Arc behaviour recognition and characterization analysis by using machine learning

In this paper, we propose a new method based on the combination of local binary pattern (LBP) and GoogLeNet neural network to identify the arc patterns in the monitoring images of three types of arc states, namely, stable arc, swinging oscillation, and circumferential oscillation, in the wire arc additive manufacturing process. The results show that obtaining the texture features in the arc pattern image via local binary pattern, and then building the GoogLeNet neural network model can effectively identify the arc length, arc width, and left and right maximum inclination with the number of stacked layers, which can be used to accurately identify the arc state compared with the direct training of neural network on the original image. For the arc morphology images in where are influenced by droplets, complex background and other factors, the proposed method can achieve a clear arc edge, whichbenefics boundary identification of melt pool, droplets and arc morphology. The extract accuracy of arc state is up to 99.50%. The research outcomes will provide a theoretical reference for monitoring arc state during wire arc additive manufacturing process.