基于卷积神经网络的化工旋转机械故障识别

Fault recognition of chemical rotating machinery based on convolutional neural network

针对化工旋转机械故障识别困难的问题,以旋转机械离心泵为例,通过分析二维卷积神经网络存在的问题,对卷积神经网络故障识别模型进行优化。首先利用小波变换等方法对机械振动信号进行特征提取,得到显著特征集,再利用降维方法去除不必要的环境噪音等冗余特征,使特征信息更容易被卷积神经网络识别;然后构建基于一维卷积神经网络的旋转机械故障识别模型,进行特征识别;最后,通过仿真试验,对上述改进后的故障识别模型进行验证。结果表明,通过特征提取优化和降维优化的一维卷积神经网络模型,训练时间极短,模型构建快速,同时具有更高的旋转机械故障识别准确率,证明改进后的旋转机械故障识别模型具有很高的可行性。

Aiming at the difficulty of fault recognition of chemical rotating machinery,taking centrifugal pump of rotating machinery as an example,the fault recgnition model of convolutional neural network was optimized by analyzing the existing problems of two-dimensional convolutional neural network.Firstly,wavelet transform and other methods were used to extract features from mechanical vibration signals,thus the significant feature sets were obtain.Then,dimensionality reduction method was used to remove unnecessary redundant features such as environmental noise,making feature information easier to be recognized by convolutional neural network.Moreover,a fault recognition model of rotating machinery based on one-dimensional convolutional neural network was constructed for feature recognition.Finally,the improved fault recognition model is verified by simulation.The results show that the one-dimensional convolutional neural network model optimized by feature extraction and dimensionality reduction has very short training time,fast model construction,and higher fault recognition accuracy of rotating machinery,which proves that the improved fault recognition model of rotating machinery has high feasibility.