Use cases

Production Optimization

production

Use Case Description

Optimizing Industrial Processes remain a continuous industrial concern, especially with the transition from mass production to customized serial production. With the 4th industrial revolution, many companies are digitalizing their processes that allow to obtain a huge amount of data in different process levels. There are many opportunities for AI to inject intelligence in such optimizations of industrial processes. On the other side, there are also many challenges for AI adoption in this industrial field.

This use case addresses many of these challenges starting form hybrid data acquisition & preparation to a fully deployable & automated AI flow. Furthermore, interactions between design-production-products for process optimization and multi-process production processes will be investigated. In order to generate relevant datasets, formalize the challenges description, and demonstrate AI capabilities, two production infrastructures are linked through the design & production of complex parts. 

Results, Highlights & Research Outputs

Production data architecture and AI-enabled data flow

Optimization of industrial processes requires consistent and reliable data availability. An architecture has been designed that will be used to acquire, understand and integrate data from the adhesive bonding process as well as from the additive manufacturing. This backbone gives us the full data flow from design to end of line quality where different AI agents will be used to demonstrate how AI can optimize multi-step complex manufacturing processes.

Knowledge base and symbolic reasoning consultant

This will also allow to build a knowledge base with information from different levels (expert knowledge, process data, design data, simulation data). An interactive consultant driven by symbolic reasoning was build upon a first version of this knowledge base to support the domain expert in selecting good design parameters. Knowledge graphs were coupled with this interactive consulant to allow user friendly updating of the available knowledge in the knowledge base.

Chatbox for technical-sheet data and real-time quality inspection

In addition, a chatbox has been developed to interact with data coming from technical sheets, where data from both process is being combined. The operator can interact with the chatbox to query technical information. The chatbox combines a large language model (LLM) with Graph-based Table Representation Learning (GTRL) framework. GTRL combines graph-based visual modeling with sequence-based language modeling to learn granular per-cell embeddings that are sensitive to the semantic meaning of cells within their corresponding table context. During the process, AI-enabled in-situ qulaity inspection system is running on an FPGA, fusing data from multiple sources using variational encoders. The FPGA implementation enables real-time process control due to its low latency.

Outputs & Resources

Publications

Brian Booth, Sam Leroux, Dries Verhees, Mathieu Vandecasteele, Cyril Blanc, Abdellatif Bey-Temsamani and Wilfried Philips (2025). Hybrid Data Fusion for Low-Cost High-Speed Monitoring in Powder Bed Fusion 3D Printing. International Conference on Computer and Automation Engineering (ICCAE 2025).

Mathieu Vandecasteele, Domenico Iuso, Milad Hamidi Nasab, Dries Verhees, Joaquim Sanctorum, Mohsen Nourazar, Abdellatif Bey-Temsamani, and Brian Booth. 2024. “High-Speed Deformation Prediction in Selective Laser Melting Using Context-Adaptive Neural Networks.”, 2024 International Conference on Machine Learning and Applications (ICMLA). IEEE. 

Morales-Hernández, A., Rojas Gonzalez, S., Van Nieuwenhuyse, I., Couckuyt, I., Jordens, J., Witters, M., & Van Doninck, B. (2024). Bayesian multi-objective optimization of process design parameters in constrained settings with noise: an engineering design application. Engineering with Computers, 1-15.

Vandevelde, S., Vennekens, J., Jordens, J., Van Doninck, B., & Witters, M. (2023). Knowledge-Based Support for Adhesive Selection: Will it Stick?. Theory and Practice of Logic Programming, 1-21.

Demonstrators / Applications

  • Data architecture for adhesive bonding and additive manufacturing data flow.

  • Knowledge base and interactive consultant driven by symbolic reasoning.

  • Chatbox for querying technical-sheet data.

  • AI-enabled in-situ quality inspection system running on FPGA for real-time process control.

Videos & Media

Some more info on (previous) realization can be found in the videos below.

AI for Quality Control in 3D metal Printing:

AI enabling Adhesive Bonding process:

AI + Domain knowlegde for accurate and robust process optimization: 

Team & Contact

Contact

Cyril Blanc (Additive Manufacturing Expert); Jeroen Jordens (Adhesive Bonding Expert) - Flanders Make IM

Industry Use Cases follow up & Valorization: Wouter De Nijs / Abdel Bey - Flanders Make IM

Principal Investigators / AI Expert Leads

Cyril Blanc - Flanders Make IM

Jeroen Jordens - Flanders Make IM

Research Groups

Flanders Make IM

pr