Real-time implementation of an intelligent diagnostic and control system for bioreactors
Cardello, Ralph Joseph
Doctor of Philosophy
Bioprocesses often require considerable expertise to insure proper operation and performance. However, they are also rich in both data and factual knowledge. Supervision and control can be improved by utilizing information from all available sources. Knowledge about dynamics may be incomplete, particularly during upsets or equipment failure. Additional information derived from past data trends, expert operators and the literature can be very useful in process analysis, fault detection, process salvaging and optimization. In this work an expert system which utilizes on-line real-time data as well as archived data trends and operator expertise to analyze a process is developed. A modular format (the hierarchical modular structure) is designed to organize the large body of information--typical of the knowledge likely to be encountered in a bioprocess. The highly modular format allows for knowledge to be classified as general, process configuration specific or system specific. The expert system is challenged with process faults which are not easily noticed without a qualitative understanding of process dynamics. The supervisor is able to detect the unusual situations by first generating a process assessment based on a synthesis of all available on-line measurements in conjunction with known qualitative information. Once a global process picture is established, the expert system can decide upon appropriate strategies for recovery or backup control. Process analysis, recovery and optimization were experimentally demonstrated using Clostridium acetobutylicum fermentation as a sample system. Successful expert system process analysis was able to detect sensor failures and media feed flow failures. In addition, effective backup pH control and process recovery strategies were suggested and implemented in real-time by the expert system supervisor. The expert system ability to aid on-line optimization by shortening the time needed to analyze process transients may result in faster attainment of the region of optimal performance.
Chemical engineering; Biomedical engineering; Artificial intelligence