Fault Identification for Missing Data Systems with a Dynamic Bayesian Network Approach

For the problems in monitoring the industrial systems, such as non-Gaussianity, dynamic nature and missing data, a fault identification method based on dynamic Bayesian network is proposed. A dynamic Bayesian network with mixture of Gaussian output (DBNMG) is constructed, and a parameter learning strategy based on expectation maximization algorithm is deduced. For the missing data issue, a non-imputation inference method for DBNMG is proposed to conduct the fault detection and identification with the partially observed data. The proposed approach is evaluated with the continuous stirred tank reactor (CSTR). Simulation results demonstrate the effectiveness of the proposed method.