Large-scale plant-wide process modeling and hierarchical monitoring: A distributed Bayesian network approach

Abstract

In this work, a systematic distributed Bayesian network approach is proposed for modeling and monitoring large-scale plant-wide processes. First, to deal with the large-scale process modeling issue, the entire plant-wide process is decomposed into blocks and Bayesian networks are constructed for different blocks. Subsequently, distributed Bayesian network blocks are fused into a global Bayesian network with a proper designed algorithm. For fault detection, a missing data approach is proposed for state estimation, based on which the T2 and Q statistics are constructed. Finally, a Bayesian decision fusion mechanism is established for hierarchical monitoring of variables, unit blocks and the global industrial plant. For fault isolation, a Bayesian contribution index is further developed and the corresponding isolation scheme is proposed. Simulation results on the plant-wide Tennessee Eastman process show that the distributed Bayesian network approach can be feasible for modeling large-scale process. Furthermore, the proposed hierarchical monitoring scheme provides informative multi-level reference results for further diagnosis and isolation.