Fault Diagnosis Based on Bayesian Networks for the Data Incomplete Industrial System

In the data-incomplete industrial systems, the existing data-driven fault diagnosis techniques cannot be applied directly due to the missing of sampled data. In this paper, we propose a method based on bayesian networks to realize the fault diagnosis of systems with incomplete sample data. Our method uses the Expectation-Maximization (EM) algorithm to estimate the missing part of incomplete sample data, then selects the features based on the mutual information technique, and finally, constructs the bayesian network classifier to achieve the fault diagnosis of systems. We used the Tennessee Eastman Process as the simulation model, and analyzed the diagnostic performance under different degrees of missing data. Both the normal case and three faults had been considered in the simulation. Compared with the data-complete case, our method achieved a good diagnosis performance in the case within 10% rate of missing sample data.