COBEM99

S18 Processos Mecânico-Metalúrgicos

Title:

COMPARISON BETWEEN MULTIPLE REGRESSION ANALYSE AND NETWORK TECHNIQUES ON WELDING GEOMETRY MATHEMATICAL MODELS USING SYNERGIC PULSED MIG PROCESS

Summary :

ABSTRACT. THIS PAPER AIM TO COMPARE GEOMETRIC WELDING BEAD MODELS OBTAINED BY NEURAL NETWORK BACK PROPAGATION TYPE WITH THOSE OBTAINED BY MULTIPLE REGRESSION ANALYSE TECHNIQUES. THE TECHNIQUE OF “FACTORIAL DESIGN” WAS USED FOR EXPERIMENTAL DESIGN. IT WAS ANALYSED FOUR WELDING PARAMETER (CURRENT, VOLTAGE ADJUSTMENT, WELDING SPEED AND STAN OFF) AT 3 LEVELS, USING THE COMPLETELY DESIGN 34, HAVING IN TOTAL 81 WELDING RUNS. THE SYNERGIC PULSED MIG PROCESS INSIDE OF ROBOTIC CELL WAS USED TO WELD THE SPECIMENS. THE INDEPENDENCE VARIABLES WAS THE WELDING PARAMETERS AND THE DEPENDENT VARIABLES OR RESPONSE WAS WELD BEAD DIMENSIONS. THE MODELS WAS OBTAINED BY USING STEPWISE REGRESSION ANALYSES AND COMPARED GRAPHICALLY WITH THESE OBTAINED BY NEURAL NETWORK. THE RESULTS HAS SHOWN, AFTER THE WELDING BEAD VALIDATION, THAT THE NEURAL NETWORK MODELS ARE MORE PRECISE IN PREDICTING WELDING GEOMETRY AND COULD BE CONSIDERED AS EXCELLENT TOOL FOR WELDING PROCESS OPTIMISE. KEY-WORDS: SYNERGIC PULSED MIG, WELDING PARAMETERS, SYNERGIC CONTROL, NEURAL NETWORK, MULTIPLY REGRESSION ANALYSE

Author :

Alfaro, Sadek C.A.
Silva, José H. F. da

Paper View :