Real-time monitoring of laser welding based on multiple sensors

Adequate penetration is a basic property of qualified welds. The proposed monitoring system was constructed to classify the welding state as adequate penetration or inadequate penetration by using infrared, ultraviolet and sound sensors. In order to achieve a smooth data acquisition and a quick response of the system, the circular buffer technology and multithreading/multitasking technology were applied. Since on-line parameter adjusting is very important for a high quality monitoring system, the particle swarm optimization algorithm was used to condition the weights of the neural network. Experimental results show that the system is feasible for monitoring laser welding process in real time.