Research On Fault Diagnosis Of Hydraulic Pump Based On Simulation Data And Deep Transfer Learning

Hydraulic pump is the power source of the hydraulic system,is the key parts of many high-end engineering machinery,numerical control equipment.Hydraulic pump failure will affect product quality,property and personnel safety,so accurate and reliable diagnosis of hydraulic pump health state is very important.In recent years,with the development of artificial intelligence,deep learning has been widely applied in the field of fault diagnosis and achieved good results.However,the training of deep learning model requires a large number of fault samples,and the training samples and test samples are required to have the same probability distribution.However,it is difficult to meet the above two conditions in practical engineering.Insufficient data set samples will lead to small sample classification problem,and different data set sample distribution will lead to unbalanced data set classification problem,both of which will lead to the decline of model diagnosis performance.Aiming at the problem of hydraulic pump fault diagnosis with limited sample set and unbalanced data,this paper adopts one-dimensional convolutional neural network to propose a fault diagnosis model based on simulation data and deep transfer learning,so as to realize fault diagnosis of hydraulic pump with small sample and unbalanced data.The main contents of this paper are as follows:Firstly,the basic theory of deep transfer learning is introduced,including the structure and basic principle of convolutional neural network,the basic concept of transfer learning,and the calculation formula of transfer component analysis is given.Secondly,according to the hydraulic system condition monitoring test bed,the hydraulic system condition monitoring data set is obtained as the actual fault data set,and four kinds of hydraulic fault mechanisms are analyzed.Based on the hydraulic principle diagram of the test bed,a fault simulation model of the hydraulic system was proposed by combining AMESim and C language to obtain the simulated fault data of the hydraulic system under different health conditions.The difference between the simulated fault data and the actual fault data is analyzed to verify the rationality of the simulation model.Thirdly,a multi-scale one-dimensional convolutional neural network and multi-sensor information fusion deep neural network(MS1DCNN-MSIF)fault diagnosis model was proposed to solve the difficulties of complex and difficult diagnosis of hydraulic signals.On the basis of one-dimensional convolutional neural network,this model adopts multi-scale convolution strategy to extract fault features,adopts multi-sensor information fusion strategy to fuse multiple sensor signals for fault diagnosis,and uses Softmax for classification and recognition.Through the fault diagnosis experiment of hydraulic pump,the proposed model is compared with support vector machine,deep confidence network,stack self-coding network and other models under normal sample condition,and the accuracy and effectiveness of the proposed model are verified.Then the problems of small sample classification and unbalanced data set classification are simulated to show the difficulties in diagnosing these problems.Finally,a hydraulic pump fault diagnosis model based on simulation data and deep transfer learning is proposed,which combines the hydraulic system fault simulation model with the proposed neural network model to solve the problems of small sample classification and unbalanced data set classification.The method takes the simulation fault data set as the source domain and the actual fault data set as the target domain.Transfer component analysis(TCA)and maximum mean difference(MMD)were used to reduce the difference between source domain and target domain data.The deep learning model is used to learn fault feature knowledge in the source domain data set,and then the target domain data set is used to fine-tune the model to complete the fault diagnosis of the hydraulic pump.Finally,the proposed model is used for fault diagnosis of hydraulic pump under the condition of small sample and unbalanced data set respectively to verify the effectiveness of the proposed model.