Mig Welding Defect Detection and Classification Based on VGG16 Convolutional Neural Network

In the field of high-precision welding manufacturing, Metal Inert Gas (MIG) welding can provide a stable melt pool and low deformation rate, resulting in sturdy welds. However, MIG welding faces some problems in practical applications, such as welding quality monitoring relying more on manual labor, being affected by various materials and process parameters, and low levels of automation and intelligence. It is difficult to capture complex welding nonlinear features and requires a lot of manual recognition and intervention. In response to the above issues, an online monitoring method for MIG welding process has been developed, using deep learning techniques, especially convolutional neural networks (CNN), to accurately identify multiple welding states, and combining image enhancement techniques with center loss measurement learning strategies to improve the accuracy and robustness of the welding model. Through classification training, testing, and verification of welding images, the classification of welding states that occur during the welding process, such as normal, burnt through, incomplete fusion, and incomplete penetration, has been achieved with an overall accuracy rate of over 98.75 %, meeting the requirements of actual production.