A Deep Learning Approach for Condition-based Monitoring and Fault Diagnosis of Rod Pump System

Petrochemical industry is one of the key industry areas where Internet of Things (IoTs) and big data analytics could be widely applied to support smarter production and maintenance. In oil and gas exploitation, sucker-rod pumping systems are used in approximately 90 percent of artificially lifted wells. An automatic pipeline is crucial for real-time condition monitoring and fault detection of the system to save costs. Here we used convolutional neural network (CNN), a deep learning framework, to identify the working conditions of pump wells based on dynamometer cards and the corresponding sensor data. Two schemes, namely, data-based CNN and image-based CNN are proposed and compared with traditional machine learning algorithms such as k-Nearest Neighbors and Random Forests. Through experiments on a real dataset from oil fields, we show that CNN based approach could significantly outperform traditional methods without any need of manual feature engineering that requires domain expertise. Besides, we proposed a semi-automatic method for labeling big datasets of dynamometer cards, which could significantly reduce the labor work by manual labeling. Our work provides a feasible and efficient method for fault detection in oil pump systems and paves the way to applying deep learning techniques in IoTs related industries.