Layered Rod Pumping System Downhole Fault Diagnosis Technology

A sucker-rod pumping system of double-layer separate recovery is developed in order to solve radically the interbedded disturb problem in the Daqirg oi;-field. The system has realized the synchronous recovery of double-layer and erihanced the pumping efficiency. It is must to research the fault diagnosis technique in order to insure the popular application of the sucker-rod pumping system of double-layer separate recovers; consequently the work state of the pumping system can be found out well and truly, so that the pumping system can be debugged and maintained suitably. In the paper, the theory and application of neural-network-based fault diagnosis technique are firstly studied, and the various problems about down-hole fault diagnosis expert system of the sucker-rod pumping system of double-layer separate recovery are discussed.In the paper, the topological structure and algorithm of the BP neural nets are discussed The shortcomings of the BP algorithm are summed up. The various influencing factors of the generalization capability of the BP neura! nets are analyzed emphatically. It is explained how to gained the top-quality and representative samples.Along with the sucker-rod pumping system of double-layer separate recovery is applied popularly, its static and dynamic behavior need analyzing, and its prediction model needs to establish. In the paper, an explicit difference method has been proposed in order to solve the prediction model, and then the mechanical state of each section of sucker rod string can be gained, the dynagrams can be drawn. These provide theory basis for design, parameters selection and fault diagnosis of the pumping system.The key to the down-hole fault diagnosis of the pumping system lies in obtaining the down-hole pump dynagrams. In this paper, a numerical analysis method is given by combining iterative algorithm with finite difference method, the algorithm may be used to solve the diagnosis model of the pumping system, gain dynagrams of the eccentric pump and the tube pump, so that the neural nets is applied to recognize the dynagrams. A pretreatment method of MATLAB-Toolbox-based for ground dynagrams is proposed in order to realize theintelligentization of fault diagnosis of the sucker-rod pumping system of double-layer separate recovery in the paper. The method can realize to recognize the real dynagrams by a computer and gain two boundary conditions that are applied to solve the diagnosis model. One condition is relation between load and time, other condition is relation between displacement and ime.In the paper, the down-hole fault diagnosis expert system of the sucker-rod pumping system of double-layer separate recovery is established, it consist of knowledge pretreatmer.t module, neural nets diagnosis module of the eccentric pump, neural nets didgnosis module of the tube pump, knowledge after treatment module, system coni.ro] module and accessories. The BP neural nets has been built with the MAILAB-Toolbox, and trained with the improved BP algorithm that has self-adapting leam rate and momentum factor. The course to initialize, train and test for neural network diagnosis modules has been illustrated in detail. A unifying method is proposed, it is used to produce the initialization data documents of the neural nets with typical work state patterns. Finally, the fault diagnosis expert system software has been designed.