Fault diagnosis in industrial processes using principal component analysis and hidden Markov model

An approach combining hidden Markov model (HMM) with principal component analysis (PCA) for on-line fault diagnosis is introduced. As a tool for feature extraction, PCA is used to reduce the large number of correlated variables to a small number of principal components in an optimal way. HMM is applied to classify various process operating conditions, which is based on pattern recognition principles and consists of two phases, training and testing. The moving window for tracking dynamic data is used. The impact of the window length is studied by simulation. The sampling rate used in training data and in test data is different for correct and quick fault diagnosis. Case studies from the Tennessee Eastman plant illustrate that the proposed method is effective.