TurboAgent: An LLM-Driven Autonomous Multi-Agent Framework for Turbomachinery Aerodynamic Design
The aerodynamic design of turbomachinery presents a myriad of challenges due to its complex and tightly coupled multi-stage processes. These processes typically encompass geometry generation, performance prediction, optimization, and high-fidelity physical validation. Traditional approaches in intelligent design have often concentrated on individual stages or relied on loosely coupled pipelines, which complicates the development of fully autonomous, end-to-end design systems. To tackle these challenges, researchers have introduced TurboAgent, a pioneering framework that leverages large language models (LLMs) for autonomous turbomachinery aerodynamic design and optimization.
Overview of TurboAgent
TurboAgent operates as a multi-agent system where the LLM plays a pivotal role in task planning and coordination. This framework integrates specialized agents that are responsible for various functions, such as:
- Generative Design: Utilizing advanced algorithms to create innovative design solutions.
- Rapid Performance Prediction: Quickly simulating performance metrics to inform design decisions.
- Multi-Objective Optimization: Balancing various performance criteria to achieve the best overall design.
- Physics-Based Validation: Ensuring that the designs meet the necessary physical and engineering standards through high-fidelity simulations.
This integrated approach allows TurboAgent to transform traditional trial-and-error methodologies into a more efficient, data-driven collaborative workflow. High-fidelity simulations are retained for final verification, ensuring that the designs are robust and reliable.
Validation and Results
The effectiveness of the TurboAgent framework was validated using a transonic single-rotor compressor, a standard benchmark in turbomachinery design. The results from the TurboAgent framework demonstrated a remarkable alignment with target performance metrics, as evidenced by:
- Coefficients of determination (R²) for mass flow rate, total pressure ratio, and isentropic efficiency all exceeding 0.91.
- Normalized root mean square error (RMSE) values remaining below 8%.
- An optimization agent that enhanced isentropic efficiency by 1.61% and total pressure ratio by 3.02%.
These achievements underline the capability of TurboAgent to facilitate a seamless closed-loop design process, transforming natural language requirements into finalized design outputs.
Conclusion
In conclusion, TurboAgent represents a significant advancement in the field of turbomachinery aerodynamic design. The framework not only streamlines the design process but also enhances scalability and efficiency, making it a valuable tool for engineers and researchers alike. With the ability to execute the complete workflow in approximately 30 minutes under parallel computing, TurboAgent sets a new standard for autonomous design in complex engineering fields.
For further details, refer to the original study available on arXiv:2604.06747v1.