COBEM99

S19 Usinagem dos Materiais

Title:

TOOL WEAR DETECTION IN END-MILLING PROCESS VIA ARTIFICIAL NEURAL NETWORKS

Summary :

THIS WORK ADDRESSES THE VIABILITY STUDY OF A REAL TIME TOOL WEAR MONITORING SYSTEM FOR VERTICAL END MILLING PROCESSES BASED ON NEURAL NETWORK TECHNIQUES USING, INEXPENSIVE, ROBUST, NON-INVASIVE CURRENT SENSOR AND ACCELEROMETERS. THE APPLICATION OF MODERN ARTIFICIAL INTELLIGENCE TECHNIQUES BASED ON NEURAL NETWORKS COMBINED WITH PROPERLY CHOSEN PRE-PROCESSING ALGORITHMS MAKES POSSIBLE THE USE OF SUCH LOW COST SENSORS WITH VERY ACCEPTABLE DETECTION PERFORMANCE. THE SELECTED PRE-PROCESSING ALGORITHM IS THE LINEAR PREDICTIVE CODING (LPC) FOR ITS ROBUSTNESS, SPEED EFFICIENCY, DIMENSIONALLY REDUCTION AND ITS RELATION TO THE FREQUENCY SPACE. FOR THIS FIRST VERSION, WE USED THE FEEDFORWARD NEURAL NETWORK PARADIGM, WITH BACKPROPAGATION AS ITS LEARNING ALGORITHM. THE TOOL WEAR DETECTION RESULTS FAVOUR THE APPLICATION OF THIS SYSTEM TO BETTER PRODUCTION EFFICIENCY. WE SHOW THAT, WITH PROPERLY CHOSEN PRE-PROCESSING ALGORITHMS COMBINED WITH THE POWER OF NEURAL NETWORKS, GOOD TOOL WEAR MONITORING RESULTS ARE OBTAINED IN SPITE OF THE SENSORY SUBSYSTEM. THE MILLING PROCESS IS DESCRIBED, SO AS THE ACQUISITION AND SENSORIAL SYSTEM USED FOR THIS WORK. FOLLOWS THE LPC PRE-PROCESSING WITH ITS ADVANTAGE AS A FEATURE EXTRACTOR, DIMENSIONALLY REDUCTOR AND ITS RELATION TO THE FREQUENCY SPACE. THEN, WE MAKE A BRIEF PRESENTATION OF NEURAL NETWORKS FOCUSING ON THE KIND USED, FOLLOWED BY THE NEURAL TRAINING METHODOLOGY, DESIGN CONSIDERATIONS AND THE NETWORK TOPOLOGY FOR THE BEST TRAINING RESULT OBTAINED. THE FAST DETECTION OF TOOL WEARINESS CONDITION IS VERY IMPORTANT FOR PRODUCTION EFFICIENCY AND QUALITY, IN THAT THE TOOL CAN BE RAPIDLY REPLACED, REDUCING PRODUCT REJECTION. THIS IS OF THE MOST IMPORTANCE FOR FULLY AUTOMATED, UNATTENDED MILLING PROCESSES. DUE TO THE USE OF ONLY ONE SENSOR (VARIABLE), EFFICIENT PRE-PROCESSING ALGORITHM IN TIME AND QUALITY, AND CONSEQUENTLY SMALL NEURAL NETWORK SIZE, THE WHOLE PROCESS CAN BE EASILY IMPLEMENTED IN A CONVENTIONAL COMPUTER FOR REAL TIME OPERATION.

Author :

Almeida, Ailson Rosetti de
Dos Santos, Marcelo Teixeira
Tu, Carlos Chien Chin

Paper View :