A NEW METHODOLOGY FOR FAULT DETECTION OBSERVRS IN VAV SYSTEMS

EQUIPMENT INSTALLED IN BUILDINGS FOR HEATING, VENTILATION, AND AIR CONDITIONING IS BECOMING INCREASINGLY COMPLEX. OPERATION WITH CONFLICTING GOALS IS GETTING MORE DIFFICULT. METHODS ARE NEEDED TO SIGNIFICANTLY IMPROVE THE REAL-TIME DETECTION AND DIAGNOSIS OF FAULTS IN BUILDING EQUIPMENT. CURRENT APPROACHED TO SOLVING THIS PROBLEM DO NOT WORK WELL IN NONLINEAR SYSTEMS SUCH AS VAV, AND CAN GIVE INCORRECT RESULTS DUE TO SENSOR BIAS, MODEL INACCURACIES AND MODELLED DISTURBANCES OR OTHER NOISE. EXTENSIONS OF TRADITIONAL TECHNOLOGY TO NONLINEAR SYSTEMS ARE OFTEN NOT ROBUST. FURTHERMORE, THE IMPLEMENTATIONS ARE TIMECONSUMING AND ARE DIFFICULT FOR NONSPECIALISTS TO APPLY, LIMITING WIDESPREAD USE. THE PROPOSED RESEARCH WILL DEVELOP A NEW METHODOLOGY FOR CREATING FAULT DETECTION OBSERVERS (FDO'S) FOR DETECTING AND DIAGNOSING THESE PROBLEMS. IT WILL OVERCOME THE WEAKNESSES OF BOTH PURE MODEL-BASED APPROACHES AND PURE PATTERN RECOGNITION APPROACHES, BY A COMBINATION OF MODEL-BASED AND NEURAL NETWORK TECHNIQUES. IT WILL EMBED THESE TECHNIQUES IN AN OVERALL OBJECT ORIENTED GRAPHICAL EXPERT SYSTEM ENVIRONMENT TO SIMPLIFY FUTURE WIDESPREAD DEPLOYMENT. FOR PHASE 1, A PROTOTYPE WILL BE PROGRAMMED AS PART OF THE RESEARCH. PHASE 2 WILL RESOLVE ADDITIONAL OUTSTANDING QUESTIONS, PRIOR TO ULTIMATE COMMERCIALIZATION.