Year: 2026
Type: dissertation
Abstract: This Bachelor's thesis deals with the design and implementation of a dialogue-based control system for a programmable logic controller using large language models. The main objective of the thesis is to create an architecture enabling an operator to issue control commands in natural language, which are subsequently interpreted by an agent layer and safely converted into deterministic operations in the PLC. The thesis describes the design of a two-domain architecture consisting of a Python API domain and a PLC domain on a Siemens S7-1500 device, interconnected via the OPC UA standard. The agent layer utilizes the GPT-4 Turbo and Claude models to extract the operator's intent and generate structured JSON output, which undergoes deterministic validation prior to transmission to the PLC. The thesis further analyses the risks of large language model hallucinations in an industrial environment and proposes a two-layer protection mechanism combining validation at the API level and saturation logic directly in the PLC. The functionality of the prototype was verified by testing on ten simulated pumps, including bulk control, status queries, and a voice interface.
Author: Mičko, Adrián
Language: Slovak
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7165510235