An integrated approach for system functional reliability assessment using Dynamic Bayesian Network and Hidden Markov Model

This paper presents a novel methodology to estimate and predict the functional reliability of a system using system functional indicators and condition indicators of components. Instead of 'system reliability', the paper uses the terminology 'system functional reliability' because the functional indicators used in the methodology principally represent the system performance level or system functionality. The proposed model relates the degradation state of components to the system functional state. The model allows the use of system functional indicators and condition data of components in continuous time domain. The proposed methodology uses both Hidden Markov Model and Dynamic Bayesian Network for estimating and predicting system functional reliability. HMM helps in mapping the continuous data into hidden state probabilities while the system DBN helps in finding the posterior system state probability by considering the component dependencies within a system. The study is also extended to show how the external covariates can be incorporated into the proposed model. Since the external covariates accelerate the degradation of a component, the component state transition probability in the second model is adjusted to vary with respect to the covariates. A case study based on Tennessee Eastman Chemical Process is conducted to demonstrate the proposed methodology for system functional reliability estimation and prediction. Another simulation based case study is presented to describe how the external covariates are included in the presented methodology.