Abstract: In the fault diagnosis of agricultural machinery rolling bearings, aiming at the nonlinear and non‑stationary characteristics of bearing vibration signals and the not obvious fault characteristics. This paper proposes a novel method based on Continuous Wavelet Transform (CWT), Convolutional Neural Network (CNN) and Support Vector Machine (SVM) of rolling bearing fault diagnosis (CWT—CNN—SVM). Firstly, multi‑scale time‑frequency analysis of rolling bearing vibration signals is carried out using CWT to provide more detailed characteristics for subsequent fault diagnosis. Secondly, the extracted time‑frequency graph is used as input to learn the fault feature information by CNN. Then, SVM is used to classify the output results to achieve accurate fault type identification. Compared with CNN, CWT—CNN and CWT—ResNet, the test results show that CWT—CNN—SVM has the highest fault diagnosis accuracy, with a single accuracy of 100% and a 5‑time repeated accuracy of 99.62%. CWT—CNN—SVM not only achieves diagnostic accuracy when dealing with complex rolling bearing fault diagnosis problems, but also shows the advantages of combining deep learning with fault diagnosis, which can further improve the performance of small data sets. The CWT—CNN—SVM method proposed in this paper has beneficial theoretical value and practical application prospect for improving the fault diagnosis performance of agricultural machinery rolling bearings.

Key words: fault diagnosis, agricultural machinery, rolling bearing, continuous wavelet transform, CNN, support vector machine

摘要： 针对农用机械滚动轴承故障诊断中轴承振动信号非线性、非平稳特性以及故障特征表征不明显的问题，提出一种基于连续小波变换(CWT)、卷积神经网络(CNN)和支持向量机(SVM)的滚动轴承故障诊断方法(CWT—CNN—SVM)。首先，利用CWT对滚动轴承振动信号进行多尺度时频分析，为后续故障诊断提供更详细的特征；然后，将提取到的时频图作为输入，利用CNN深层次学习故障特征信息；最后，采用SVM对输出结果进行分类，以实现精确的故障类型识别。与BPNN、SVM、CWT—CNN以及CWT—ResNet等方法比较，试验结果表明，CWT—CNN—SVM故障诊断准确率最高，单次准确率达到100%，5次重复试验准确率为99.62%。CWT—CNN—SVM在处理复杂的滚动轴承故障诊断问题时，不仅诊断准确，同时展现出深度学习与故障诊断相结合的优势，能进一步提升小数据集的性能。所提出的CWT—CNN—SVM方法对于提升农机滚动轴承故障诊断性能，具有一定的理论价值和实际应用前景。

关键词: 故障诊断, 农机, 滚动轴承, 连续小波变换, 卷积神经网络, 支持向量机