Oil well diagnosis by sensing terminal characteristics of the induction motor

Oil well diagnosis usually requires dedicated sensors placed on the surface and the bottom of the well. There is significant interest in identifying the characteristics of an oil well by using data from these sensors and neural networks for data processing. The purpose of this paper is to identify oil well parameters by measuring the terminal characteristics of the induction motor driving the pumpjack. Information about oil well properties is hidden in instantaneous power waveforms. The extraction of this information was done using neural networks. For the purpose of training neural networks, a complex model of the system, which included 25 differential equations, was developed. Successful application of neural networks was possible due to the proposed signal preprocessing which reduces thousands of measured data points into 20 scalar variables. The special input pattern transformation was used to enhance the power of the neural networks. Two training algorithms, originally developed by authors, were used in the learning process. The presented approach does not require special instrumentation and can be used on any oil well with a pump driven by an induction motor. The quality of the oil well could be monitored continuously and proper adjustments could be made. The approach may lead to significant savings in electrical energy, which is required to pump the oil