Contrastive Learning for Fault Detection and Diagnostics in the Context of Changing Operating Conditions and Novel Fault Types

Abstract

Reliable fault detection and diagnostics are crucial in order to ensure efficient operations in industrial assets. Data-driven solutions have shown great potential in various fields but pose many challenges in Prognostics and Health Management (PHM) applications: Changing external in-service factors and operating conditions cause variations in the condition monitoring (CM) data resulting in false alarms. Furthermore, novel types of faults can also cause variations in CM data. Since faults occur rarely in complex safety critical systems, a training dataset typically does not cover all possible fault types. To enable the detection of novel fault types, the models need to be sensitive to novel variations. Simultaneously, to decrease the false alarm rate, invariance to variations in CM data caused by changing operating conditions is required. We propose contrastive learning for the task of fault detection and diagnostics in the context of changing operating conditions and novel fault types. In particular, we evaluate how a feature representation trained by the triplet loss is suited to fault detection and diagnostics under the aforementioned conditions. We showcase that classification and clustering based on the learned feature representations are (1) invariant to changing operating conditions while also being (2) suited to the detection of novel fault types. Our evaluation is conducted on the bearing benchmark dataset provided by the Case Western Reserve University (CWRU).

Contrastive Feature Learning for Fault Detection and Diagnostics in Railway Applications

“…These can cause variations in the data that are larger than the variations expected due to the changes in the asset's health condition. If all possible variations are not sufficiently represented in the training dataset, they can harm the performance of data-driven fault detection and diagnostics models [4]. For example, stones on the sleeper can easily be mistaken for a spalling defect.…”

“…Feature learning, both supervised and unsupervised, has attracted a lot of attention in recent years [23,24]. The learned feature representation has been typically combined with classification models in supervised setups [4], clustering methods in unsupervised settings [25], or One-Class Classification (OCC) models in unsupervised settings without fault data [5]. Different types of autoencoders (AE) have been proposed for feature learning in unsupervised settings [5,25].…”

“…The objective of training any AE is to compress all information contained in the data, not distinguishing based on the type of information -whether it is descriptive of a health condition or not. It was demonstrated that AEs are not robust when applied to varying OCs [4]. Therefore, these approaches might not be suitable if the data sample is affected by large variations that are non-informative of health conditions.…”

“…Contrastive feature learning has been shown to achieve robustness against changing OC [4] and is, therefore, a promising feature learning paradigm. The learning objective is to group data with similar semantic meaning close to each other in the feature space and spread dissimilar data far apart [26].…”

Fuzzy-Logic-Based Recommendation System for Processing in Condition Monitoring

“…Other reviewed works were dedicated to the monitoring of bearings. In [ 11 , 12 ], the authors proposed a quadratic classifier; in [ 13 ], features were extracted using an autoencoder and applied for state assessment. Other approaches are depicted in [ 14 , 15 ], where the authors describe classifiers trained on the dataset with a low number of labels.…”