利用改进CNN实现轴承自适应抗噪故障诊断

Adaptive Anti-Noise Fault Diagnosis Algorithm for Bearings Based on an Improved Convolution Neural Network

近年来对轴承诊断方法的研究表明,大多数方法都是将特征提取和故障识别相结合。但是,这些方法对专业知识有较高的要求,同时需要花费大量的时间开发诊断算法,且算法的性能会受到工况变化、噪声等因素的影响。为了解决上述问题,这里提出了一种信号的预处理方法,可以将一维振动信号转为二维灰度图,以及自适应的抗噪卷积神经网络(AA-CNN)模型。该模型,无须解析机械设备内在故障机理,并且将特征提取过程与故障分类过程有机结合,实现故障诊断。为了保证结果的通用性,利用公开的轴承数据集对模型进行验证,结果表明:该故障诊断模型的识别准确率保持在99%以上,并且在含噪声的时域信号中也能准确识别出故障种类和程度,有良好的适应性。

According to the research of bearing diagnosis methods in recent years,most of existing methods are developed by combining feature extraction and fault recognition.However,these fault diagnosis algorithms require expert experience and time-consuming to design.Moreover,the performance of the algorithm will be affected by working conditions,signal noise,and other factors.In order to solve the above problems,it proposes a Signal-to-Image conversion method and a fault diagnosis method,which based on the convolutional neural network(CNN)structure.The model does not need to analyze the intrinsic fault mechanism of mechanical equipment,which organically combines the feature extraction process and the fault classification process to realize fault diagnosis.Experiments on the bearing data set show that the proposed method can not only achieve more than 99%fault recognition accuracy,but also obtain good adaptability under noisy signal.