Research on Curved Parts Surface Quality Detection during Laser‐Directed Energy Deposition Based on Blurry Inpainting Network

Abstract

The laser‐directed energy deposition technology can be used for additive/subtractive hybrid manufacturing (ASHM). ASHM can realize the manufacturing of some complex parts, such as curved parts. Curved parts will inevitably have some defects during the manufacturing process. However, it is difficult to detect these defects, due to the edge blur of the surface. Therefore, a curved surface quality detection method is proposed. First, the error effect of the curved surface on the surface quality detection is quantitatively analyzed. An efficient channel attention network–DPD network (ECANet–DPDNet) blurry inpainting network model is proposed to effectively reduce the adverse effect of edge blurring. Then, the feature parameters of the repaired image are extracted. The backpropagation (BP) neural network trained by the feature parameters is used to predict curved surface roughness. Finally, two kinds of surface defects are identified using our proposed method based on adaptive threshold segmentation matrix and interference region filtering. The constructed support vector machine (SVM) defect type recognition model is trained using the 15 feature parameters extracted from the defect region. The experimental results show that the accuracy rates for the judgment of scratch defects and pit defects can reach 96.00% and 94.00%, respectively.