DiagFit

The failure prediction software that monitors your equipment's health

DiagFit is a blind failure prediction software, an essential step in the predictive maintenance process allowing to anticipate and plan machine downtime.

Based on state-of-the-art unsupervised machine learning algorithms for industrial time series management, DiagFit enables the extremely rapid implementation of a solid and reliable failure prediction solution.

Feature generation, the core invention of the software, extracts particularly rich mathematical properties that allow DiagFit to efficiently detect in real time the weak signals of failure or aging of these equipments.

Based on a unique incremental approach, our DiagFit tool relies on artificial intelligence to enable dynamic learning of equipment operation throughout its life, driven by feedback from business experts.

Based on state-of-the-art unsupervised machine learning algorithms for industrial time series management, DiagFit enables the extremely rapid implementation of a solid and reliable failure prediction solution.

Feature generation, the core invention of the software, extracts particularly rich mathematical properties that allow DiagFit to efficiently detect in real time the weak signals of failure or aging of these equipments.

Based on a unique incremental approach, our DiagFit tool relies on artificial intelligence to enable dynamic learning of equipment operation throughout its life, driven by feedback from business experts.

Advantages of DiagFit in failure prediction :

Speed

its “blind mode” approach allows failure prediction models to be generated in a few hours/days

Accuracy

its high-performance technological core leads to an extremely low false alarm rate

Explicability

its sensor information and failure recognition facilitate efficient and targeted equipment intervention

Universality

DiagFit works with all types of equipment and sensors, as long as they produce time series

Usability

its No-Code interface requires no knowledge of coding or data science

Autonomy

its intuitive handling allows users to use the software independently

Examples of DiagFit interface

DiagFit, an iterative process in 4 steps

The software allows to generate a predictive model dedicated to a device without any knowledge of coding or data science. This step, called “BUILD “*, takes only a few minutes/hours.
Manual modification of some of the model parameters is available for more experienced data science users.

* This step can be performed by our experts, please contact us.

Once generated, the model can be tested on one or several new data sets. This step aims to test the robustness of the model, to refine the normality space previously generated in step 1 and to initialize the fault dictionary which will be used in step 3 and enriched in step 4.

The tested model is then associated with working equipment in this operation step called “RUN”. Technicians and other experts are alerted when operational deviations are observed in the data. They then benefit from the software’s indications to identify the origin of these deviations, called anomalies, and intervene effectively to restore the affected part or equipment to working order.

Maintenance operators, knowing the functional aspects of the equipment involved, can then accept or reject the detected anomalies. This step aims to complete the model’s learning process and enrich the dictionary of anomalies that will enable automatic recognition of failures in the future.

Build

The software allows to generate a predictive model dedicated to a device without any knowledge of coding or data science. This step, called "BUILD"*, takes only a few minutes/hours. Manual modification of some of the model parameters is available for more experienced data science users.

* This step can be performed by our experts, please contact us.

Test

Once generated, the model can be tested on one or several new data sets. This step aims to test the robustness of the model, to refine the normality space previously generated in step 1 and to initialize the fault dictionary which will be used in step 3 and enriched in step 4.

Operate

The tested model is then associated with working equipment in this operation step called "RUN". Technicians and other experts are alerted when operational deviations are observed in the data. They then benefit from the software's indications to identify the origin of these deviations, called anomalies, and intervene effectively to restore the affected part or equipment to working order.

Enrich

Maintenance operators, knowing the functional aspects of the equipment involved, can then accept or reject the detected anomalies. This step aims to complete the model's learning process and enrich the dictionary of anomalies that will enable automatic recognition of failures in the future.

“A virtuous loop is then set up as the equipment's life goes on.”

Failure prediction software with high deployment flexibility

Public or private cloud

Azure, OVH, AWS or any other proprietary cloud, DiagFit can be integrated to all your environments

On-site

DiagFit is designed to be deployed on the most demanding infrastructures

Integrated with a third party platform

The power of our technology can be redirected to a third party application thanks to our REST APIs

Embedded

Contact us for more information

A wide range of use cases

Thanks to the blind mode and its easy implementation, DiagFit can cover a large number of use cases in our different key markets (transportation, manufacturing, energy):

Rotor blades

Transport

Nuclear pipelines

Energy

Wipers

Transport

Cryopumps

Energy

Industrial PCs

Industry

Train doors

Transport

Welding robots

Industry

Circuit breakers

Energy

Radars

Transport

Injection pumps

Energy

FAQ

Frequently Asked Questions

Yes, the power of our technology can be redirected to a third-party application through our REST APIs. We are setting up a downstream API and an upstream API.

More about DiagFit

Our ability to avoid failure labels, the speed of model creation and their performance.

Our models are built on the basis of healthy data, so it allows you to start creating models without fault labels. DiagFit is often used as a productivity tool to build many models in a very short time. In addition, the performance of our algorithms allows the identification of weak signals undetectable by conventional approaches, and allowing us to characterize the anticipation of failures

More about blind mode failure prediction

Yes, DiagFit is robust to context changes as long as the data has been seen in training. In addition, our feedback functionality makes it possible to enrich the models previously built.

  • When a new anomaly appears, the user has the option of approving or rejecting this anomaly:
  • If it is a real anomaly, the user can label it and thus enrich a dictionary of failures. DiagFit is able to calculate the probability of its belonging to one or more categories of failures already seen. It then gives the user a confidence score with regard to this labelling.
  • If it is a false anomaly (or false positive), the user can reject it and re-train the model taking this behavior into account

More about the failure dictionary

No. We train the models on healthy data, however, failure labels can be useful to evaluate the performance of the models created by DiagFit.

More about blind mode failure prediction

Yes, this eliminates a certain number of FPs and fault detection is only better.

Amiral Technologies takes advantage of the best of both approaches by working on dynamic learning throughout the life of the equipment. This methodology is particularly relevant in the industrial context where occurrences of failures are rare and occur over time.

Supervised + unsupervised = Incremental Model Learning

Request a demo