Logo image
Equation Informed Neural Networks with Bayesian Inference Improvement for the Coefficient Extraction of the Empirical Formulas
Conference paper

Equation Informed Neural Networks with Bayesian Inference Improvement for the Coefficient Extraction of the Empirical Formulas

Cadmus Yuan, Jing Yu Wang, Cheng En Lee and Kuo-Ning Chiang
2023 24th International Conference on Thermal, Mechanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsystems, EuroSimE 2023
2023

Abstract

Bayesian inference Coefficient extraction Empirical functions Equation-informed neural networks Monte Carlo Markov Chain. Fluid Flow and Transfer Processes Computer Science Applications Electrical and Electronic Engineering Mechanical Engineering Safety Risk Reliability and Quality Modeling and Simulation
Equation-informed neural networks (EINS) method provides an efficient way for the coefficient extraction of the complicated empirical function from large observation data. The Bayesian inference is applied to improve the extraction from a statistical point of view. The empirical material model with multiple coefficients is first converted into a neural network representation, and the aforementioned coefficients are assigned as the weightings of the neural network. The training of this neural network provides a continuous improvement of the weighting.This paper presents the framework of EINS and Bayesian inference. The coefficient extraction of the Chaboche function for the material plasticity of SAC 305 is used as the test scenario for EINS, to investigate its applicability. Moreover, Bayesian inference is realized by the Monte Carlo Markov Chain method (MCMC) with Gibbs sampling to validate the robustness of the coefficients obtained by EINS.

Metrics

1 Record Views

Details

Logo image