Study Of The Fault Diagnoses Of Flexible Wire Rope Sucker Rod Pumping Unit Based On Artificial Neural Network

The fault diagnoses of the pumping unit is an important part of the research in the field of petroleum production. It is of much importance to enhance the efficiency and automatization level of the whole pumping system. At present, The principle of this technology is to analyze and judge the condition of the pumping unit utilizing its dynagraphs. Based on the dynagraphs of ground and the mathematic model, was the downhole dynagraphs calculated.The computer image processing technology and pattern recogniting technology have achieved much progress in lots of fields and been paid much more attention as time going on. They have promoted the development of imitated intelligence.The development of artificial neural network(ANN) provides a valuable chance to change this situation. It is distinguished by parallel processing, self-study ability, high stability, accommodation and reflection ability and it can be calculated respectively. All of these virtues can providean new theory and method for fault diagnoses.The combination of the technology of the computer image process, pattern recognition and ANN have greatly improved the ability to recognize the dynagraphs of the whole pumping unit.Based on the technology of computer image processing, A method to execute the process of pattern recognition to the dynagraph is put forward in this thesis. The dynagraphs are dealt with a series of image processing, after they are achieved and converted to the proper vector which will be input to the ANN. A basis for the practical use of the ANN has been offered.The BP network is adopted based on the theory and experience of some researchers. The structure of the network is decided based on the practical situation of this thesis. A series of methods of dynamic alterable step and disposal of the specimens of are used. The structure of the network enhanced as well as the training efficiency of the network.A practical example by changing the training number to a dynagraph has been given. They have been showed the correctness of the choosing of the type of network and the error of training can satisfy the need.The image processing is designed for all the colorful images. For all the dynagraphs, they would deal with following processes: the grayscale transformation, smoothing, sharpening, unity of size, binary and plot detaching.