Speaker
Description
In the life cycle of marine engineering systems, the system performance may differ across different structural and load parameters. To identify the sensitivity, Sobol's global sensitivity analysis (GSA) of the system parameters can be practiced. Possibilty of parameter changes can be quantified and ranked with the Sobol indices decomposed from the total variance. The traditional Sobol's GSA requires computationally expensive and time-consuming Monte-Carlo simulations. This study firstly adopts a Machine-learning-based surrogate model for intensive Monte-Carlo tasks, and then sensitivity indices are obtained in an efficient fashion.
Furthermore, to overcome the obtaining and readness difficulty of sensitivity results
at the workspace, a large language model (LLM) engaged reporting framework is proposed in this study. The surrogate-based Sobol analysis can be executed by the user at the workspace in a human-trusted manner, while the LLM-agent is engaged to perform the sensitivity task and interpret the sensitivity results to the report development. The introduced LLM-agent-engaged framework is deemed as a viable solution for improving the quality and efficiency in diagnostic reporting. This approach sets a foundation of the Human-AI-in-the-loop (H-A-I-L) framework for smart engineering solutions.
| Topic Area | Advanced Computational Methods and Applications in Marine, Subsea and Offshore Technology |
|---|