Harnessing AI for Scalable Analysis of Simulation Data

Engineers face the growing problem of how to analyse the large ensembles of simulation data generated by design optimisation, off-design operability studies, or snapshots of unsteady computations. To address this challenge, this paper presents a tools-focussed approach that enables large language models (LLMs) to orchestrate the analysis pipeline while retaining a trusted, verifiable and reproducible process. Each tool is a processing building block that we formulate, control and understand. The tools are exposed in three ways: a conventional web API, a web-chat application, and a Model Context Protocol (MCP) server. The web-chat application: receives a question from the user as input; uses an LLM to formulate a multi-step plan to answer the question using the tools; executes the plan; and returns an answer or document, generated by an LLM, using the results of the foregoing analysis. The MCP server offers the same tools via the emerging MCP standard, allowing engineers to analyse their data with modern MCP clients in a conversational exploration backed by quantitative findings and linked to a curated set of technical documents. The approach is demonstrated using two case studies: a database of 179 axial turbine rotors; and a database of 590 axial compressor simulations. The first case study uses the web-chat to identify the highest efficiency design and the geometric parameters that correlate with this. In the second case study, the user asks a series of questions using an MCP client-server system. The exploration of the data shows how increasing compound lean reduces endwall loss and, using a relevant academic paper, describes the key aerodynamic mechanisms responsible. The framework described in the paper, providing access to tools that retrieve, process and plot data, represents a scalable strategy for engineers to harness the rapidly growing capabilities of LLMs in the analysis of their data.