Research on recognition method of laser cladding dilution rate and contour quality based on Semantic Segmentation Visual Geometry Group network

Dilution rate and contour quality are important indicators for studying the forming quality of laser cladding technology. Currently, the acquisition of the dilution rate mainly relies on manual marking under the metallographic assistant software, and the evaluation of contour quality mainly relies on human observation. In order to solve the problem that the manual labeling of the dilution rate and the manual evaluation of the cladding layer contour quality rate are low and subjectively affected, this paper proposes the use of artificial intelligence to identify the cladding layer dilution rate and perform a parameterized evaluation of the cladding layer contour. The Semantic Segmentation Visual Geometry Group (SSVGG) network is proposed by improving the VGG16, and the improvement of the convolutional layer reduces the calculation of the convolutional network. Adding a deconvolution layer enables the features to be displayed to the original image dimension. The feature fusion layer is added to increase semantic information. The experiment proves that the accuracy of Intersection over Union for extracting cladding topography by the SSVGG network is 93.89%. The neural network can accurately output the size of the cladding layer and the dilution zone, the accuracy rate of obtaining the dilution rate is 93.33%, and the dilution rate obtained by the area method can better characterize the cladding effect compared with the length method. The arc of the cladding layer output by the neural network can be parameterized to indicate the quality of contour forming.