基于自适应MCKD与CNN的滚动轴承故障诊断

Fault Diagnosis of Rolling Bearing Based on Adaptive MCKD and CNN

为了解决强背景噪声下故障特征提取困难及传统方法依赖经验和知识的问题,提出了一种基于自适应最大相关峭度解卷积(MCKD)与卷积神经网络(CNN)的滚动轴承故障诊断方法。首先,利用粒子群算法(PSO)优化MCKD的参数。其次,对滚动轴承故障信号进行信号滤波,得到降噪后的信号。最后,将降噪后的信号输入到构建的CNN模型中进行训练和测试,得到轴承故障诊断的分类结果。通过轴承寿命试验台的故障数据集的测试和评价,将提出的方法与未经过降噪的CNN方法进行比较,验证了该方法具有较高的诊断精度。

In order to solve the problem of the difficulty of extracting fault features under strong background noise and the tradi⁃tional methods relying on experience and knowledge,a rolling bearing fault diagnosis method based on adaptive maximum corre⁃lation kurtosis deconvolution(MCKD)and convolutional neural network(CNN)is proposed.Firstly,particle swarm optimiza⁃tion(PSO)is used to optimize the parameters of MCKD.Secondly,the rolling bearing fault signal is filtered to get the denoising signal.Finally,the de-noising signal is input into the constructed CNN model for training and testing,and the classification re⁃sult of bearing fault diagnosis is obtained.Through the test and evaluation of the fault data set of the bearing life test rig,the pro⁃posed method is compared with the CNN method without noise reduction,and it is verified that the method has high diagnostic accuracy.