Recognition of Primary Dendrite Spacing in Laser Cladding Based on Deep Learning
S.R. Guo, K.X. Wang, L.J. Cui, X.L. Li, H.Y. Li, W.H. Zeng and Y.L. Cao
Measurement of primary dendrite spacing is very important in the morphology of the laser cladding samples. In this work a new automatic monitoring method based on the deep learning is applied to the measurement of primary dendrite spacing in laser cladding metallographic maps and the primary spacing network based on Mask R-CNN deep learning model is built. According to the returned boundary box and mask optimization model parameters, the test set is used to detect the network training results. The results show that the average absolute error of the test set is 0.691 μm, and the accuracy of the boundary frame is 0.967. This study is of great significance for the subsequent research on the relationship between microstructure and coating performance. The network output results can be fed back and guide the subsequent laser cladding experiment parameters, so as to obtain a suitable primary dendrite spacing.
Keywords: Fibre laser, laser cladding, primary dendrite spacing, deep learning, Mask R-CNN, metallographic diagram