Wavelet Theory-based Model for Early Warning of Chemical Process Faults
by
Manish Misra
University of South Alabama, Mobile, AL 36688
Coauthors: Rudramurty Balla and Satya N. Mishra

Chemical processing industry relies heavily on efficient monitoring and control of its processes. Several considerations, such as chemical process plant safety, production specifications, environmental regulations, operational constraints, and plant economics are some of the reasons driving an upward interest in research and development of more robust methods for process monitoring and control. Recent research in chemical process monitoring has focused on process and sensor fault detection and diagnosis, where the formulated approaches undertake corrective actions only after a fault has been detected.

However, it is rather desired to develop an intelligent system that can predict the future state of the process, thereby initiating corrective actions before a fault occurs. In this paper, a wavelet-based model is developed for chemical process monitoring. The inspiration for the wavelet based model stems from the fact that most chemical process are multi-scale in nature, in that they present a convoluted picture of events that have occurred at multiple time-frequency scales, and at different resolutions. Wavelets provide a joint time-scale representation of a time-series signal through its projection onto nested orthogonal subspaces, with finer subspaces containing coarser subspaces in a multi-resolution framework. With their time-frequency localization and multi-resolution property, wavelets have offered an attractive framework for multi-scale representation of data. Such properties of wavelet transform have been used in developing a model for predicting the future states of chemical processes. In tests with industrial data, the wavelet based prediction model consistently provided superior results as compared to time-series based regression models.

Date received: October 6, 2004