Enhancing maintenance strategies for elevators through fine-tuning large language models

Abstract

This thesis investigates how elevator maintenance strategies can be improved using parameter-efficient approaches such as Low-Rank Adaptation (LoRA) to fine-tune large language models (LLMs). It explores several methods for fine-tuning these models for domain-specific applications with the goal of enhancing precision and lowering computing overhead in AI-driven maintenance insights. By integrating specialized datasets linked to elevator systems, the research explores how models might be tuned to deliver accurate diagnoses, boost maintenance strategies, and limit potential AI hallucinations that could threaten safety. The research also highlights how crucial it is to strike a balance between model accuracy and scalability, efficiency, and meet a range of operational requirements.