EARLY WARNING FAILURE DETECTION USING NEURAL NETWORKS

MACHINERY HEALTH MONITORING IS CRITICAL IN SYSTEMS WHERE REDUNDANCY CAPABILITY IS LIMITED AS IN A HELICOPTER GEARBOX. IN ORDER TO DECIDE ON THE MOST EFFICIENT, SAFE, AND COST-EFFECTIVE PROGNOSIS, THE DEGREE OF SEVERITY OF A FAULT MUST BE DETERMINED. AT PRESENT, CONDITION ASSESSMENT TECHNIQUES ARE BEING EXPLORED WHICH REQUIRE A LARGE AMOUNT OF PROCESSING AND PRODUCE A LIMITED CATEGORIZATION OF THE DEGREE OF SEVERITY. THEY ARE ALSO LIMITED BY LOW SIGNAL TO NOISE LEVELS AT THE EARLY FAULT STAGE. A DIAGNOSTIC SYSTEM FOR SUCH CONDITION ASSESSMENT IS PROPOSED USING A NEURAL NETWORK TO IMPROVE ON THE SPEED AND PRECISION OF DIAGNOSIS. PSI TECHNOLOGY COMPANY PROPOSES TO DEVELOP A TECHNIQUE FOR REAL-TIME, CONTINUOUS HEALTH MONITORING OF ROTATING MACHINERY USING NEURAL NETWORKS TO PROCESS RAW TIME TRACES. IN PHASE I, EXISTING IN-HOUSE, WELL-CHARACTERIZED BEARING DATA FROM A CONTROLLED EXPERIMENT ON BEARINGS TESTED UNTIL DESTRUCTION, WILL BE USED TO TRAIN AND TEST A NETWORK WHICH WILL PROVIDE A DECISION ON WHETHER A CERTAIN LEVEL OF THE SEVERITY OF A FAULT HAS BEEN REACHED. THE HIGH SPEED PERFORMANCE OF SUCH A NETWORK WILL BE CHARACTERIZED FOR ACCURACY AND RANGE OF FAULT DIAGNOSIS. HARDWARE REQUIREMENTS FOR A NEURAL NETWORK IMPLEMENTATION WILL ALSO BE DETERMINED. IN PHASE II, CONSTRUCTION AND SITE TESTING OF A PROTOTYPE, AS WELL AS AN EXPANDED REPERTOIRE OF COMPONENT DIAGNOSTICS WILL BE CARRIED OUT. THE ABILITY TO DETECT FAULTS AND TO ASSESS THE SEVERITY OF SUCH FAULTS IN ROTATING MACHINERY IN REAL-TIME MAY BE USED TO MONITOR THE ASSEMBLY LINE QUALITY OF MACHINERY. IT MAY BE USED ON ANY TURBINE OR PUMP SYSTEM WHOSE HEALTH IS TIME CRITICAL. THE ELECTRIC UTILITIES COULD USE SUCH A SYSTEM TO MONITOR TURBINES AND GENERATORS AND MINIMIZE DOWN TIME AND REPLACEMENT COSTS. ADDITIONAL USES INCLUDE MONITORING THE HEALTH OF COMPLEX AUTOMATED MANUFACTURING EQUIPMENT TO REDUCE THE RISK OF ASSEMBLY DOWN TIME OR POOR QUALITY COMPONENTS.