Development and Validation of a Dynamic Bayesian Network‐Based Adaptive Reliability Optimization Model Using Design of Experiments and Expected Information Gain

Abstract

This study proposes a dynamic Bayesian networks (DBN)‐guided adaptive reliability experimental design framework for dynamic reliability learning and uncertainty reduction in complex engineering systems under operational and budgetary constraints. The methodology integrates DBN, Bayesian experimental design, entropy‐based uncertainty quantification, expected information gain, and utility‐based decision theory into a unified probabilistic framework capable of sequentially selecting the most informative experiments according to their expected contribution to uncertainty reduction while considering implementation costs. The proposed approach transforms reliability analysis from a passive estimation process into an active learning methodology capable of guiding adaptive experimentation under uncertainty. Owing to its structure, the framework can be applied to a wide range of industrial systems operating under uncertainty. For validation purposes, the framework was implemented on a pumping system composed of three critical components, an electric motor, a pump, and a control valve. Reliability evolution was modelled over successive 31 temporal slices, while adaptive experiments were selected according to entropy reduction and cost‐aware utility maximization. Candidate experiments are evaluated through entropy reduction and EIG calculations, while a utility function balances informational value and experimental cost to identify the optimal tests at each stage. The results demonstrate significant uncertainty reduction across all subsystems. The motor achieved a 99% reduction in mean entropy, while the valve and pump achieved uncertainty reductions of 86% and 75%, respectively. These uncertainty reduction and optimal experiment selection process under cost constraint and reliability evolution confirmed the performance, accuracy and superiority of the proposed methodology in adaptive decision‐making and experiment prioritization.