29D-13 |
A hybrid approach to detecting and isolating sensor faults and process faults |
X. DONG1, M. T. Morgan1, and T. A. Haley2. (1) Dept. of Agricultural & Biological Engineering, Purdue Univ., 1146 Agricultural & Biological Engineering Bldg., West Lafayette, IN 47907-1146, (2) Food Safety & Regulatory Compliance, Bush Brothers & Co., 1016 E. Weisgarber Rd., Knoxville, TN 37909-2683 Food sterilization in batch retort remains as one of the largest food processes. Optimal time-temperature profile is determined by considering different objective functions involving nutrient retention, process time and energy consumption as well as constraints reflecting the required degree of microbial inactivation. Any deviations from the specified processing condition, which may be due to the erroneous operations, changing upstream conditions such as filling tank temperature or sudden drop of supply steam pressure, or equipment faults, will lead to either over-processed or under-processed products. Most of these abnormalities will remain unnoticed by the operator until it’s too late to take actions. We have presented an approach based on Multi-way Principle Components Analysis (MPCA) for the detection and isolation of sensor faults on the 2002 IFT Conference. The power of this approach on distinguishing sensor faults and other process faults has remained unclear. The objective of this study is to improve the data-driven diagnosis system to be able to detect and isolate different types of process abnormalities. A hybrid approach that makes use of both system intrinsic dynamics and historical data was developed. First, a statistical "fingerprint" of normal operations was obtained using MPCA for all process inputs and outputs from historical data. Next, the verified process model was used to generate similar data array for various device and process faults. By simulating various fault scenarios and subsets of process signals, statistical patterns that match particular device and process faults were identified, with which a library of patterns associated with specific faults was built. The simulation study showed that process faults that of importance such as air leaking, and sensor faults could be detected and isolated promptly. The diagnosis system is being tested in the pilot plant. By distinguishing sensor faults and process faults, different accommodating actions can be taken either by the computer or by the operator to ensure food and plant safety.
Session 29D, Food Engineering: Thermal processes
|