JOURNAL OF MANUFACTURING PROCESSES, v.80, pp.75 - 86
Abstract
Laser welding of aluminum alloys is a highly complex and unstable process owing to the high reflectivity and high thermal conductivity of aluminum and vaporization of alloying elements. Laser-beam absorptance is one of the factors that greatly influences the process stability, but it is very difficult to measure with conventional monitoring systems. In this study, the laser-beam absorptance and keyhole behavior during the fiber laser welding of Al 5052-H32 was investigated using a deep-learning-based monitoring system. In this method, two synchronized coaxial high-speed cameras were used to simultaneously observe the top and bottom specimen surfaces. From the obtained images, top and bottom keyhole apertures were detected by using an objectdetection deep-learning model. Then, a ResNet-based deep-learning model was applied to the detected keyhole top and bottom apertures to predict the time-varying laser-beam absorptance inside the keyhole considering multiple reflections. By analyzing the monitoring results, we classified the keyhole behavior and absorption patterns into three types, and investigated the process stability and defect formation mechanisms. It was found that, when the keyhole opened constantly under excessive energy conditions, the keyhole size continued to increase and burn-through defects were generated if the aperture area exceeded the threshold values. Furthermore, process stability was improved and defect formation was suppressed by using helium as a bottom-side shielding gas under high-energy-density conditions.