An Intelligent Metallographic Processing System for Identifying Primary Dendrites Based on Deep Learning
L.J. Cui, M.Y. Sun, S.R. Guo, Y.L. Cao, W.H. Zeng, X.L. Li and B. Zheng
A primary dendrite identification method based on deep learning is proposed. The primary dendrite identification is an important part of laser cladding surface performance test which is the base of mechanical properties research. Traditional primary dendrite identification depends primarily on artificial vision detection. With the rapid development of deep learning, industrial automatic detection has become an inevitable trend, because deep learning technology can greatly improve the accuracy and efficiency, therefore, the study of primary dendrite automatic detection technology has great practical value. In this paper, the semantic segmentation neural network in deep learning is used to label the primary dendrite by using only small-scale dataset training the neural network. The semantic segmentation neural network is constructed based on the improved U-Net structure. The experimental results show that after training, the mean Intersection over Union (IoU) index of the test set results is 0.789, and manual recognition takes about 18 times longer than deep learning.
Keywords: Laser cladding, primary dendrite, deep learning, semantic segmentation, additive manufacturing (AM), intelligent manufacturing