Amiral Technologies launches version 1.5 of DiagFit, its flagship solution for blind failure prediction of industrial equipment – without the need for historical failure data. Sébastien Le Gall, CTO of Amiral Technologies, tells us more specifically about this software and its use cases.
Hello Sébastien, first of all, can you explain the concept of DiagFit?
DiagFit is a meeting between scientific research and the needs of industrial customers. Predicting industrial equipment failures is a complex subject and several approaches are possible to address it, from the simplest (simple monitoring of thresholds for a few parameters) to the most complex (complete digital twin with heavy-duty processing neural network base).
DiagFit is in the “very high performance” and “low resource consumption” box. Let me explain: the CNRS research on which our software is based makes it possible to automatically extract – and at a lower cost as regards the computing power and the need for sensors – the very relevant indicators of the state of health of industrial equipment from the data generated by it, without the need for an exhaustive history of fault data. This last point is major. This is what characterizes DiagFit’s blind approach.
From there, DiagFit sets the automation of data processing to music: in “Build” mode, a prediction model is automatically generated for the equipment in question, and in “Run” mode alerts on its state of health are raised automatically.
DiagFit, for whom is it exactly?
DiagFit meets the needs of industrial customers in several sectors. For the transport sector, we have dealt with the prediction needs of breakdowns and component wear in connected vehicles, aeronautical and rail equipment. In the energy sector, we have addressed the needs of prediction of cracks in pipelines, and aging of wind turbines. In the manufacturing sector, we have tackled issues of optimizing press output quality, and issues of predicting failures for electronic equipment. I would say that we are aimed at industrial customers who wish to greatly increase the uptime and operating time of their equipment, or to optimize the quality of their production.
What are the benefits of DiagFit compared to other solutions?
- Blind prediction capability, without historical failure data
- The ability to process data from any sensor, with a small number of sensors – or even with a single sensor – therefore without significant equipment intrusion
- The ability to produce predictive models without having to know the physical models of the equipment
- Model production time (minutes / hours): The automation built into DiagFit allows rapid deployment and results.
- The accuracy of the models (very low rate of false positives)
- The processing is not resource intensive like it is for other methods such as digital twins or neural networks. DiagFit can be operated in the cloud, edge or embedded in equipment / sensors
How do you use it?
Our goal is to empower customers to use DiagFit. The “Run” or “diagnostic” part must be able to be used by maintenance staff. The “Build” part is easily accessible by IIoT managers.
The creation of models is done by loading files containing the time series data of the proper functioning of the equipment. The diagnosis is made either in “online” mode by the streaming of data via the DiagFit Rest API, or in “offline” mode by loading files. To see a live demonstration, you can contact us at firstname.lastname@example.org.
How does the product work from a technical point of view?
It is a “web application” type software based on a micro-service architecture allowing scalability and facilitating deployment. There is therefore nothing to install on client computers, just open a web browser, all software management is centralized on a server. We have also made the choice to separate the frontend which is responsible for the graphical interface from the backend which integrates our algorithms and databases. This makes it possible to provide a customer with either DiagFit as a whole or only the backend with its Rest API allowing it to easily interface with an existing application. One final piece of information, all alerts raised can be sent to a broker whose address can be configured via the MQTT protocol.
For what types of data?
We work with cyclical or non-cyclical time series. Cyclic means that the data is split into windows of identical size which are repeated over time. For example, we can imagine the cycle of starting an engine, the rotation cycle of a transmission shaft, the operating cycle of a robot, etc.
We have no constraints on the sample rate, but our strength lies in being able to detect barely noticeable changes that conventional algorithms cannot detect. For cyclic data, we still recommend that the cycles have at least twenty values.
What are the new specificities of version 1.5?
The big novelty of version 1.5 is the notion of operator feedback.
In the “Run” phase, as DiagFit detects and retains the signature of anomalies not previously seen, it is now possible for the operator to enter the type of failure encountered and the necessary repair. This information, which we call feedback, helps to enrich the model over time in order to be more precise in future predictions and in prescribing necessary repairs.
Are you already working on the next version?
Absolutely. We have a very rich roadmap in terms of functionality, algorithms, user experience and integration capacity. We continue with an approach of continuous research to resolve customer issues and to facilitate the use and integration of our software.