Data-driven analysis of surface roughness influence on weld quality and defect formation in laser welding of Cu–Al

Abstract

The laser welding of Cu–Al alloys for battery applications in the automotive industry presents significant challenges due to the high reflectivity of copper. Inadequate bonding and low mechanical strength may occur when the laser radiation is directed toward the copper side in an overlap configuration welding. To tackle these challenges, a laser surface treatment technique is implemented to enhance the absorption characteristics and overcome the reflective nature of the copper material. However, elevating the surface roughness and heat-energy input over threshold values leads to heightened temperature and extreme weld. This phenomenon escalates the formation of detrimental intermetallic compounds (IMC), creating defects like cracks and porosity. Metallurgical analysis, which is time-consuming and expensive, is usually used in studies to detect these phases and defects. However, to comprehensively evaluate the weld quality and discern the impact of surface structure, adopting a more innovative approach that replaces conventional cross-sectional metallography is essential. This article proposes a model based on the image feature extraction of the welds to study the effect of the laser-based structure and the other laser parameters. It can detect defects and identify the weld quality by weld classification. However, due to the complexity of the photo features, the system requires image processing and a convolutional neural network (CNN). Results show that the predictive model based on trained data can detect different weld categories and recognize unstable welds. The project aims to use a monitoring model to guarantee optimized and high-quality weld series production. To achieve this, a deeper study of the parameters and the microstructure of the weld is utilized, and the CNN model analyzes the features of 1310 pieces of weld photos with different weld parameters.