Share

Smart maintenance for optimized production

By continuously and automatically monitoring industrial processes and equipment, smart maintenance can predict and minimize the risk of malfunctions during production. CEA-List is bringing data analysis, machine learning, and knowledge modeling technologies to smart maintenance systems.

Approaching nominal operation

A piece of production equipment operating nominally is safer, consumes less energy and raw materials, lasts longer, and runs less risk of unexpected shutdowns.

Smart maintenance has the capacity to bring equipment closer to nominal operation and optimize the maintenance process itself, limiting downtime and bolstering the security of field operations.

Continuous, automated equipment monitoring

Smart maintenance covers every step of the maintenance process by:

  • Detecting malfunctions and stoppages.
  • Identifying a problem’s causes (diagnosis) and effects (prognosis) on the production environment.
  • Recommending corrective and/or preventive measures.

Digital technologies make automation of this complex process possible. Monitoring can be carried out continuously, directly on or very close to production equipment, for safer, more accurate maintenance processes.

Continuous monitoring of data—either collected by sensors installed on the production equipment or generated by the equipment itself—enables early detection of process drift and prevention of stoppages. This means that maintenance interventions can be as safe and efficient as possible.

Implementing data science tools capable of analyzing large quantities of heterogenous data (e.g. physical, mechanical, or electric measurements, machine data, etc.) enables sophisticated detection strategies that broaden the scope of investigations by human operators. By defining dependencies between data and leveraging data logs, these tools can help predict stoppages before they occur and diagnose likely causes.

Advantages for every industry

Virtually every industry can benefit from the potential of smart maintenance. The nature of the data can vary from one industry to another, but with personalized settings and learning approaches, similar strategies and tools can be used.

 

Use cases

  • Mechanical industries: monitoring of rotating machinery
  • Chemistry: monitoring process stability
  • Plastics: monitoring injection molding processes
  • Cyber monitoring: server log analysis, and more

Our solutions

Actionable data

  • Physical measurements (temperature, humidity, etc.), electric, acoustic, and mechanical data (vibrations, pressure, etc.) from sensors near or on machines.
  • Images from cameras (visible light, infrared, etc.).
  • Production data from supervision systems.
  • Technical data from server logs.

Analysis tools

  • AI systems that implement machine learning and deep learning
    These systems can learn to predict malfunctions by recognizing past examples (supervised learning mode) or by detecting deviations from “nominal” behavior (unsupervised learning mode). These tools require substantial data logs. CEA-List has developed two tools to aid in the development of these systems:

    • Streamer, an open-source toolkit for managing data streams, designed for creating and evaluating AI modules that implement continuous learning and analysis based on physical signals (like electric, audio, or vibration data) or on sequences of events (machine logs).
    • DNAi (Distributed and Networked AI), a distributed AI platform allowing machines to share collected information and models learned locally by each machine without having to directly share data.
  • Symbolic AI and expert systems
    These technologies allow for automated reasoning through the modeling of expert knowledge. They are suitable for domains where human expertise is well established and/or data logs are rare. Unlike deep learning tools, they have the advantage of being able to produce explainable decisions. CEA-List’s answer to this is:

    • ExpressIF®, a symbolic AI platform that can model human knowledge, collect or update knowledge based on observations, and reason automatically, applying various reasoning models in order to process data based on the knowledge.
Example

RATP sets sights on operational excellence with the help of predictive maintenance

RATP, the operator of the Paris metro network, set itself the goal of ensuring 97% service availability.
The network is managed by a supervision system. This system manages traffic from thousands of data points coming from infrastructure equipment on the lines (signals, switchers, station doors, etc.).

RATP turned to CEA-List for help reaching its operational excellence targets. Together, the two partners created a predictive maintenance system implemented across the infrastructure as well as in the supervision system itself, which tended to freeze when alert volumes were particularly heavy.

The system was developed by adapting CEA-List data analysis and ML technologies to the RATP use case. It processes a large variety of field data as well as technical data from logs created by the monitoring system. In order to avoid stoppages, it assesses the current status of the supervision system and predicts how the status will evolve over time.

The system was tested on three years of data. It detected 90% of labelled stoppages with a few hours’ advance notice. At the same time, the thousands of false alarms triggered exponentially by fixed thresholds (native to the supervision system) were reduced to a few instances per week. This simplified the use of the tool by employees working in the field.

Veolia models its expert knowledge

To optimize the operation of its wastewater treatment facilities, Veolia decided to automate process monitoring. CEA-List’s ExpressIF® was used to model Veolia experts’ knowledge to enable the monitoring system to:

  • Ensure correct operation of the facility.
  • Diagnose detected problems.
  • Recommend solutions.

The demonstrator that was developed for Veolia is available as a SaaS or on-premise solution, delivering the advanced decision making and uncertainty measurement capabilities of ExpressIF®. The system offers advanced interactive features for domain experts—if missing information compromises the decision-making process, the tool can request human intervention to resolve the issue.