Fault diagnosis of pumping system based on multimodal attention learning (CBMA Learning)

Abstract

Grasping the pumping system's operational state in time is crucial to the petroleum extraction sector. Rapid and efficient identification of indicator diagrams is a direct method of determining the operating status of oil wells. A new method of multimodal attention learning model(CBMA Learning) using superimposed images of the indicator diagrams and its frequency domain maps combined with numerical data of the indicator diagram is proposed. By processing the specified type of data through each branch network of the model in this method, the model could extract the characteristics of various modal data, and the attention mechanism introduced can efficiently capture the detailed information of the data and increase the precision of fault diagnosis. The proposed method is validated by applying it to an actual dataset from an oil field in northwest China. The experimental results demonstrate that the average classification accuracy of CBMA Learning is increased by 1.17 percent compared to the simple multimodal learning model, and the average accuracy of CBMA Learning reaches 93.05 percent, which is 10.47 percent higher than other models.