Adaptive fourth-order phase field analysis using deep energy minimization
By Somdatta Goswami, Cosmin Anitescu, Timon Rabczuk. 2020
Abstract Phase field modeling of fracture is computationally expensive as it demands a very fine mesh to resolve the damage region. Hence, the practical application of such models are severely limited. Local refinement techniques are often necessary. In our recent work (Goswami et al., 2019), for solving brittle fracture problems using physics informed neural network (PINN), the crack path is resolved by minimizing the variational energy of the system. However, in Goswami et al. (2019) we used a pre-refined domain based on prior information of the failure path, which is not always available. In this work, we propose an adaptive h -refinement scheme to locally refine the domain along the path of the growth of the crack. The phase field parameter, ϕ and a residual-based posteriori error estimator are the proposed convenient measures to determine the need for refinement. For ϕ , a critical threshold is chosen such that it is lower than the value at which crack nucleation occurs and the fracture region is easily identified. On the other hand, for the residual-based error estimator, elements contributing to the highest error are marked for refinement. The proposed algorithm takes as input the geometry described using NURBS patches. For the evaluation of the basis functions, we develop a procedure based on the Bezier representation and integrate it with the adaptive refinement formulation. The results obtained using the adaptive refinement integrated variational energy based PINN approach is validated with the available analytical solution for several examples from the literature. The proposed approach is implemented on several two and three-dimensional examples to illustrate the effectiveness of the formulation. Code and data necessary for replicating the results of the examples in the article will be made available through a GitHub repository.
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
Citations in the corpus
- Physics informed deep learning for computational elastodynamics without labeled data (2020, background): “It is important to note that a few recent attempts show the promise and power of PINN for addressing computational mechanics relevant challenges such as solving mechanical problems [39, 46], modeling fracture in materials [40, 47], and detecting cracks via ultrasound nondestructive testing [48].”
- A physics-informed variational DeepONet for predicting the crack path in brittle materials (2021, methodology): “To efficiently approximate the solution of PDEs with discontinuity, variational energy-based PINN (VE-PINN) was proposed in [27, 28], where the network is trained by minimizing the variational energy of the system (defined using the weak formulation).”