Artificial intelligence orchestration for text-based ultrasonic simulation via self-review by multi-large language...

Area of Science:

• Computational Engineering
• Artificial Intelligence
• Ultrasonic Testing

Background:

• Traditional ultrasonic simulation systems often use complex graphical user interfaces (GUIs) or scripting.
• These methods demand significant user time and present accessibility challenges for new users.
• There is a need for more intuitive and efficient control mechanisms in simulation software.

Purpose of the Study:

• To develop a novel text-based control architecture for ultrasonic simulation systems.
• To enhance the accessibility and efficiency of simulation configuration and scenario generation.
• To improve the reliability of AI-driven simulation processes.

Main Methods:

• Modularizing simulation functionalities into discrete functions for text-based control.
• Leveraging a large language model (LLM) for natural language command interpretation.
• Implementing a ground artificial intelligence (AI) approach with self-review and multi-agent collaboration for scenario generation.

Main Results:

• Reduced average simulation configuration time by approximately 75%.
• Significantly decreased the scenario generation error rate from 23.89% to 1.48% using the ground AI approach.
• Demonstrated enhanced reliability and efficiency in ultrasonic simulation control.

Conclusions:

• The proposed text-based control architecture offers an efficient and scalable alternative to traditional GUI-based methods.
• AI-driven approaches, particularly LLMs and ground AI, can substantially reduce operational costs and improve reliability in simulation frameworks.
• This innovation enhances the accessibility of complex simulation systems, benefiting time-sensitive applications like digital twins.