Li L., Yang J., Bai X., Huang Q., Giunta G., Belouettar S., Li H., Hu H.
Composite Structures, vol. 377, art. no. 119900, 2026
Often, a failure prediction relies on criteria that evaluate current deformation state satisfying both conservation law and material law. To achieve this, two challenges arise: (1) it is difficult to obtain the material law when the constitutive behavior is complex, especially near or beyond the critical state, and (2) it is difficult to establish the failure criterion that performs well in the case of new failure modes. Based on data-driven computational mechanics, a criterion-free and model-free framework is proposed to predict the failure of composites. Data-driven computational mechanics minimizes the distance function between material data and admissible state satisfying the conservation law. After failure occurs, the admissible state is beyond the material database regime, leading to a high value of distance function. Thus, the proposed framework predicts failure by significant increase in value of the distance function. The failure predicting results based on the proposed framework are compared with those of other failure criteria. Moreover, the test data from the First World-Wide Failure Exercise are considered to evaluate the proposed framework. Different from classical failure prediction where the material models and failure criteria are calibrated based on test data, the proposed framework can predict failure directly with test data.
