Automatic intelligent plant operation and control requires methodology that uses plant measurements in some optimal manner. The visual and auditory systems in human beings preprocess information before presenting it to the brain. For example, large sets of high frequency raw data are analyzed to extract data trends which are as detailed as necessary for the situation at hand. The purpose of this research project is to develop procedures to mimic this preprocessing of data. In chemical plants a number of factors need to be considered the time constants of various responses are typically different and one is required to consider each variable at a different scale, there is multirate sampling a temperature measurement is typically available much more frequently than a composition measurement, sensor signals vary greatly for each process, etc. This project is aimed at developing methodology to preprocess raw sensor data and develop viable automatic control systems for chemical plants. Two functional analysis methods the Wavelet bases and the Frazier Jawerth bases will be used to preprocess data which will then be fed into a neural net to yield process trend analysis used to detect failed sensors and abnormal operating conditions. The combination of Frazier Jawerth transforms and Wavelet decomposition enables description of a signal in a single domain which has information about its time evolution and frequency context. Neural networks capture information from examples which typically come from the past operational history or online information.
