Tekoälyavusteinen työstökoneiden kunnossapitodatan analyysi : dataohjautuvan kunnossapidon vaatimukset ja mahdollisuudet

The increasing complexity and high utilization rates of industrial production equipment increase maintenance requirements. Reducing production downtime and predicting failures requires up-to-date and reliable information about the condition of the equipment. The aim of this master's thesis is to evaluate the suitability of artificial intelligence-based methods for developing maintenance in the machine tool environment. The work consists of a literature review and a case study. The literature review compiles key maintenance strategies and applications of digital and intelligent maintenance and creates a basis for the methodological choices of the case study. The case study evaluates the suitability of a large language model for structuring free-form failure records of machine tools to support maintenance development. The case study analyzed the failure histories of 18 machine tools. A total of 1768 failure records were classified into 14 predefined failure categories in the Python software environment using OpenAI's ChatGPT API interface and the GPT‑4.1 model. The reliability of the classification was assessed by comparing a random sample of 200 failure descriptions with the classification made by an expert, measuring the agreement between the classifiers using Cohen's kappa. The results show that by utilizing a language model, failure descriptions can be structured into categories, which enables the identification of recurring failures and machine-specific failure profiles. The method used is suitable for maintenance development, but it does not replace expert judgment in individual cases.