Industrial measurements / Signal processing
Deep Learning for Industrial Signal Processing
Autoencoder-based denoising and learned representations for interpreting industrial signals.
Problem statement
Distinguish useful signal structure from noise and identify observations that warrant further attention.
My contribution
Development of an autoencoder-based approach to denoising industrial signals.
Technical challenges
Domain-level considerations, without implementation-specific details.
- Reducing noise without losing meaningful signal characteristics.
- Interpreting unusual observations without assuming every deviation is a failure.
- Evaluating reconstruction quality alongside downstream usefulness.
Methodology
Use autoencoder representations for denoising, with reconstruction-based analysis as a conceptual route to anomaly indication.
- Learn a representation that reconstructs the relevant signal structure.
- Evaluate anomaly indications as candidates for further review, not automatic diagnoses.
Conceptual architecture
Input signal
Learned reconstruction
Anomaly indication
Review
Technologies and methods
- Autoencoders
- Denoising
- Anomaly Detection