, a fusion algorithm based on machine learning was used to construct a three-dimensional failure envelope of the composite material. The results show that the data mining methods can identify correlations in failure data and express them in the form of three-dimensional stress function equations. Data-driven methods based on physical information can further improve the accuracy of composite material failure behavior prediction and provide a technical basis for failure analysis and strength prediction of composite material structures.