RTF-Skip-GANomaly: A Novel Deep Learning Framework for Anomaly Detection in Run-to-Failure Data Using Skip-GANomaly Backbone

Anomaly detection in run-to-failure (RTF) data is one of the pivotal tasks in a variety of real-world signal processing applications. Vast majority of the existing anomaly detection methods in the literature are based on generative adversarial networks (GANs), but they are ineffective and problematic when applied to the RTF data. To break through such a limitation, in this paper, we proposeRTF-Skip-GANomaly, a novel and high-performing deep learning (DL) framework for anomaly detection in the RTF data, by leveraging the off-the-shelf Skip-GANomaly as backbone. Key innovations in our approach are two-fold: (i) we introduce an extra dual-stream DL network calledWeibull-Netand integrate it into the Skip-GANomaly backbone such that a pseudo Weibull function is predicted from the latent representation of the reconstructed RTF data with time embedding, and (ii) we also devise an effective strategy to train the Weibull-Net jointly with the Skip-GANomaly backbone through a customized loss function design. Experimental results validate the superior performance and enhanced efficacy of our framework over the existing approaches.