A Bayesian Fault Diagnosis Method Incorporating Background Knowledge of Abnormality Signs

In process industries, fault diagnosis is to isolate the problem source that degrades control loop performance. In order to reduce dependence on the amount of historical data, this paper focuses on the use of background knowledge and incorporating it into Bayesian diagnosis. We formulated a kind of knowledge, known as Abnormality Signs, as constraints on the underlying probabilities and introduced those constraints into Bayesian diagnosis framework. In this way, the observation space dimensionality can be reduced. The proposed approach is evaluated by an application to a diagnosis problem of a simulated oil sand slurry handling system. The advantage of combining background knowledge with historical data is achieved even when the abnormality is sparse.