Online monitoring of GMAW welding defect based on deep learning

Utilizing the aluminum alloy exterior plate of the driver's cab of high-speed railway in rail transit as the substrate, the research is conducted on key intelligent welding technologies, focusing on the issue of online monitoring of welding defects. With the help of process test platform and welding procedure specification, welding defect experiment design, batch data collection, expert experience calibration and database construction are implemented. The convolutional neural network algorithm is used to construct multi-dimensional information fusion models for different types of data, and parameters of the fusion models are optimized. Finally, training, verification and testing of fusion models are completed. The results show that the fusion model after training has better recognition results for welding defects than the single information model. The monitoring accuracy of welding defects in the training set and the testing set is 99.0% and 88.3%, respectively. The data acquisition and model response total time for this monitoring system is less than 100 ms, which meets the requirements for engineering applications, enhances the level of intelligence in robotic welding, and drives the digital transformation and upgrading of enterprises.