HyCont – Hybrid AI methods for industrial process control

“HyCont” stands for “Hybrid AI Methods for Industrial Process Control.” The goal of the project is to develop a transferable framework for the safe, adaptive, and economically optimized control of complex industrial production processes using hybrid AI methods.

The challenge

Rising energy and raw material costs, high quality requirements, and a shortage of specialists are increasing the pressure on companies to operate their facilities more efficiently and with less reliance on individual experience. Since critical process decisions are often still made manually or using simple heuristics, this frequently results in quality fluctuations, scrap, high energy consumption, and unplanned downtime.

Methodological approach

HyCont addresses these challenges by combining model predictive control (MPC), data-driven models, and reinforcement learning (RL) to optimize processes beyond the state of the art.

MPC forms the safety-critical core by explicitly incorporating safety, quality, and process limits into the control model. Data-driven models capture nonlinear relationships and changes in process behavior. Reinforcement learning does not serve as an uncontrolled “black box,” but rather supports the validated adaptation of models and parameters in the event of process drift, disturbances, or changing material properties.

Objectives and benefits

The method is being evaluated in a prototype on an injection molding machine at the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation IOSB. By combining process data, multimodal sensor technology, and quality feedback, a robust closed-loop control system is being developed. The goals include reducing scrap and rework by approximately 5–15%, lowering energy consumption and wear by 5–10%, and minimizing unplanned downtime.

HyCont provides a modular toolchain, including integration guidelines for application in industries such as plastics, metals, chemicals, and food, as well as high-tech manufacturing.

The expert team

The consortium brings together the expertise of Fraunhofer Austria, Fraunhofer IOSB, and Fraunhofer IZFP: Fraunhofer Austria is responsible for the methodological integration of optimization, modeling, and reinforcement learning. Fraunhofer IOSB manages industrial AI engineering, automation, and IT/OT integration, and provides the pilot environment. Fraunhofer IZFP complements the project with expertise in sensor technology, non-destructive testing, data ecosystems, and AI-based quality assessment.