Fault Diagnosis Algorithm of Beam Pumping Unit Based on Transfer Learning and DenseNet Model

Abstract

The difficulty of collecting fault data samples is one of the application problems of the deep learning method in fault diagnosis of mechanical production; the second is that when the depth of the learning network increases, the network accuracy is saturated or even decreased. Therefore, based on the deep learning algorithm and the DenseNet model, this paper establishes a fault diagnosis model for the beam pumping unit through the transfer learning method. The model uses the global pooling layer as the classifier. The model is used to classify and test various working conditions such as wax deposition, pump leakage, insufficient liquid supply, and pump leakage in oil wells. The results show that the model can obtain a classification model with high accuracy by learning a limited number of sample data; in the case of uneven data samples, the model can also basically complete the task classification task accurately. Through the evaluation of the test set, the model has an average accuracy of more than 95% in identifying various working conditions.