ENGIE Digital Uses Amazon SageMaker for Predictive Maintenance at Power Plants
ENGIE builds on Amazon Web Services (AWS) to develop its predictive maintenance platforms. Ultimately, nearly 10,000 pieces of equipment will be connected, each with dozens of models, representing an estimated savings of €800,000 per year for the company.
The choice of an AWS architecture was guided by the constraints of our business, which by its industrial nature incorporates the concept of predictive maintenance. With several thousand pieces of equipment in operation, each with several dozen models, we scale up very quickly, and scalability is a significant issue.”
Mihir Sarkar, chief data officer at ENGIE Digital, describes the role of this particular entity within the ENGIE Group: “ENGIE Digital is the Group's software company. We develop digital platforms and products for the various Global Business Lines with a focus on specific verticals, such as renewables or thermal. The purpose of digital products is to provide internal entities with solutions to enable them to improve their operational efficiency. They can also enhance the offer for external customers to contribute to their energy transition towards carbon neutrality.”
It is on the basis of these objectives that ENGIE Digital developed the Robin Analytics and Agathe platforms. The goal of these digital platforms is to develop predictive maintenance models to prevent equipment malfunctions and schedule maintenance more effectively. Robin Analytics is aimed at equipment within the Group's thermal power plants, while Agathe is offered to B2B customers to ensure their equipment is well maintained.
Challenges of Predictive Maintenance
ENGIE wanted to benefit from the infrastructure and tools enabling it to develop, train, and deploy predictive maintenance models in order to anticipate breakdowns and malfunctions of equipment within the Group's power plants and for its customers, while keeping control over the allocation of resources and costs. It also wanted to be able to stay up to date and benefit from the latest technological innovations to guarantee the industrialization and scalability of its models.
To implement these projects successfully, ENGIE Digital partnered with Mangrove, an AWS Advanced Consulting Partner specializing in machine learning and AWS services in France.
For ENGIE, leveraging machine learning technologies to address equipment maintenance issues is a key issue. “The challenge of predictive maintenance is primarily optimizing costs. Systematic maintenance forces us to conduct site visits at regular intervals. Better planning and optimization of maintenance cycles is an important issue for our B2B customers and for our power plants,” says Sarkar.
ENGIE Digital currently has three use cases specific to predictive maintenance, as detailed by Céline Mallet, head of the predictive maintenance platform Agathe at ENGIE Digital: “The first is predicting the useful life of our equipment. Equipment is liable to wear out and lose efficiency. Predictive maintenance allows us to predict when we will reach the efficiency thresholds justifying a maintenance action and replacement of a part, several days, weeks, or months in advance. The second is early detection of anomalies. Predictive maintenance allows us to use operating data from equipment (power plants, compressors, etc.), which allow us to detect abnormal behavior early on. Finally, we also have the possibility of developing models that take advantage of the data generated by IoT sensors (measuring ultrasound vibration, for example) to estimate the state of health of the equipment from the recorded variables.”
Predictive maintenance is also taking place against a backdrop of changes in the Group’s overall strategy, aimed at favoring the production of renewable energy. It helps facilitate this transition, as Sarkar explains: “One of the recent developments in the organization of electricity production is that thermal power plants must now ensure a baseload and adjust their production in relation to that produced by intermittent renewable energies. ENGIE's shift towards energy transition means the thermal power plants are used differently, with more frequent shutdowns when there is wind and sun, for example. The use of the equipment in these plants, such as valves and pumps, is therefore different, which can produce behaviors and failures for which it is essential to have visibility. Predictive maintenance models make it possible to better anticipate breakdowns and not wait for systematic maintenance cycles to take action or, worse, have to deal with unanticipated shutdowns.”
The Choice of Services to Solve ENGIE’s Challenges
ENGIE Digital has been using AWS for more than 3 years now: “Of the 12 platforms in ENGIE Digital's portfolio, 11 are built on AWS. These services are also used within other Group entities such as Data@ENGIE, which is behind the Common Data Hub, the Group's distributed data lake on which Agathe and Robin Analytics now operate,” says Sarkar. ENGIE Digital turned to these solutions when developing Agathe and Robin Analytics: “The choice of AWS architecture was guided by the constraints of our business, which by its very nature incorporates the concept of predictive maintenance. With several thousand pieces of equipment in operation, each with several dozen models, we scale up very quickly, and scalability is an important issue,” says Sarkar.
