Hybrid physics-based and data-driven models for smart manufacturing: Modelling, simulation, and explainability

Abstract

To overcome the limitations associated with purely physics-based and data-driven modeling methods, hybrid, physics-based data-driven models have been developed, with improved model transparency, interpretability, and analytic capabilities at reduced computational cost. This paper reviews the state-of-the-art of hybrid physics-based data-driven models towards realizing a higher degree of autonomous and error-free operation in smart manufacturing. Recognizing the complementary strengths of pure physics-based and data-driven models, hybrid physics-based data-driven models are categorized as consisting of three types: (1) physics-informed machine learning, (2) machine learning-assisted simulation, and (3) explainable artificial intelligence. The principles and characteristics of these three types of hybrid physics-based data-driven models are summarized to address three aspects of smart manufacturing: product design, operation and maintenance, and intelligent decision making. Finally, the prospective directions and challenges of hybrid physics-based data-driven models are discussed from the perspective of data, scientific insights, interpretability of hyperparameters, and trading off between accuracy and explainability.