A novel fault diagnosis method based on improved multiscale range entropy and hierarchical prototype (HP) is proposed in this paper. Firstly, considering that range entropy cannot analyze the complexity of time series from multiple perspectives, the coarse-grained process is combined with range entropy. In addition, to make the coarse-grained process more comprehensive, the selection of its starting point is improved. Secondly, to extract more feature information, the dimension reduction of eigenvectors is carried out by using singular value decomposition. Finally, HP is trained with the eigenvectors and its performance is tested. To test the performance of the proposed fault diagnosis method, testing bearing vibration signals collected by sensors from Case Western Reserve University and Southeast University are used for experimental analysis in this paper, and the experimental results show high accuracy of the proposed fault diagnosis method. To verify the suitability of the improvement proposal, the superiority in feature extraction ability and the classification capability of the classifier, the proposed fault diagnosis method is compared with another seven fault diagnosis methods. The results show that the proposed fault diagnosis method has the highest fault diagnosis accuracy.

提出了一种基于改进的多尺度范围熵和层次原型（HP）的故障诊断方法。首先，考虑到范围熵不能从多个角度分析时间序列的复杂性，将粗粒度过程与范围熵相结合。此外，为了使粗粒度过程更加全面，改进了其起点的选择。其次，为了提取更多的特征信息，利用奇异值分解对特征向量进行降维。最后，用特征向量训练 HP 并测试其性能。为了测试所提出的故障诊断方法的性能，本文采用凯斯西储大学和东南大学传感器采集的轴承振动信号进行试验分析，实验结果表明所提出的故障诊断方法具有较高的准确性。为了验证改进方案的适用性、特征提取能力的优越性和分类器的分类能力，将所提出的故障诊断方法与另外七种故障诊断方法进行了比较。结果表明，所提出的故障诊断方法具有最高的故障诊断准确率。将所提出的故障诊断方法与另外七种故障诊断方法进行了比较。结果表明，所提出的故障诊断方法具有最高的故障诊断准确率。将所提出的故障诊断方法与另外七种故障诊断方法进行了比较。结果表明，所提出的故障诊断方法具有最高的故障诊断准确率。