DOE OSTI · 1514382
Failure Analysis on Multivariate Time-series Data given Uncertain Labels
Abstract
Machine failure analysis and detection is critical to today’s industrial society. Given a number of failure events on multivariate temporal data, the ability to identify 1) the discriminative patterns prior to the events and 2) the most relevant features to the failure has practical use for early warning and root cause analysis. However, since these patterns are not necessarily adjacent to the onset of failure in time, faulty labels are often with uncertainty, which makes traditional supervised detection methods inapplicable. To address the label uncertainty and learn the complicated correlation in multivariate time series, we design Failure Analysis on Multivariate Time-series Data (MAMT) that jointly selects the most failure-relevant features and time-instances by a novel dynamic and directional label diffusion process. Extensive experiments demonstrate that MAMT is more effective while more efficient than popular baselines.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Huang, Hao, Yoo, Shinjae. 2019-07-20. Failure Analysis on Multivariate Time-series Data given Uncertain Labels. https://www.osti.gov/biblio/1514382
Cite the original work for its findings. Save a collection to share your selection of sources.