Information Fusion and Machine Learning for Sensitivity Analysis Using Physics Knowledge and Experimental Data

Summary

This paper investigates physics-informed machine learning strategies for global sensitivity analysis (GSA) when both a physics-based model and experimental observations are available. Two machine learning techniques, deep neural networks (DNN) and Gaussian process (GP) modeling, are considered, and two strategies for incorporating physics knowledge are explored: enforcing physics constraints via loss functions, and pre-training and updating the model with simulation and experimental data. Four different models are built for each type, and uncertainties are included in Sobol indices computation. The DNN-based models yield smaller bounds on sensitivity estimates compared to GP-based models. The methods are illustrated with additive manufacturing and lake temperature modeling examples.