Explainable and generalizable AI-driven multiscale informatics for dynamic system modelling.

Area of Science:

• Engineering
• Artificial Intelligence
• System Dynamics

Background:

• Existing AI and first-principle models lack explainability and data fidelity for demanding industrial applications.
• Ultra-precision machining requires advanced system modeling that balances interpretability with accuracy.

Purpose of the Study:

• To develop an explainable and generalizable 'grey-box' AI informatics method for real-world dynamic system modeling.
• To create a multiscale 'world model' integrating physical principles and data-driven approaches.

Main Methods:

• Developed a grey-box AI model integrating a white-box architecture (physical principles) with black-box data-fitting components.
• The physical principles provide an explainable global structure, while black boxes improve local accuracy using training data.
• Encapsulated implicit variables and relationships missed by standalone models.

Main Results:

• The grey-box method demonstrated superior performance compared to existing modeling techniques.
• Validated through a case study on an industrial cleanroom high-precision temperature regulation system.
• The model proved suitable for diverse operating conditions.

Conclusions:

• The developed grey-box AI method offers a robust solution for dynamic system modeling in industrial settings.
• This approach enhances both explainability and data fidelity, addressing limitations of current methods.
• The model's generalizability makes it applicable to various real-world systems.