In order to solve the problems that the bearing of the cutting part of the shearer is prone to failure in complex environments and the practical application effect of the existing fault diagnosis model is not good, a fault diagnosis method for the bearing in the cutting part of the shearer based on the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and the improved Support Vector Machines (SVM) was proposed. Then, the energy features of the reconstructed IMF components are extracted, and a high-dimensional feature matrix is formed with the time-domain frequency domain features extracted from the reconstructed signal, and the PCA dimensionality reduction algorithm is used to reduce the dimensionality. Finally, the improved Support Vector Machine (SVM) classification model is used to diagnose and identify the faults of the low-dimensional feature matrix, and compared with a variety of mainstream classification algorithms. The training results show that the proposed method has a fault diagnosis accuracy of 99.3%, which is 3.9%, 1.1% and 1.7% higher than that of SVM, PSO-SVM and GA-SVM, respectively, and still has a classification accuracy of 95.2% under the noise condition, which is 8.9%, 3.9% and 3.1% higher than the other three classification models, respectively, and the convergence speed is faster. It has a classification accuracy of 94.7% in practical engineering applications, which can effectively improve the intelligence of coal mines.