As one of the most vital parts of rotating equipment, it is an essential work to diagnose rolling bearing failure. The traditional signal processing-based rolling bearing fault diagnosis algorithms rely on artificial feature extraction and expert knowledge. The working condition of rolling bearings is complex and changeable, so the traditional algorithm is slightly lacking adaptability. The damage degree also plays a crucial role in fault monitoring. Different damage degrees may take different remedial measures, but traditional fault-diagnosis algorithms roughly divide the damage degree into several categories, which do not correspond to the continuous value of the damage degree. To solve the abovementioned two problems, this paper proposes a fault-diagnosis algorithm based on “end-to-end” one-dimensional convolutional neural network. The one-dimensional convolution kernel and the pooling layer are directly applied to the original time domain signal. Feature extraction and classifier are merged together, and the extracted features are used to judge the damage degree at the same time. Then, the generalization ability of the model is studied under a variety of conditions. Experiments show that the algorithm can achieve more than 99% accuracy and can accurately give the damage degree of the bearing. It has good performance under different speeds, different types of motors, and different sampling frequencies, and so it has good generalization ability.

中文翻译：

---

多任务卷积神经网络在滚动轴承故障识别中的应用

作为旋转设备最重要的部件之一，诊断滚动轴承故障是一项必不可少的工作。基于信号处理的传统滚动轴承故障诊断算法依靠人工特征提取和专家知识。滚动轴承的工作条件复杂多变，因此传统算法略有不足。损坏程度在故障监控中也起着至关重要的作用。不同的损坏程度可能会采取不同的补救措施，但是传统的故障诊断算法将损坏程度大致分为几类，这些类别不对应于损坏程度的连续值。针对上述两个问题，提出了一种基于“端到端”一维卷积神经网络的故障诊断算法。一维卷积核和池化层直接应用于原始时域信号。将特征提取和分类器合并在一起，并使用提取的特征同时判断损伤程度。然后，在各种条件下研究模型的泛化能力。实验表明，该算法可以达到99％以上的精度，并能准确给出轴承的损伤程度。在不同的速度，不同的电动机类型和不同的采样频率下具有良好的性能，因此具有良好的泛化能力。提取的特征可同时判断损伤程度。然后，在各种条件下研究模型的泛化能力。实验表明，该算法可以达到99％以上的精度，并能准确给出轴承的损伤程度。在不同的速度，不同的电动机类型和不同的采样频率下具有良好的性能，因此具有良好的泛化能力。提取的特征可同时判断损伤程度。然后，在各种条件下研究模型的泛化能力。实验表明，该算法可以达到99％以上的精度，并能准确给出轴承的损伤程度。在不同的速度，不同的电动机类型和不同的采样频率下具有良好的性能，因此具有良好的泛化能力。