Submersible Screw Pump Fault Diagnosis Method Based on a Probabilistic Neural Network

Abstract

The core components of submersible screw pump are concentrated underground, so it is difficult to judge the fault timely and accurately. Based on this, this paper proposes a fault diagnosis method of submersible screw pump based on probabilistic neural network (PNN), and develops the corresponding fault diagnosis system. Wavelet packet theory is used to decompose and reconstruct the active power signal of submersible screw pump, extract the main fault information contained in the power signal, and construct the fault feature vector of submersible screw pump combined with the parameters such as output, oil pressure, casing pressure and dynamic liquid level. Based on the historical data of submersible screw pump, a probabilistic neural network model is constructed. The mapping relationship between fault feature vector and fault form is obtained through the historical data training model, so as to judge the fault form to be logged according to the fault feature vector to be logged. 142 groups of fault data are collected to train the model and verify the accuracy of the diagnosis model and the test shows that the accuracy of the diagnosis model is 90.5%. The test results show that the fault diagnosis method of submersible screw pump based on PNN can timely and accurately judge the fault type of submersible screw pump and reduce the economic loss of oilfield production.

Keywords

• fault diagnosis
• probabilistic neural network
• submersible screw pump
• wavelet packet