A technological and scientific core

Our approach to predictive maintenance

Feature generators at the heart of our DNA

Amiral Technologies was founded on scientific inventions from the GIPSA laboratory at the CNRS in Grenoble: highly discriminating feature generators.

The essential starting point in processing time series is feature generation. On the quality of these features depends the resulting model, and thus the accuracy of the predictions.

These systems are specifically designed for industrial equipment. They exploit both the most advanced signal processing techniques and the proven existence of physical laws governing the equipment, in order to identify invariants in the signals.

This clever combination allows DiagFit to extract relevant information from the equipment sensor time series to produce and update prediction models, even in blind mode.

Between proprietary algorithms and state-of-the-art

Amiral Technologies’ R&D also leverages the state of the art in machine learning algorithms to make them more specialized and therefore more effective than standard market solutions for industrial equipment monitoring.

Our constant research has led us to not only design new proprietary anomaly detection algorithms, but also to adapt the latest relevant algorithms to our feature generators to take advantage of all the market innovations.

This dual approach allows DiagFit to achieve unprecedented detection performance.

From anomaly detection…

By combining our feature generators and our algorithms, we are able to create a normality space representing the equipment’s nominal operation. 

This method, known as anomaly detection in machine learning, is the starting point of our incremental approach and allows us to detect drifts in the signals.

… to failure prediction

Our methodology allows us to detect drifts but also to anticipate them! 

Thanks to its precise and discriminating technological core, DiagFit detects weak signals in the data and raises alerts before the failure actually occurs. This ability to anticipate opens the field of failure prediction.

Chalk & talk

Makia Zmitri Ph.D., Data Scientist
at Amiral Technologies

The health score of an equipment

The challenge, in this case, is to determine the threshold that will best meet the trade-off between false alarms and true detections.

Play Video

Maxime Le Gall, Data Scientist
at Amiral Technologies

Feature generation

This step is paramount in the processing and characterization of time series.

Play Video
Play Video

Maxime Le Gall, Data Scientist
at Amiral Technologies

Feature generation

This step is paramount in the processing and characterization of time series.

Request a demo