Temporal signals to images: Monitoring the condition of industrial assets with deep learning image processing algorithms

The ability to detect anomalies in time series is considered highly valuable numerous application domains. sequential nature of objects responsible for an additional feature complexity, ultimately requiring specialized approaches order solve the task. Essential characteristics series, situated outside domain, are often difficult capture with state-of-the-art anomaly detection methods when no transformations have been applied series. Inspired by success deep learning computer vision, several studies proposed transforming into image-like representations, used as inputs models, and led very promising results classification tasks. In this paper, we first review signal image encoding found literature. Second, propose modifications some their original formulations make them more robust variability large datasets. Third, compare on basis a common unsupervised task demonstrate how choice can impact same architecture. We thus provide comparison between six algorithms without modifications. selected Gramian Angular Field, Markov Transition recurrence plot, grey scale encoding, spectrogram, scalogram. also achieved raw input another model. that encodings competitive advantage might be worth considering within framework. performed dataset collected released Airbus SAS, containing complex vibration measurements from real helicopter flight tests. different detection.