The SCS4AI project focuses on industrial research into an innovative method for AI-driven software development with formally verified results. By combining AI with Supervisory Controller Synthesis (SCS), software for critical systems is automatically verified and corrected. This enhances both reliability and productivity in the development of control software for sectors such as infrastructure and industrial automation.

The challenge: reliability of AI in software development

Safety-critical systems such as bridges, airplanes, and power plants are becoming increasingly complex and reliant on software. Errors in control software can have severe consequences, as seen in incidents involving the Den Uyl Bridge and the Boeing 737 MAX. AI offers a (partial) solution to these challenges and contributes to increased labor productivity. However, error-free output cannot be guaranteed, making formal verification essential. This project explores how AI-driven software development can be combined with SCS to ensure guaranteed safety.

Our approach: combining AI and SCS

The core of this research is the combination of AI-driven software development with SCS. This approach enables the automatic analysis and correction of AI-generated software models, allowing errors to be detected and resolved early. This involves several concrete challenges, such as ensuring a correct semantic translation between models suitable for machine control and SCS, enabling AI to perform model adaptations, and making synthesis results suitable as input for AI. A key requirement is that, despite modifications made by AI to the software model, it must continue to comply with the safety requirements of SCS.

Collaboration and impact

TU/e and Cordis are researching, testing, and validating this technology on a limited scale using the demonstration setup of the Brainport Digital Factory. By utilizing realistic industrial systems, we test the effectiveness and scalability of the approach. This marks an important step toward a broader application in sectors such as energy, manufacturing, and transportation. The results of this project could enable safe AI-driven software development and strengthen the Dutch position in Smart Industry and safe industrial automation.