Detecting abnormal process trends by wavelet‐domain hidden Markov models

A novel method for detection of abnormal conditions during plant operation uses wavelet‐domain hidden Markov models (HMMs) as a powerful tool for statistical modeling of wavelet coefficients. By capturing the interdependence of wavelet coefficients of a measured process variable, a classification strategy is developed that can detect abnormal conditions and classify the process behavior on‐line. The method is extended to include multiple measured variables in detection and classification. Two case studies illustrate the potential of this method.