采用分数阶域MFL-Net的机械智能故障诊断方法研究

Study on Intelligent Fault Diagnosis of Machinery Using MFL-Net in Fractional Domain Method

提出了一种分数阶域多尺度特征卷积神经网络的智能诊断方法。首先,对原始振动信号进行分数阶傅里叶变换(FRFT),获得多个分数阶次下振动数据的时频特征;其次,构建具有多尺度特征学习模块的轻量级卷积神经网络(MFL-Net),进一步从分数阶域时频特征中提取故障信息,通过训练获得诊断模型并应用于故障识别;最后,通过离心泵和滚动轴承故障数据集对所提方法的有效性进行验证。结果表明:所提方法可以有效提取非平稳信号中的故障特征,并实现故障的准确诊断。

A novel intelligent fault diagnosis method was proposed based on multi-scale feature learning convolutional neural network(MFL-Net)and fractional Fourier transform(FRFT).First,FRFT was applied to original vibration signal for obtaining fractional domain feature data under multiple fractional orders.Then,the MFL-Net with multi-scale feature learning modules was constructed to further extract fault features from fractional domain data,and the fault type can be diagnosed by the model.Finally,the effectiveness of the proposed method was verified by fault datasets of centrifugal pump and rolling bearing.Results show that the proposed method can effectively extract the fault features from non-stationary signals and thus identify faults accurately.