The blind failure prediction software
DiagFit is our predictive maintenance software that enables manufacturers to anticipate and plan their machine shutdowns. The software is based on state-of-the-art machine learning algorithms for industrial time series management.
These properties are extremely rich and allow the efficient detection of weak signals of failure, ageing or approaching end of life of equipment.
The specification generation, an invention at the heart of the software, allows the construction of mathematical properties of time series from industrial equipment or IIoT sensors.
Predictive maintenance: blind failure prediction
The "blind mode" means that DiagFit does not need to know the type of equipment it is monitoring (technical or physical specifications) and does not need to be enriched with a failure history in order to operate. This allows the software to be implemented extremely quickly as the industry is equipped with increasingly reliable equipment that rarely fail.
Thanks to its unsupervised approach, the software learns from the sound data of the equipment and automatically builds a robust and reliable normality space. This operation allows the detection of anomalies with high accuracy and a low false alarm rate. It is the management of discriminating properties from our specification generators that makes detection so effective
The advantages of DiagFit
The advantages of DiagFit
Request a demoBlind mode
Works without historical failure data
No-code
No data science skills required
Agnostic
Sensor and equipment agnostic
Quick
Models built in hours/days instead of weeks/months
Accurate
Reduced false alarm rate
Software based on the knowledge of domain experts
DiagFit is incredibly fast and easy to use as it has been designed for maintenance experts.
A step-by-step process allows you to generate a predictive model dedicated to a piece of equipment without any knowledge of coding or data science. This step* called "BUILD" usually takes only a few minutes/hours.
Once generated, the predictive model can then be used and linked to operational equipment in the “RUN” step. Domain experts are then alerted when deviations from the normality space occur and benefit from the indications given by the software to identify their origin. These deviations, called anomalies, can be accepted or rejected by maintenance operators who have functional knowledge of the equipment concerned.
This enrichment by experts allows the model to learn over the life of the equipment and to be even more efficient.
*can be carried out by our experts if desired, please contact us for more information.
Labelling of anomalies
The labelling function allows the maintenance expert to enter an anomaly in the dictionary associated with the equipment.
The software will then be able to recognise its signature automatically in the future and give the user a confidence score for its diagnosis.
As with all alerts, the expert can then confirm or deny the prediction.
Deployment
- Public cloud
- Private cloud
- On-premise
- Integrated with a third party platform
- Embedded (please contact us)
Architecture and interfaces
- Web microservices architecture
- On linux server with Dockers
- Minimum configuration: 4 core processors at 3 GHz
- REST API with interactive online documentation
- Alert notification via MQTT protocol
Clients
Clients
Use Cases
Use Cases
Failure prediction on an anonymised ship system
Using anonymised health data acquired over a year on the ship, DiagFit was used to create a unique model within hours that captures correlations between sensor data and produces a health status for each sensor
Quality prediction and normative maintenance for plastic presses
Using the data from the variable speed drive on the press and the quality measurements at the output of the press, DiagFit analyses the deviation of the specifications used to derive a virtual quality sensor.
Pipe crack detection
Based on the eddy current data used to monitor the surface of pipelines, DiagFit can classify healthy and unhealthy pipe sections