Neural network of plume and spatter for monitoring high-power disk laser welding

Abstract

Using multi-characteristic information fusion based on a BP (back propagation) NN (neural network) of the plume and spatters to monitor the high-power disk laser welding of type 304 austenitic stainless steel is presented. An ultraviolet and visible sensitive highspeed video camera was used to capture the dynamic images of laser welding plume and spatters during laser welding. The number and area of spatters, and the area, height, tilt angle and centroid of plume were calculated by using image processing technology and defined as the characteristic parameters of plume and spatters, which were used as inputs of the neural network. The weld bead width was considered as a parameter reflecting the welding status, which was used as output of the neural network. Relations of plume and spatters with laser welding status was established by a BP neural network and experimental results showed that the proposed method could effectively estimate the high-power disk laser welding status when the laser power ranged from 2 kW to 10 kW.

Progress and perspectives of in-situ optical monitoring in laser beam welding: Sensing, characterization and modeling

“…By analyzing the geometrical characteristics of molten pool shadow, the relationship between the morphology of molten pool and the LBW stability was investigated quantitatively. Another reported studies [88]- [90], [209] used two high-speed cameras in different wavelength to capture the vapor plume and spatters in high-power disk laser welding, the result shows that the measurement of UV/VIS-camera light was more appropriate for plume and spatter detection. Meantime, the high-speed NIR-camera was setup at the position of 60 °to the horizontal direction to clearly monitor the behaviors of keyhole and melt pool.…”

Online evaluation of resistance spot welding quality and defect classification

“…An analytical model for predicting weld penetration is established via physical modeling to achieve online measurement capability. Gao Xiangdong's team [16][17][18][19][20] proposed a new nondestructive testing technique based on magneto-optical imaging detection; the test results indicate that the proposed method for identifying weld defects in welding parts yields favorable results. Wan et al [21] proposed a generalized regression neural network to evaluate the weld quality of small-scale titanium alloys by using features extracted from voltage signals.…”

Detection of ceramic tile defects based on improved YOLOv5

“…Although the traditional visual inspection method has many applications, its adaptability and accuracy are weak. [6][7][8] At present, deep learning methods are widely used in industrial inspection, but there are still few detection models suitable for ceramic tiles. Zhao Chu et al [9] proposed an improved Faster RCNN algorithm for tile surface defect detection.…”