An Image Segmentation Algorithm for Extracting Melt Pool Parameters in Wire Arc Additive Manufacturing

Tungsten inert gas (TIG) welding is an important subset of wire arc additive manufacturing technology where tungsten serves as the electrode, argon as the protective gas, and the heat produced by the electric arc melts the target material, shaping it into a stack following a predetermined path. The molten pool is crucial for determining the quality of TIG additive manufacturing weld seam formation. However, a lack of effective and efficient sensing and extraction techniques makes it difficult to obtain the characteristic parameters of the melt pool. To tackle this challenge, a method based on image segmentation is proposed to accurately extract melt pool parameters, providing essential data support for developing melt-through prediction models. To achieve this, the optimized Otsu algorithm based on genetic algorithm (GA-Otsu) and the improved STDC-BiSeNet algorithm (ES-BiSeNet) were used to segment the melt pool, respectively. The ES-BiSeNet introduces a self-attention mechanism as an enhancement to the convolution module. Meanwhile, a Rep-VGG style structural reparameterization is designed to separate the training time and inference time architectures. The accuracy of the extraction algorithm was validated through process experiments using 304L steel material on a robotic additive manufacturing system. The results show that the improved BiSeNet can accurately segment the melt pool. By examining the weld cross-section, it can be determined that the maximum error in extracting melt width is 5.83%, which is generally sufficient for accurately extracting melt pool features. This method can effectively extract the characteristic parameters of the melt pool, which plays a crucial role in establishing the melt penetration prediction model and advancing the automation of wire arc additive manufacturing.