Motivation and Goals

Uncertainties are ubiquitous in the design and analysis of engineered structures. Reliability-based designed optimization (RBDO) is an optimization method that can provide optimum designs in the presence of uncertainties. The application of RBDO procedures to structural systems is challenging due to the need to handle multiple (noisy) cost functions, as well as probabilistic constraints defined by the critical structural failure events. Monte Carlo simulation has emerged as a versatile solution technique for RBDO problems. Machine learning (ML) tools, in particular, surrogate models, can alleviate the high computational cost of simulation-based methods. The project aims to propose efficient strategies that combine Monte Carlo methods and ML techniques for optimizing structural design under uncertainty.

Recent Results 

In a recent study, we developed an advanced Monte Carlo framework to solve the RBDO problem using cross-entropy based importance sampling. The approach requires one to evaluate the structural reliability constraints for every candidate design generated during optimization. To accomplish this, we proposed a method that efficiently and accurately estimates the (small) structural failure probabilities using the principles of importance sampling and line sampling. The research findings are documented in a manuscript, to be submitted to the journal Reliability Engineering and System Safety. In future, this method will be integrated with the envisaged RBDO procedure using cross entropy-based importance sampling. In an ongoing work, we are developing an active learning-based surrogate modelling framework to enhance the efficiency of Monte Carlo simulation methods for structural reliability assessment and design optimization. Preliminary results were presented in the 2024 GNI symposium.