Physics-Constrained Machine Learning for Reliability-Based Design Optimization

To aid and improve the reliability of product designs, repeated safety tests are required to find out the safety performance of the product with respect to design variables. A large number of design variables involved in the performance evaluations often leads to enormous testing costs. A method that can effectively utilize partially available information from multiple sources of varying dimensions and fidelity is a pressing need for reliability-based product design. Moreover, in the product design and safety estimation process, it is beneficial to take into account the manufacturing policies and physical principles. Therefore, it is desirable to have a framework that allows the incorporation of physical principles and other prior information to regularize the behavior of the predictive model. This paper presents a new physics-constrained machine learning method for reliability-based product design and safety estimation considering partially available limited reliability information.