Optimization of Mechanical Pump Units from Character Recognition through Self-Organized Maps

Abstract

This paper presents an alternative for identification and prediction, qualitative and quantitative, of the problems in the equipment subsoil system rod pumping, reflected in the downhole dynamometer charts to establish a methodology for data analysis of 512 dynamometer charts of 8 wells the northern región of Huila, provided by Ecopetrol. After making a collection and manual classification of dynamometer charts, and to form a database, if the problem identified was suitable to be solved by a neural network. Subsequently clarifying the method for extracting the most significant characteristics of each diagram by a signal processing method called Wavelet Transform.

Was designed and tested different algonthms transfonn to the conesponding coefficients of approximations and details of the signals, and then apply them to the training of a neural network of self-organizing maps (SOM), capable of performing his own interpretation was faster and more assertive. The research results were adjusted to final software also provides identification, prediction swiftty and accurately to problems in mechanical pumping equipment in order to minimize operating costs and maximize production in the industry.