A Bayesian belief-rule-based inference multivariate alarm system for nonlinear time-varying processes

This study considers the multivariate alarm design problem of nonlinear time-varying systems by a Bayesian belief-rule-based (BRB) method. In the method, the series of belief rules are constructed to approximate the relationship between input and output variables. Hence, the method does not require an explicit model structure and is suitable for capturing nonlinear causal relationships between variables. For the purpose of online application, this study further introduces sequential Monte Carlo (SMC) sampling to update the BRB model parameters, which is a fast and efficient method for approximately inferring nonlinear sequence models. Using the model parameters obtained by SMC sampling, the series of output variable tracking errors can be estimated and employed for multivariate alarm design. The case study of a condensate pump verifies the effectiveness of the proposed method.