Deep Learning Based Fault Detection of Natural Gas Pumping Unit

Natural gas pumping unit is a very difficult object for diagnosis. A lot of combinations of technical equipment, different operational conditions, and other factors require design and implementation of reliable diagnosis methods. Acoustic signal based fault diagnosis of natural gas pumping units is well known and widely used in a number of applications. Statistical modeling and frequency analysis are among the most popular. In this paper, we share our experience in the use of the classification model based on an artificial multilayered dense feed forward neural network and a deep learning approach for software-implemented diagnosis of a GTK-25-i type of pumping unit. The model predicts three states of the unit: “nominal”, “normal”, and “faulty”. It this research we used combination of vibration signal and acoustic emission signal as features for neural network model. In this paper we present the descriptive statistics, time and frequency domains analysis of vibration and acoustic emission signals. We present the developed structure of input data pipeline and the architecture of the deep neural network. The paper shows the detailed training, validation, and test class-wise metrics. Also we present the final classification performance as F1-score for each of three classes and compare results with other well-known approaches. The paper reports the overall accuracy of 0.9864 and minimum F1-score of 0.8113. This is competitive compared to the latest industry research findings.