Publications by authors named "Gabriel Michau"
Fully learnable deep wavelet transform for unsupervised monitoring of high-frequency time series.
Proc Natl Acad Sci U S A
February 2022
Article Synopsis
- High-frequency (HF) signals are important for monitoring industrial assets, but traditional deep-learning tools often struggle with their size and complexity.
- This paper presents a fully unsupervised deep-learning framework that extracts meaningful representations from raw HF signals by incorporating properties of the fast discrete wavelet transform (FDWT).
- The proposed architecture is designed to be learnable, allowing for effective denoising and feature extraction without needing prior knowledge or additional processing, and it outperforms existing methods in various machine-learning tasks on sound datasets.
Article Synopsis
- Reliable fault detection is essential for efficient industrial operations, but variations in condition monitoring data due to external factors and rare fault occurrences can lead to false alarms.
- The proposed solution involves using contrastive learning with feature representation trained by triplet loss to improve fault detection and diagnostics, addressing issues from changing conditions and novel fault types.
- Evaluation on the CWRU bearing benchmark dataset shows that the approach maintains invariance to operating condition changes while effectively detecting new types of faults.
Interpretable Detection of Partial Discharge in Power Lines with Deep Learning.
Sensors (Basel)
March 2021
Partial discharge (PD) is a common indication of faults in power systems, such as generators and cables. These PDs can eventually result in costly repairs and substantial power outages. PD detection traditionally relies on hand-crafted features and domain expertise to identify very specific pulses in the electrical current, and the performance declines in the presence of noise or of superposed pulses.
View Article and Find Full Text PDF