CERTIF-AI: Certification of the quality of the production process through artificial intelligence

A consortium of academic and industrial partners, led by Jheronimus Academy of Data Science (JADS) and Technische Universiteit Eindhoven (TU/e), is collaborating on the development of an Artificial Intelligence (AI) toolkit, consisting of algorithms and methods, for industrial applications. Production processes can be made 'smarter' by leveraging the data streams generated by the machines used in production. In particular, these data streams can be mined to build a model of the production process as it was actually executed - as opposed to how it was intended.

In late April 2023, we spoke with project leader Boudewijn van Dongen from TU/e and community manager Renato Calzone from JADS. Boudewijn is a professor of computer science and chair of the Process Analytics group at TU/e. This research group distinguishes itself in the field of Information Systems by focusing on the modeling, understanding, analyzing, and improving of processes. Renato is a program manager for 'Data Science for Social Good' and 'Smart Industry' at JADS. JADS provides a collaboration and co-location environment where research, education, and business come together to create value with Data Science and business analytics and to solve complex societal challenges. JADS is facilitated by Tilburg University (TiU) and Technische Universiteit Eindhoven (TU/e).


Boudewijn explains the idea behind the project. “The goal of the CERTIF-AI project was actually to see if we can bring back a part of the manufacturing industry, especially the high-tech manufacturing industry, to the Netherlands or Europe. What is needed for that? We aim to use the large amount of data generated by production equipment and machines to certify processes, improve product quality, and diagnose problems.” Quality controls in manufacturing processes are often destructive and therefore not sustainable. This project explores how, by applying process mining to real-time sensor data, quality control can be performed during the production process, detecting potential issues early on. For industrial end-users, this means fewer errors in production, resulting in lower costs or greater process reliability and, consequently, a better product or service. The result is an AI toolkit, consisting of algorithms and methods, for industrial applications.

The Partners

The project is not only academically innovative by applying AI techniques to real-time sensor data, but JADS and the consortium partners also contribute to the use of AI for concrete industrial applications throughout the project. With algorithms and methods, the process is actually optimized at several partners.

The AI toolkit is developed and implemented by a research team from JADS, Technische Universiteit Eindhoven (TU/e), Tilburg University (TiU), and Hogeschool Utrecht (HU). There is collaboration with four industrial end-users; Damen Shipyards, Omron, Additive Industries, and VTEC.
Sioux Mathware, Bright Cape, UNIT040, and Lambda Function - the solution providers - assist in implementing the toolkit, along with Hogeschool Utrecht and practical researchers from JADS. “These companies also actively contribute; they are keen to incorporate solutions into their portfolios to serve other end-users, contributing to the necessary scaling. Many of these partners are also involved in related projects,” says Renato.

Closeness to Practice

The PhD candidates are making good progress. Various approaches are used within the departments to examine the models and processes and to observe the differences revealed by comparing those models with the actual data collected. The impact of models and their predictive value is also under investigation.

In addition to the fundamental aspect of the research, the actual implementation of solutions is part of CERTIF-AI. This allows companies to genuinely improve their processes. Renato explains: “Working on implementable solutions for end-users is quite a unique approach. On the one hand, we want to shift boundaries on a fundamental level, and at the same time, we want to provide solutions that the industry can benefit from. We closely collaborate with the participating companies, using their data and addressing the questions these companies have; we are deeply rooted in practice. It's a two-way street, highly valuable for our PhD candidates.”

No Easy Path

As in many projects, establishing a good connection between fundamental research and implementation is a challenge. “The end-users are not all at the same level,” Renato explains. “It's not an easy path we're entering; the advantage is that we can learn a lot from each other. Eventually, this is also beneficial because we cannot tackle everything at once. There is sufficient data that the PhD candidates are working with.” The difference between theory and practice is also a challenge, in a positive sense. “As a university, you tend to look at problems mainly from a fundamentally theoretical perspective. Once you enter a company, see real data, and observe a real model, the scale and the difference in scale are so enormous; that is a very important lesson for the PhD candidates,” adds Boudewijn.

Ideal Scenario

Renato: “We have always said that we will deliver a toolkit that allows you to predict whether the processes are heading in the right direction or not. That toolbox consists of best practices, algorithms, but also a kind of plug protocols or plug software that ensures it can actually be implemented or can be implemented at the end-user. Or has been implemented, preferably. On a proof of concept level, of course, we are not a consultancy or IT company.” Boudewijn adds: “So, we are really creating something that helps companies move forward. That is also what distinguishes a Technical University; we like to create things and deliver not only knowledge but also an implementation. A piece of infrastructure on which the next project can build to ultimately - perhaps in 30 years - have a fully automated quality management of production processes.”

For more information, email Boudewijn b.f.v.dongen@tue.nl or Renato r.f.calzone@jads.nl