Virtual modelling integrated phase field method for dynamic fracture analysis

Abstract

A non-deterministic phase field (PF) virtual modelling framework is proposed for three-dimensional dynamic brittle fracture. The developed framework is based on experimental observations, accurate numerical modelling, and virtually foreseeable dynamic fracture prediction module through the machine learning algorithm. The uncertain system inputs, including variabilities of material properties, are incorporated into dynamic fracture analysis. Phase field method is implemented to simulate the dynamic fracture behaviours of 3D cracked structures with variabilities and then to create the training database for virtual damage model. The virtual damage model omits the physical finite element approximation process and reveals the virtual governing relationship between the variational system inputs and fracture responses. This advantage enables the virtual model to provide reliable crack propagation prediction based on either experiment-based or numerical simulation and greatly improves the computational efficiency of dynamic fracture analysis. To establish the accurate virtual model in the training process, a newly developed extended support vector regression (X-SVR) method with T-spline polynomial kernel functions is adopted for its outstanding performance in handling complex high-dimensional problems. Based on different real-world engineering scenarios, multiple fracture failure criteria, both strength-based and serviceability-based, are selected and demonstrated in numerical investigations to visualise the proposed framework's workflow. The effectiveness as well as accuracy are verified by these examples, and it is observed that the computational efficiency of dynamic fracture analysis is greatly improved. With the proposed framework, a continuously updating dynamic fracture surveillance system can be potentially built for practical applications.