Research On Fault Diagnosis For Submersible Reciprocating Pump Based On Deep Learning Network

At present,the oil is still an extremely important energy sources in the world,how to mine the crude oil efficiently and reliably has always been the focus of researchers at home and abroad.In order to overcome the shortcomings of the traditional oil extractor,people propose the submersible reciprocating pump.Although the submersible reciprocating pump is easy to control and its work efficiency is high,its application scale was greatly limited because that it was put forward in the short time and its supporting technology development is not perfect.It has an very important significance to promote the submersible reciprocating pump's own advantages and widely used in oil field so as to improve the oil production efficiency.Thus,through the depth analysis of the structure and working principle of submersible reciprocating pump,fault diagnosis methods that are suitable for the submersible reciprocating pump are put forward.First of all,since the submersible reciprocating pump is a rodless device,it is impossible to diagnose its failure by using the conventional ground dynamometer method based on the relationship between load and displacement.However,the submersible reciprocating pump's drive motor is directly connected to the pump through the plunger,that is,the motor's power output is directly on the load,so the motor current data values can indirectly reflect the size of the load.Drawing on the principle of ground dynamometer,the current of submersible motor is selected as the characteristic information to determine the submersible reciprocating pump's fault status.Secondly,since the traditional fault diagnosis methods need to be combined with certain feature extraction method to achieve classification and identification and these feature extraction methods often contain human experience in different degrees,they will reduce the diagnosis model's effect in a certain extent.In order to avoid the impact of human factors on the original fault data,a fault diagnosis model of deep belief network(DBN)is proposed.Thirdly,A DBN fault diagnosis model based on the motor current for submersible reciprocating pump is established.Simulate different fault states on the experimental platform,collect and record the the motor operating current value so as to build a sample library.Using DBN fault diagnosis model with multiple hidden layers to diagnose six typical oil well faults which occurred in submersible reciprocating pumping unit,and the effectiveness of the proposed method is verified by comparison with the traditional fault diagnosis method.Finally,limited to the fault diagnosis method based on the current only can realize the identification of part of the fault states of the submersible reciprocating pump well,a fault diagnosis method based on the downhole multi-parameter acquisition device and DBN network for submersible reciprocating pumping unit is proposed.The seven relevant parameters of the underground unit monitored by the system are combined with the motor current measured on the ground as the characteristic information for determining the operating status of the submersible reciprocating pump well.The effectiveness of the method is verified through the experimental analysis.Therefore,the fault diagnosis method based on multi-parameters can determine the operating status of the submersible reciprocating pump unit in an all-round way and ensure its normal long-term reliable operation.