SIGNAL VALIDATION BY COMBINING MODEL-BASED AND EVIDENTIAL REASONING APPROACHES

THIS PROJECT WILL MERGE METHODS DERIVED FROM ARTIFICIAL INTELLIGENCE TECHNOLOGY WITH CONVENTIONAL ANALYTIC REDUNDANCY METHODS, THE OBJECTIVE BEING TO OBTAIN IN REAL-TIME HIGHLY SENSITIVE AND COST-EFFECTIVE IDENTIFICATIONOF DEGRADATION IN POWERPLANT INSTRUMENTATION. THE RESEARCH WILL START WITH THE EXPERIENCE THAT IS AVAILABLE IN THE USE OF SIMULATION MODELS AND OBSERVERS/FILTERS TO ESTIMATE UNMEASURED STATES. STATISTICAL ANALYSIS OF THE RESIDUAL DIFFERENCE BETWEEN MEASURED AND PREDICTED VALUES WILL BE USED TO IDENTIFY ANOMALIES. THE WEAK LINK IN THE APPLICATION OF THESE METHODS IN THE PAST HAS BEEN THE CODIFICATION OF THE EXPERIENCE NEEDED TO INTERPRET THE RESULTS ACCURATELY AND AUTOMATICALLY. THE USE OF EVIDENTIALREASONING EMBEDDED INTO AN EXPERT SYSTEM WILL BE INVESTIGATED IN THIS PROJECT TO FACILITATE THE IMPLEMENTATION OF SUCH HEURISTICS.