Abstract: Aiming at the limitation and instability of the single detection method among the three non-destructive detection methods of eddy current detection, magnetic flux leakage detection and Barkhausen detection, a CART decision tree data fusion algorithm based on supervised learning is proposed to grind typical work pieces. Burns are evaluated. Experiments are performed on the sample data of 946 tooth surface detection signals of a straight tooth work piece. 758 sets of data are used as training samples to establish a model after pruning. The remaining 188 sets of data are evaluated. The results show that the prediction accuracy is 99.5%. It shows that the CART decision tree algorithm based on supervised learning has high recognition accuracy and provides new ideas for the electromagnetic non-destructive evaluation of gear grinding burns.