LAUSR

Welding status determines the post-weld quality and is crucial to high-power disk laser welding. A low-cost monitoring system based on two photodiodes is developed to monitor the real-time welding statuses in this paper. A deep learning architecture based on stacked autoencoder (SAE) is proposed to automatically learn more representative features of welding statuses from the raw signal features captured by the visible light photodiode and the reflected laser light photodiode without any manual operations. The maximum correntropy loss function is applied to improve the learning ability of the proposed SAE method. Furthermore, a genetic algorithm is applied to optimize the key parameters of the proposed SAE method. The proposed SAE is applied to the high-power disk laser welding experiments and shows better performance in welding status monitoring than the standard AE framework and the conventional SVM and BP method. Additional experiments with different welding parameters validate the effectiveness and robustness of our proposed SAE method.