Focal point position optimization with neural networks in CO 2 laser welding

Abstract

CO<SUB>2</SUB> lasers are increasingly being utilized for quality welding in production. Considering the high equipment cost and high productivity, the start-up time and set-up time for new products should be minimized. Today most parameters involved in laser welding still have to be manually fine-adjusted when initiating welding of a new product. Ideally the parameters should be set and optimized more or less automatically. In this work the feasibility to automatically optimize the focal point position, one of the most critical parameters in laser welding, is analyzed. In a number of systematic laboratory experiments, a 1150 W CO<SUB>2</SUB> laser is used to weld 2 mm sheets of mild steel. In the experiments the focus point position is continuously changed during a welding trial, and the process is simultaneously monitored by two photo diodes, one on either side of the workpiece surface. In a number of systematic investigations, the welding speed and the power level are varied. After welding, a number of artificial neural networks are designed to recognize the optimum focus point position. The efficiency and accuracy of the neural networks are then tested on new welds, performed with similar parameter settings as the first set of welds performed. The results show good agreement between the real position of the optimum focus point and the calculated values. Finally a trained neural network has been implemented into a closed-loop control system with one top side photo diode as a sensor. Preliminary test demonstrate that neural networks can be used to optimize the focus point position with good accuracy in cw CO<SUB>2</SUB> laser welding.