Time-series and sensor data analysis

Ezako labels events and detects anomalies in sensor and measurement series. Its work covers engine test signals, satellite telemetry, the inspection of radio sites and vibration monitoring on embedded boards; the same methods apply to machines, production lines and fleets of equipment.

On sensor series

  • Similar patterns

    From a labeled event, search for the ranges that resemble it, on one or several columns and across series; each result carries a score and is reviewed by the analyst before it receives a label.

  • Label candidates

    Rare patterns of a column proposed as candidates, scored, and submitted to the analyst for validation.

  • Computed columns

    Combination of columns, derivative, low-pass denoising, normalisation, and the remaining time before a labeled event, the target of a remaining-useful-life model.

Fields of application

  • Rotating machines and turbines

    Detecting a drift in vibration, pressure and temperature series before it becomes a failure.

  • Production lines

    Spotting a quality drift in the sensors of a line before it reaches the parts.

  • Fleets of equipment

    Ranking which sites, charging stations or vehicles to inspect first, as for ANFR’s radio sites.

Software and services

What Ezako provides

  1. Upalgo Labeling Timeseries

    Labeling of sensor time series, with the search for similar patterns across series.

  2. AI Algorithms

    Automatic detection and pre-labeling, validated by analysts.

  3. UpalgoDB

    The central server for recordings, labels and labeling teams.

References

Examples of projects carried out

  1. DeepGreen

    Vibration monitoring on an embedded board

    Research and development within DeepGreen, the project led by CEA that builds the open-source Aidge platform for embedded AI: fault detection by machine learning on a microcontroller board of the STM32 class. Power generation machines were fitted with sensors to detect a departure from their expected behaviour; Ezako brings its expertise in time series.

  2. Safran Aircraft Engines

    Abnormal observations in engine test signals

    A deep learning method to detect abnormal observations in the high-frequency sensor series of engine development tests, where they make the spectrograms unusable. Published with Safran Aircraft Engines.

  3. CNES

    Anomalies in satellite telemetry

    Detection of anomalies in satellite telemetry by machine learning, in addition to the conventional surveillance methods.

  4. ANFR

    Which radio sites to inspect first

    Prediction of the radio sites most likely to reveal an anomaly, to set the order of the inspections of a national park of more than 76,000 sites.

See what Ezako can do with your data

Describe your signals and your requirements to arrange a demonstration.