Abstract
In the field of modern petroleum industry, fault diagnosis and classification of pumping units are among the most crucial research topics. The conventional diagnostic method involves installing sensors on the pumping unit to analyze the dynamometer cards(DCs). However, due to the sensor cannot work stably for a long time and is easily damaged, a more stable and effective data source is required for diagnostics. Compared with the DCs, the parameters of the driving motor of the pumping unit can also effectively reflect the running state of the pumping unit. And the electrical parameters can be easily collected for a long time. However, due to the difficulty in obtaining sufficient fault electrical parameter samples, intelligent methods find it difficult to learn the characteristics and patterns of electrical parameter data. Therefore, this paper proposes a new method for fault diagnosis. Firstly, establish a mechanism model to convert the existing faulty DC into electrical parameter data. Subsequently, an improved Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS) method was used to make multiple predictions of the fault state to expand the dataset. Finally, use OMNI-SCALE Convolutional Neural Network (OS-CNN) for fault diagnosis. According to experimental results, this proposed method effectively addresses insufficient electrical parameter data issues and achieves superior classification performance.