Development of Oilwell Fault Classifiers Using a Wavelet-Based Multivariable Approach in a Modular Architecture

Abstract

Summary Fault detection and diagnosis are fundamental problems in the process of abnormal event detection in oil wells. This paper describes an open-source modular system that enables the efficient design of fault detectors and classifiers based on machine learning techniques. Events considered in this work are part of the publicly available 3W database developed by Petrobras, the Brazilian oil holding. Seven fault classes are considered, with distinct dynamics and patterns, as well as several instances of normal operation. We also show the effectiveness of the use of wavelet-based features, which provide multiscale time-frequency analysis, targeting a more realistic event modeling. A few challenges imposed by the 3W data set are addressed by combining both wavelet and statistical features, resulting in more accurate and more robust classifiers, with a 98.6% balanced accuracy in the multiclass problem, a significant improvement over the 94.2% previously reported in the literature.