A machine learning aided non-deterministic damage prediction framework against both 2D and 3D fracture problems is presented in this paper. By introducing a newly developed extended support vector regression (X-SVR) with generalized Dirichlet feature mapping into the phase field crack growth model, a damage assessment method that contains both crack diagnosis and prognosis is designed. Within the proposed analysis framework, the intricate fracture mechanism of practical engineering system can be learnt by the X-SVR model so a continuous damage diagnosis-prognosis loop can be established to assess the latest working condition of the structure. The proposed framework is applicable not only for quantifying and then assessing the current working conditions, but also for predicting the potentially crack propagation against the future forecasted information. Compared with the established experimental records and numerical result, the accuracy, effectiveness, and computational efficiency of the proposed framework are fully verified.

本文提出了一种针对 2D 和 3D 断裂问题的机器学习辅助的非确定性损伤预测框架。通过将新开发的具有广义狄利克雷特征映射的扩展支持向量回归（X-SVR）引入相场裂纹扩展模型，设计了一种包含裂纹诊断和预测的损伤评估方法。在提出的分析框架内，实际工程系统的复杂断裂机制可以通过 X-SVR 模型学习，从而可以建立连续的损伤诊断 - 预测回路来评估结构的最新工作状态。提议的框架不仅适用于量化和评估当前的工作条件，也用于根据未来的预测信息预测潜在的裂纹扩展。与已建立的实验记录和数值结果进行比较，充分验证了所提出框架的准确性、有效性和计算效率。