A Naive Bayes Classifier for Fault Detection and Classification Using Dimension Reduction Technique

Fault detection and classification is critical to the reliability of modern control systems in different industries, where detecting and classifying faults in operational processes are very important things while failure to detect and classify them, may cause irreparable damage. In this paper, fault detection and classification approach is presented. The first step, multi stage recursive least squares parameter estimation approach for controlled autoregressive autoregressive moving average systems (CARARMA) is developed with a view to estimate the parameters of the system, additionally, improve the effectiveness of the computation. By means of multi stage approach, the (CARARMA) system is decomposed into three simple identification models, and the parameters of each simple model is identified one by one. These parameters estimated by this approach are referred to as features, and not all of them have the same useful data about the system. In order to select the valuable features and improve a classification accuracy, the Linear Discriminant Analysis (LDA) approach based on scattering matrices is applied for dimension reduction. The classification between these reduced classes is done based on the Naive Bayes classifier. Finally, the obtained results explain the performance of this proposed approach.