Fault detection for sucker rod pump based on motor power

Computer-aided fault detection for sucker rod pump is a crucial technology to monitor wells in the oil production. According to the detection results, engineers could take corresponding measures to ensure wells operating in a safe and productive state. Generally, the conventional approaches to address this problem are mainly based on dynamometer cards, but these methods have obvious defects in security risks and high maintenance cost in the application. Noteworthy, the motor is the energy resource of sucker rod pump and the change of motor’s power could reflect the variation of the working states. Therefore, in this paper, a novel method based on motor power for detecting the working state of the sucker rod pump is proposed. In this method, a set of the labeled motor power curve is essential. To obtain this vital information resource, the motor power curves are labeled by transforming them into dynamometer cards, which fully consider many crucial factors in this process. Moreover, to obtain useful information from motor power data, eight novel features are defined by analyzing the mechanism of motor work and the data distribution of the curve. Subsequently, the hidden Markov model (HMM), a probabilistic model with the double stochastic process, is employed to map the relationships between motor power data and working states. At last, the proposed method is verified experimentally using an oil dataset collected from oil field including six different working states, and then this technique is compared with some other methods. In the comparison, the proposed method gives 91.7% correct diagnosis that is higher than the 72.9% of SVM and the 62.5% of ANN. The experimental results show that the performance using the method proposed in this paper is satisfactory.