Online Model Discovery under Parameter Uncertainty for Risk and Reliability-Based Decision Support

This paper presents an online identification framework for uncovering the governing equations of dynamical systems operating under degradation to support risk and reliability-informed decision making. To enable accurate discovery from large basis function libraries that capture the complexity of process systems, the wide-array of nonlinear dynamic approximation (WyNDA) method is employed and enhanced through the integration of recursive sparse regression (RSR) and recursive sparse Bayesian learning (RSBL). The proposed approach is validated through numerical simulations of a subsea system under varying degrees of sparsity. In addition, parameter drifts is introduced to emulate degradation effects and their estimates are utilized as evidence for decision supports. Specifically, remaining useful life (RUL) prediction based on a Wiener degradation process and the use of a Bayesian belief network (BBN) are demonstrated. The results demonstrate the potential of the proposed approach as a realtime solution for decision support in systems with limited data availability, while maintaining high accuracy.

Keywords: System identification, decision support, WyNDA, subsea seawater injection system, RUL, and BBN.