In order to take full advantage of AWS, ENGIE Digital enlisted the support of Mangrove, a consulting company specializing in AWS services: “We turned to Mangrove so they could lead the transformation to AWS services that we weren't using at the time, such as Amazon Simple Storage Service (Amazon S3) and Amazon SageMaker. We can also count on Mangrove as an expert to help us make the most of the opportunities that arise from new releases and developments of AWS services,” says Mallet.
For the Agathe and Robin Analytics platforms, Mangrove uses several AWS services, as explained by Bastien Murzeau, Mangrove’s chief technology officer and tech lead at ENGIE Digital: “Amazon S3 is an essential service for us, all of our data goes through it. It is also thanks to this service that we can then analyze the data using services such as AWS Glue and Amazon Athena, and other lighter transformation processes via AWS Lambda."
These services allow users to keep control of the resources used, as well as the costs, as Murzeau explains: “AWS Glue allows us to run Spark easily and inexpensively with dynamic scalability. We can handle very small tasks just as easily as very large ones.”
“Cost control is essential because maintenance is a very competitive sector,” adds Mallet.
Using Amazon SageMaker to Train Maintenance Models
An example of the collaboration between ENGIE Digital and Mangrove is the adoption of Amazon SageMaker. A year ago, ENGIE Digital wanted to improve its predictive maintenance models and turned to Mangrove to gage the potential benefits of use. Murzeau now believes the service is of central importance: “Amazon SageMaker is a key service for us. The advantage of using it is that we don't have to reinvent the wheel and can rely on a service that works and provides us with stability. Previously, we trained our models ourselves, and we didn’t benefit from the best practices that the service constrains us to follow. It also gives us a higher level of security, due to the compartmentalization of training tasks, and enables our customers' data to be isolated. Another benefit of Amazon SageMaker is cost control thanks to training using Amazon Elastic Compute Cloud (Amazon EC2) Spot Instances, which offer up to 90-percent savings in terms of computing costs.”
ENGIE Digital relies on all of these services to enable Agathe and Robin Analytics to develop and train a large number and a wide variety of maintenance models. “With Agathe, our ambition is to cover 8,000 pieces of equipment within 5 years, each with 2-10 predictive maintenance models,” says Mallet.
“For ENGIE thermal power plants, we are preparing to onboard more than 100 pieces of equipment on the Robin Analytics platform, with the aim of reaching a total of more than 1,000 pieces of equipment by 2023. These include a great diversity of types of equipment, including several types of valves, pumps, and heat exchangers, as well as geographic diversity in the deployment of the predictive maintenance solution for the Group’s various plants and business units,” says Sarkar about the development of Robin Analytics. ENGIE Digital teams are also studying the adoption of additional services that could benefit Agathe and Robin Analytics: “As we did last year for Amazon SageMaker, we investigated the feasibility and benefits of migrating our timeseries databases to Amazon Timestream, which we plan to eventually adopt,” says Sarkar.
“One AWS technology that we are not yet using but are considering adopting is Amazon SageMaker Studio to onboard data scientists from business units and enable them to access our platform and datasets. This would also ensure that their code is production-ready for the platform,” says Mallet, as an avenue for improvement for the Agathe platform.
With a presence in more than 70 countries across five continents, ENGIE is a key player in low-carbon energy production, distribution, and services on a global scale. The development of ENGIE Digital is testament to the Group's desire to create and take advantage of digital technologies to accelerate the strategy for transitioning to carbon neutral energy.
Benefits of AWS
- More than 1,000 prediction models developed and trained in a short time for a variety of equipment such as valves, pumps, ventilation, air conditioning, and heating systems.
- 10,000 pieces of equipment connected and benefiting from predictive maintenance within 5 years for ENGIE power plants and the Group's B2B customers.
- Estimated savings of €800,000 per year for the Group's business units that have adopted predictive maintenance.
Amazon Simple Storage Service (Amazon S3) is an object storage service offering industry-leading scalability, data availability, security, and performance.
Amazon SageMaker helps data scientists and developers quickly prepare, build, train, and deploy high-quality machine learning (ML) models by bringing together a broad set of features designed specifically for ML.
AWS Glue is a serverless data integration service that makes it easy to prepare data for analytics, machine learning, and application development.
Amazon Athena is an interactive query service that makes it easy to analyze data in Amazon S3 using standard SQL. Athena is serverless. There is no infrastructure to manage, and you pay only for the queries that you run.