Fault diagnosis for down-hole conditions in beam pumping units based on an improved fuzzy Iterative Self-Organizing Data Analysis Technique

Abstract

Dynamometer cards are commonly used to analyze down-hole conditions of beam pumping units in practical oil production. In the literature, supervised learning based methods heavily rely on training samples. In order to realize unsupervised learning of fault diagnosis for down-hole conditions, a method based on an improved fuzzy Iterative Self-Organizing Data Analysis Technique (ISODATA) with “merging” and “splitting” mechanisms is proposed in this paper. The Hsim similarity function is used to replace the Euclidean distance to improve the classification accuracy in high-dimensional space. In “merging” and “splitting” mechanisms, the “minimum distance between two classes” is a very important parameter to affect clustering accuracy and is difficult to be accurately set in advance. It is considered as a variable parameter and is dynamically adjusted in the clustering process. Simulated annealing algorithm is used to realize optimization and Xie-Beni (XB) validity index is used as the optimization target. An example is given to illustrate that the proposed method can realize dynamic clustering with a better effectiveness.