Year: 2026
Type: dissertation
Abstract: The integration of artificial intelligence is increasingly pursued in industrial condition monitoring and data management. However, the application of large language models in industrial environments is hindered by their inability to integrate real-time production data with expert analysis without case-specific software development and interface maintenance. This Bachelor’s thesis investigates the applicability of the open Model Context Protocol (MCP) standard to solve this integration problem. The objective is to determine how an MCP server can be utilized as a standardized interface between AI agents and information systems, as well as to evaluate the diagnostic performance of the proposed solution and examine its current limitations. The research was conducted by combining a literature review with prototype development. The literature section proposes an architectural model that connects the application layer to the industrial layer, with the MCP server acting as a mediator. The prototype was implemented as a Python-based MCP server utilizing the FastMCP library and a local database. The prototype was simulated with a large language model in three condition monitoring scenarios: the diagnosis of pipeline blockages, bearing failures, and valve malfunctions. The results indicate that the MCP architecture enhances the integration of AI agents with data sources and supports the identification of failure mechanisms. The AI agent was able to select the appropriate MCP server tools, cross-reference parameters, and consistently diagnose the simulated process disturbances. However, the study identified limitations regarding scalability, maintenance, and cybersecurity. The primary challenges relate to the contradiction between deterministic automation and the probabilistic nature of language models, bottlenecks in transfer protocols, and the tokenization of high-resolution time-series data. Future research should focus on models trained specifically on industrial timeseries data and their associated tokenization methods. The flexibility provided by the MCP server enables the utilization of multiple narrow-domain language models to further improve diagnostic capabilities.
Source: Aaltodoc (Aalto University)
Author: Ranta, Lauri
Language: Finnish
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7162220065