Towards automatic anomaly detection and diagnosis in positional arc-directed energy deposition based on deep learning

Positional arc-directed energy deposition (DED) is a highly anticipated technique with high degrees of freedom for fabricating overhang structures without support structures and positioners. Nevertheless, the immaturity of monitoring and control techniques for ensuring a stable deposition process in positional arc-DED is the main challenge in achieving reliable and automatic manufacturing of metal components. This study aims to identify the molten pool state and diagnose the occurrence of hump and drop defects in positional arc-DED based on deep learning. Five convolutional neural network (CNN) models are used to perform the classification task of molten pool states to determine the optimal architecture of the defect detection model based on their classification performance. A target recognition framework, called YOLOv5s, is used to construct a hump defect diagnosis model and a drop defect diagnosis model to diagnose the occurrence of hump and drop defects, respectively. Compared to other CNN classification models, ResNet18 can effectively balance the performance and the computational resource requirement, obtaining an excellent classification accuracy of 0.996. The maximum detection error of the hump defect diagnosis model for extracting the molten pool height and length is less than 0.68 mm. The molten pool dimensional ratio is used to evaluate the probability of the hump defect occurrence. The molten pool dimensional ratio increases gradually during one formation cycle of the hump defect. The drop defect diagnosis model successfully avoids the interference of the burning arc and obtains an accurate detection result, with a maximum detection error of less than 2.09 mm². The molten pool area is introduced to evaluate the probability of the drop defect occurrence in positional arc-DED, which decreases first and then increases as the drop falls from the molten pool. This study lays a solid foundation for controlling the stability of the deposition process in positional arc-DED.