Predictive maintenance for ladle foundry

Client: Celsa
Location: Abrera, Barcelona, España
Sector: Metal
Application: IIoT

Celsa, a metallurgical company dedicated to metal rolling, wanted to learn about the capabilities of working with serverless systems (SaaS), in order to optimize both the management of data obtained from their production processes, and improve the post-processing of this data, through the application of Artificial Intelligence, Machine Learning and predictive maintenance.

Main objective:

Reduce downtime due to breakdowns

Proposed approach:

Predictive maintenance

In depth analysis:

Initial situation:

Celsa has several production plants around the world. In many of its factories, the processes are the same or very similar, so Celsa is looking for the most efficient way to obtain these data, store them, and process them in the most optimal way to obtain metrics and be able to feedback the process with improvements.

Celsa asked Engapplic for a PoC (Proof of Concept) in which they wanted to test and analyze the capabilities of working with a service-based system (Serverless), with the following objectives:

  • Optimize data processing costs, from the machine to the database or Cloud.


  • Simplify the data collection infrastructure with Cloud services, as well as obtain a robust and secure system.


  • To have the ability to interact a posteriori with Cloud services that would allow them to process data easily and quickly, such as AI (Artificial Intelligence), ML (Machine Learning) and BI (Business Intelligence).


  • To be able to distribute data and information to the different departments according to their needs and objectives.


  • Establish the basis for implementing predictive maintenance.

Also, without being part of the scope of the PoC, the aim is to be able to transmit the feedback obtained directly to the machine in the future in order to improve the process, increase quality and reduce production costs.

Proposed solution:

Engapplic has implemented a data capture system using the following AWS services:

  • Data collection: AWS Greengass has been implemented as a bridge of the MQTT topics at the local level to communicate with AWS IoT Core, a service located in the Cloud. So we can define this part as the encrypted transfer of data to the Cloud.
  • Once in the Cloud, AWS IoT Core is used to take the data from the topic, process it following some parameters and inject it (rules) in AWS SiteWise for storage and visualization.
  • Finally, for this PoC, the use of AWS IoT SiteWise as a SaaS service, in charge of generating the data model, storing in a temporal database (time series) and generating a portal from which there is the possibility of generating various alarms and warnings, as well as the visualization of the data in graphical format.

This service infrastructure then allows to connect services such as SageMaker or QuickSight, and therefore to use AI (Artificial Intelligence) or BI (Business Intelligence) services. In this way, the scalability of the proposed solution and its ability to be scaled for the rest of the customer’s applications is demonstrated.


The main objective was to demonstrate the capabilities of working with a system focused on optimizing the entire data flow to the Cloud, since once the data are already in the Cloud, the management of these and the interaction with other services is easy, simple and fast, as well as the ability to add or remove parameters, values and attributes.


As a result of the PoC, the following conclusions have been obtained:

  • The system is scalable and flexible, allowing for expansion and/or modification according to the needs of each production process.
  • The system is robust and secure, with data encryption and a system of users and roles that is easy to define, modify and adapt to the needs.
  • Working with the data in the Cloud and with Serverless services is the most optimal way to be able to process this data and link it with other services such as AI, ML and BI.
  • In order to scale the process, in this case, it is necessary to first make an in-depth study of the data and generate a clear and defined list of “things”, including the metrics, the model and its attributes.
  • There has been a significant cost reduction thanks to the use of a serverless system as opposed to a classic infrastructure (On-premises).
  • System prepared for the implementation of predictive maintenance.

On the other hand, the possibility of generating an automatic feedback between the processed data and the production process has also been analyzed, and it has been concluded that, although it is possible to perform such feedback, opening the door to improvements in production, quality and maintenance reduction, the critical component and the type of manufacturing that Celsa carries out, still requires approval by a manager before implementing the changes, so for the time being, a series of proposals would be made that should be approved by the relevant managers, until achieving the robustness and maturity of an automated feedback system.