Identification Of Dendrite Morphology And Key Parameters In Laser Cladding Based On Deep Learnin

Laser cladding is a high-precision and high-efficiency material processing technology.It can melt and cladding metal powders through laser beams to produce metal parts and complex structures with excellent properties.The microstructure of laser cladding layer is one of the important factors affecting the properties of materials.The size and morphology of the grains in the microstructure are related to the mechanical properties and wear resistance of the cladding layer.Its size is related to laser power,scanning speed and other parameters.By optimizing these parameters,the dendrite morphology can be controlled,thereby improving the fatigue performance and crack resistance of the material.In this paper,through the literature analysis method,the research object is determined as dendrite and primary spacing,the data set is established,and the suitable basic network model is selected.After the laser cladding experiment,the metallographic collection was carried out.The collected images are preprocessed by Gaussian filtering,and the dendrite labels are made by Labelme labeling software.So as to complete the production of data sets.In order to accurately identify the dendrite structure from the metallographic image,a suitable neural network needs to be selected.Considering that the goal is to segment the image at the pixel level,the convolutional neural network is selected as the basic model.According to the microstructure characteristics of the cladding layer,for the segmentation of dendrite morphology,this paper proposes a BNC-Unet network based on U-Net,introduces the attention model and BN layer,and explores the number and location of deployment.By comparing the intersection and union ratio of various improved versions of the network,the best performance of the network is obtained.Finally,the automatic identification and segmentation of dendrite morphology in metallographic diagram are realized through the network.The final Io U value of the network reaches 84.2 %,while the original Unet network segmentation result is only 75.23 %.For the automatic identification and measurement of primary dendrite spacing,a primary dendrite spacing network based on Mask-RCNN is proposed in this paper.The data set is made and the network parameters are optimized.The boundary frame algorithm is used to identify and output the primary dendrite spacing more accurately.Finally,the automatic identification and measurement of the primary dendrite spacing in the metallographic diagram is realized through the network,and the primary dendrite spacing network is identified and transmitted.