Hydraulic System Fault Diagnosis Based on MSIF-ECACNN
To address the complex and difficulty to identify hydraulic signals accurately,an effective channel attention convolutional neural network model based on multi-sensor information fusion was proposed to diagnose the pump and accumulator in hydraulic system respectively.In this model,a parallel network structure was adopted.According to the differences in quantity and sampling frequency be-tween flow sensors and pressure sensors,as well as the different characteristics of flow and pressure signal failures,multiple pressure and flow sensor signals were input into 1D multi-channel convolutional neural networks with different convolution kernel sizes,and the effec-tive channel attention was used to adjust the weight of feature channels.Finally,feature fusion was carried out in the full connection lay-er,and classification was realized by Softmax layer.The results show that effective channel attention can effectively improve the accuracy of fault recognition.Compared with advanced research methods in this field,this method has better fault diagnosis performance,with an accuracy of 99.52％for accumulator fault diagnosis and 99.88％for hydraulic pump fault diagnosis.At the same time,it solves the problem of difficulty in accurately identifying faults caused by differences in the number and sampling frequency of non homologous sen-sors.
multi-sensor information fusionconvolutional neural networkefficient channel attentionhydraulic systemfault diag-nosis
万方数据
维普
