USING A NEURAL NETWORK TO MONITOR HIGH SPEED PRODUCTION MACHINERY FOR QUALITY CONTROL

AN APPLICATION OF A PROVEN SPEECH RECOGNITION NEURAL NETWORK TO MONITORING PRODUCTION LINE MACHINERY IS PROPOSED. THE NETWORK WILL BE USED TO DETECT MACHINE SET-UP CHANGES AND ANALYSIS WILL BE DONE TO DETERMINE IF THE TIME WAVEFORM DESCRIPTION CAN BE RELATED TO SET-UP AND TO PRODUCT QUALITY. THE PARTICULAR APPLICATION IS A MACHINE THAT PRODUCES HIGH VALUE METAL CANISTERS AT A HIGH SPEED. DATA HAVE BEEN COLLECTED FOR "GOOD" AND "BAD" SET-UP CONDITIONS FROM MICROPHONES, ACCELEROMETERS, LOAD AND FORCE SENSORS. TRADITIONAL SIGNAL PROCESSING TECHNIQUES HAVE BEEN USED ON THESE DATA AND CHANGES HAVE BEEN FOUND. HOWEVER, A PROCESS THAT AUTOMATICALLY LEARNS THE "GOOD" SET-UP AND THEN DETECTS CHANGES WOULD SIGNIFICANTLY REDUCE MONITORING SYSTEM IMPLEMENTATION COST. SINCE THE DATA AND THE PROCESSING SOFTWARE ARE AVAILABLE, CONCEPT FEASIBILITY CAN BE READILY PROVEN IN PHASE I, AND A PROTOTYPE SYSTEM DEVELOPED IN PHASE II. A MANUFACTURING SYSTEM THAT MONITORS PRODUCT QUALITY AUTOMATICALLY AND THAT HAS BROAD APPLICATION WOULD FIND A WIDE ACCEPTANCE IN INDUSTRY.