Health Assessment For Plunger Pump With Small Sample Based On Deep Learning And Transfer Learning

The plunger pump is a power element of the hydraulic system,and the healthy state of the plunger pump will directly affect the working state of the hydraulic system.Therefore,it is necessary to assess the health status of the plunger pump,monitor the working state of the equipment in real time,and grasp the working state of the plunger pump.Traditional machine learning methods have got great achievements in health assessment for plunger pump.But these methods are mainly applied in fixed working conditions.Aiming at the problem of how to set up a prediction model for the health assessment of plunger pump,there are two ways to solve this problem.Firstly,a deep transfer learning method for health assessment is proposed.First,transform raw time-domain data to frequency-domain feature.Next,deep transfer learning method based on TrAdaboost algorithm is applied to improve accuracy for health assessment.TrAdaboost algorithm uses several weak classifier to constitute a strong classifier and convolutional neural network is used to train each weak classifier.The accuracy of the method is verified by experiments,and a good result is achieved.Secondly,a generative adversarial networks based transfer learning scheme for health assessment is proposed.The model takes generative adversarial networks(GAN)to provide a large number of plausible labeled data by simulating each type of real fault data.Then we apply CNN to train a deep learning prediction model for the large number of labeled simulated data.Finally,we use transfer learning to fine tune the CNN model with real labeled data and get the final deep prediction model for limited real data.Experiments show that compared with the basic convolutional neural network,the prediction accuracy is greatly improved.