Physics-based neural network as constitutive law for finite element analysis of sintering

• DOI: 10.1016/j.ceramint.2024.02.333
• Publication Date: 2024-02-27T07:25:54Z
• Authors: Ran He; Venkat Ghantasala; Peter Polak; Baber Saleem; Jingzhe Pan
• Keywords: Distortion (music)

In the sintering of advanced ceramics, a digital twin consisted finite element model can predict final shape and microstructure sintered ceramic. However, to minimise uncertainties, it is essential continually update mechanical properties in using data collected from manufacturing process. One promising approach for achieving this through artificial neural networks (ANN). This study introduces machine learning strategy constitutive behaviour ceramics analysis deformation. A major challenge implementing material processing huge amount required by training validation an ANN, which are often unavailable or incomplete real demonstrates that requirement be reduced employing two-step technique. Firstly, ANN trained nonlinear law, describes general relationship between strain rates stresses. Subsequently, weights bias transferred retraining limited experimental actual It shown such successfully capture fine-grained alumina without demanding large data. case provided, highlighting feasibility commercial package, replace law shrinkage distortion part. particular, dumb-bell part, simulated retrained showed grain size relative density markedly different those law. important note proposed methodology generic used create ANNs laws twins wide range other engineering processes.