Deep feature learning network for fault detection and isolation

Abstract

Prognostics and Health Management (PHM) approaches typicallyinvolve several signal processing and feature engineeringsteps. The state of the art on feature engineering, comprisingfeature extraction and feature dimensionality reduction,often only provides specific solutions for specific problems,but rarely supports transferability or generalization: itoften requires expert knowledge and extensive intervention.In this paper, we propose a new integrated feature learningapproach for jointly achieving fault detection and fault isolationin high-dimensional condition monitoring data. Theproposed approach, based on Hierarchical Extreme LearningMachines (HELM) demonstrates a good ability to detectand isolate faults in large datasets comprising signals of differentnatures, non-informative signals, non-linear relationshipsand noise. The method includes stacked auto-encodersthat are able to learn the underlying high-level features, and aone-class classifier to combine the learned features in an indicatorthat represents the deviation from the normal systembehavior. Once a deviation is identified, features are used toisolate the most deviating signal components. Two case studieshighlight the benefits of the approach: First, a syntheticdataset with the typical characteristics of condition monitoringdata and different types of faults is applied to evaluate theperformance with objective metrics. Second, the approach istested on data stemming from a power plant generator interturnfailure. In both cases, the results are compared to othercommonly applied approaches for fault isolation.