Research On Fault Diagnosis Of Sucker Rod Pumping System Based On Analysis Of Dynamometer Card

The method of sucker-rod pumping accounts for a large proportion of the oil production. A majority of pumping units are dispersed in wildness area which has terrible environmental conditions. Moreover, the working condition of pumping units beneath the well is complex, so it is difficult to diagnose and handle the faults of pumping units in time. It affects the output and economic benefit of oil field severely. Hence, it is important to analyze the working condition of pumping units timely for improving the productivity of oil extraction and the economic profit. In the process of oil field production, dynamometer card is used to analyze the working condition of pumping units as the principal reference. And the feature extraction of dynamometer card is key process to analyze the working condition and diagnose faults. The content of this thesis shows as follows.(1) This paper illustrates the problems of fault diagnosis to be solved. Based on set theory model, it analyzes the mapping relation between dynamometer card set and fault set in the diagnosis. Then, it splits the whole process into comparatively simple parts. Therefore, the problem is simplified and clear.(2) This paper introduces the development and main method of fault diagnosis and the operation principle of pumping unit. After that, it analyzes the forming process of a dynamometer card, the graphic features of the typical fault dynamometer cards and the reasons of them.(3) This paper puts forward a feature extraction method of transforming the natural language into computer language. In this way, the expertise and relative knowledge can be used. Furthermore, its feature vector is easy to understand, and the values are mostly uncorrelated. As a result, it will simplify the next analysis and the processes.(4) Based on the extraction method, this paper discusses the classification method and uses a method combining the ideologies of the K-nearest and Optimal Hyper Plane to classify some samples. Thanks to the feature extraction method used above, the classification experiment used the one-dimensional classification method has gotten a good result.(5) In order to process the unlabelled dynamometer card set, this paper discusses the unsupervised learning method based on the extraction method. After that, it uses the K-means method and the density-based method to cluster some samples.