Home > TERATEC FORUM > Workshops

TERATEC 2019 Forum
Workshops - Wednesday June 12

Workshop 5 - 14:00 to 17:30

Machine learning & Maintenance prédictive

Chaired by Erick JONQUIERE, AFNet et Jean-Laurent PHILIPPE, Intel

© TERATEC 2019

Anticipate a breakdown, control the shutdown of machines, increase their service life, reduce spare parts stocks ... Here are the enticing promises of predictive maintenance.

To reliably anticipate machine appliances and vehicle component failures, organizations must implement a continuous cycle of data collection, exploration and analysis. It is important to collect the data where it is generated, that is, closer to the components themselves. The sensors installed for this purpose record the behavior of the equipment in the form of data that will feed and enrich the analysis cycle.

The word predictive maintenance implies super-intelligent algorithms able to rely on billions of data to predict or even prevent failure months in advance. This vision is far from reality for most manufacturers. "You must not be too ambitious from the start". To switch to predictive maintenance, you must first connect your machines to a data collection system. When the machines are connected, they send data. A lot of data. Therefore, the main problem is a matter of  quality rather than quantity.

The large amount of data, coupled with the increasingly powerful processing capabilities, makes it possible to develop and execute the large-scale algorithms needed to fully exploit the potential of artificial intelligence. But some software solutions also recommend using existing data from the industrial architecture as a priority to define relevant analysis models rather than multiplying sensors networks with often complex deployment. Analytical models and machine learning algorithms can then predict the probability of failure. The goal is not to create the most advanced algorithm, but to implement realistic and functional machine learning in the operational chain.

A good implementation of this predictive maintenance based on the analytic approach of the data requires a global vision of the company and a solid network architecture to process all data for the benefit of maintenance coupled with a functional organization of this maintenance. In addition, integration with business rules is as important as using effective models. In fact, it enables companies to make the link between the analytical forecasts and the recommended measures for decision-making.

What about the future of major technological developments such as augmented analytics and continuous intelligence that combine machine learning and artificial intelligence techniques to transform the way analytics content is developed, used and shared? By 2022, more than half of all major new enterprise systems will incorporate continuous intelligence that uses contextual data in real time to improve decisions.
Finally, what factors should companies pay attention to, in oder to make the most of machine learning in predictive maintenance?

During this AFNET / CNIS workshop in the framework of the TERATEC Forum, many actors, solution providers and industrial users will come to testify and share on the subject their own rich experience feedback.

« It's not intelligence that does, it's intelligence that watches us do » (Paul Claudel)

With the participation of :


Abstract & Bio
Download the presentation

Real-Time Anomaly Detection Using Deep Learning to Predict Robot Failures
Jean-Laurent PHILIPPE , DGC Sales, Senior HPC Technical Sales Specialist, Intel

Abstract & Bio
Download the presentation

Preventive or predictive maintenance of infrastructures via drones
Vincent THAVONEKHAM, Cloud Azure strategy Manager, Microsoft Regional Director
Frederick VAUTRAIN, Directeur Data Sciences, VISEO
Guilhem VILLEMIN , Directeur Technique, ALTAMETRIS

Abstract & Bio
Download the presentation

Predictive maintenance solution without additionnal sensors
Christophe BIERNACKI, Responsable. équipe de recherche MODAL, INRIA
Margot CORREARD, co-fondatrice DiagRAMS Technologies (start-up INRIA)

Abstract & Bio
Download the presentation

L’apprentissage automatique au service de la gestion d’intégrité des parties sous pression »
Michel-Ange CAMHI, Group Chief Data Officer, Bureau Veritas

Abstract & Bio
Download the presentation

Improve OEE and predictive maintenance : Machine Learning integration in manufacturing assembly lines
Serge BONNAUD, Technical Leader, IBM Europe

Abstract & Bio
Download the presentation

For any other information regarding the workshops, please contact :

Jean-Pascal JEGU
Tel : +33 (0)9 70 65 02 10
2, rue de la Piquetterie


© Ter@tec - All rights reserved - Lawful mention