Abstract: Aiming at the problem that it is difficult to accurately predict the crack defects in wide-band laser cladding, a BP neural network crack defect prediction model was established with scanning speed, overlap ratio, laser power as the input and crack density of cladding sample as the output. The weights and thresholds of BP neural network were optimized by genetic algorithm, and the relative errors before and after optimization were compared and analyzed. The results show that the relative error of GA-BP neural network model is between 0.22%-2.10%; the relative error of BP neural network model is between 2.09%-14.31%, and the prediction accuracy of GA-BP neural network model is much higher than that of BP neural network model.