【摘要】 针对当前基于机器学习的滚动轴承损伤尺寸识别精度低的问题，提出了一种基于深度残差网络的滚动轴承故障损伤尺寸识别方法。该方法以残差网络为深层特征提取主框架，建立能够将预处理后的振动样本数据映射至相应的损伤尺寸的网络模型。数据预处理时采用傅里叶变换，获得原始振动加速度时域信号的频谱图作为网络的输入；在滚动轴承加速疲劳试验机和航空发动机转子试验器等两种类型的试验器上，通过不同的损伤尺寸试验，对所提出的损伤尺寸识别方法进行了验证，并和多种方法进行了对比。结果表明：在0.3 mm的预测误差范围内，所搭建的网络模型对未参与训练的故障损伤尺寸的识别精度，内圈为91.2%，外圈为97.9%。同时，在对数据进行加噪处理后，损伤尺寸的预测误差依然能够达到0.3 mm以内的识别精度。结果充分表明该方法具有很强的故障损伤尺寸识别能力。更多还原

【Abstract】 In order to solve the problem of low accuracy of damage size identification of rolling bearing based on machine learning,a method of fault damage size identification of rolling bearing based on deep residual network is proposed. This method takes residual network as the main frame of deep feature extraction, and establishes a network model that can map the preprocessed vibration sample data to the corresponding damage size. Fourier transform is used in data preprocessing to obtain the spectrum of the original vibration acceleration time-domain signal as the input of the network. The proposed damage size identification method is verified by different damage size tests on two types of testers such as rolling bearing accelerated fatigue tester and aero-engine rotor tester, and compared with other methods. The results show that, within the prediction error range of 0.3 mm, the recognition accuracy of the network model for the fault damage size without training is 91.2% for the inner ring and 97.9% for the outer ring.At the same time, the prediction error of the damage size can still reach the recognition accuracy within 0.3 mm after the data is processed with noise. The results fully show that this method has a strong ability to identify fault damage size.更多还原