What question did this study set out to answer?

To compare various reservoir computing models for classifying bearing faults and identifying their severity levels.

February 16, 2026Scientific Reports0 citationsOpen Access

SA-ConSinGAN and reservoir computing fusion for accurate bearing fault classification and severity identification using GAF-based techniques

ASAnjil ShahGujarat National Law University VVVinay Vakharia YKYogesh KumarGujarat National Law University

Key Points

To compare various reservoir computing models for classifying bearing faults and identifying their severity levels.
Analyzed five reservoir computing models for fault classification and severity identification.
Applied three Gramian Angular Field methodologies to generate images of faulty conditions.
Employed SA-ConSinGAN to augment the bearing fault dataset.
Used ten-fold cross-validation to evaluate model accuracy.
RVFL+GADF achieved 100% accuracy in fault classifications.
RVFL+GADF combinations also reached 100% accuracy for fault severity level S3.
RVFL+GASF and RVFL+RGASF achieved 99.96% classification accuracy.

Abstract

In the present study, five reservoir computing models are compared and analyzed for bearing fault classification and severity level identification. Three Gramian Angular Field (GAF) methodologies, such as Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), and Robust Gramian Angular Summation Field (RGASF) images, were applied to generate various faulty condition images, and Deep Echo State Network (Deep-ESN), Liquid State Machine (LSM), Neuron-Astrocyte Liquid State Machine (NALSM), Random Vector Functional Link (RVFL), and Echo State Network (ESN) were trained and cross-validated. Due to the limited availability of the fault dataset, SA-ConSinGAN was employed to increase the number of transformed images of different bearing fault conditions. Ten-fold cross-validation was used to evaluate the performance of the reservoir models with correct identifications of bearing faults and severity levels. It is noticed that the combination of RVFL+GADF gave 100% accuracy in fault classifications, and fault severity level S3 achieved 100% accuracy with RVFL+GADF combinations. Also, RVFL+GASF and RVFL+RGASF combinations reached 99.96% fault classification accuracy. The results brought out the importance of GAF-based image processing techniques and SA-ConSinGAN along with reservoir computing models in predictive maintenance systems for fault diagnosis.

Ask AI

Helpful

Bookmark

View Full Paper