DesCartes Builder's FDF DSL makes AI digital twin engineering a modeling practice

Digital twins increasingly rely on AI and ML pipelines, but engineering them remains ad-hoc, with hard-to-specify, validate, and reuse pipelines and scarce tooling. To address this, researchers led by Eduardo de Conto (NTU Singapore) developed Function+Data Flow (FDF), a visual domain-specific language that explicitly represents functions (ML models) and their composition, enabling reusable and verifiable pipeline architecture. FDF is implemented in DesCartes Builder, an integrated modeling environment that supports FDF-based digital twin synthesis and validation.

The team conducted an empirical user study where participants built a representative real-time digital twin prototype using DesCartes Builder. Results showed good usability across a broad range of potential users, especially for domain experts—the intended audience. The study also surfaced concrete strengths and areas for improvement, leading the researchers to propose H-FDF, a hierarchical extension of FDF that supports iterative and modular pipelines, enabling formal specification of complex pipelines such as dual training. The findings, submitted to the SoSyM journal, suggest that integrated, model-driven platforms can transform AI-based digital twin engineering from ad-hoc practice into a disciplined modeling discipline, reducing complexity and lowering the barrier for non-AI experts.