Businesses that rely heavily on machinery encounter a significant obstacle in effectively managing equipment repairs while optimizing technician efficiency. The unpredictable nature of repair requests and potential miscommunications can lead to a strain on technical teams, resulting in delayed repairs and decreased production. This thesis proposes a novel solution to this problem: the development of a specialized chatbot powered by Large Language Models (LLMs) that is designed to assist both technicians and general employees with equipment troubleshooting. The chatbot empowers employees to attempt straightforward repairs independently, reducing unnecessary escalations to technicians and allowing them to prioritize critical issues. We examine the evolution of LLMs, focusing on Mistral and Llama2, and how they can be adapted through fine-tuning. Techniques such as quantization and LoRA are explored for their potential to streamline deployment on less powerful hardware. Successful fine-tuning on a single A100 GPU demonstrates the feasibility of adapting these models to a specialized domain—in this case, equipment troubleshooting. This research demonstrates the potential of LLMs to enhance operational efficiency in manufacturing and other equipment-intensive industries.
Businesses that rely heavily on machinery encounter a significant obstacle in effectively managing equipment repairs while optimizing technician efficiency. The unpredictable nature of repair requests and potential miscommunications can lead to a strain on technical teams, resulting in delayed repairs and decreased production. This thesis proposes a novel solution to this problem: the development of a specialized chatbot powered by Large Language Models (LLMs) that is designed to assist both technicians and general employees with equipment troubleshooting. The chatbot empowers employees to attempt straightforward repairs independently, reducing unnecessary escalations to technicians and allowing them to prioritize critical issues. We examine the evolution of LLMs, focusing on Mistral and Llama2, and how they can be adapted through fine-tuning. Techniques such as quantization and LoRA are explored for their potential to streamline deployment on less powerful hardware. Successful fine-tuning on a single A100 GPU demonstrates the feasibility of adapting these models to a specialized domain—in this case, equipment troubleshooting. This research demonstrates the potential of LLMs to enhance operational efficiency in manufacturing and other equipment-intensive industries.
Adaptation of Large Language Models to assistant chat-bots for industrial plants
FARHANGIAN, MOHAMMADARDALAN
2023/2024
Abstract
Businesses that rely heavily on machinery encounter a significant obstacle in effectively managing equipment repairs while optimizing technician efficiency. The unpredictable nature of repair requests and potential miscommunications can lead to a strain on technical teams, resulting in delayed repairs and decreased production. This thesis proposes a novel solution to this problem: the development of a specialized chatbot powered by Large Language Models (LLMs) that is designed to assist both technicians and general employees with equipment troubleshooting. The chatbot empowers employees to attempt straightforward repairs independently, reducing unnecessary escalations to technicians and allowing them to prioritize critical issues. We examine the evolution of LLMs, focusing on Mistral and Llama2, and how they can be adapted through fine-tuning. Techniques such as quantization and LoRA are explored for their potential to streamline deployment on less powerful hardware. Successful fine-tuning on a single A100 GPU demonstrates the feasibility of adapting these models to a specialized domain—in this case, equipment troubleshooting. This research demonstrates the potential of LLMs to enhance operational efficiency in manufacturing and other equipment-intensive industries.File | Dimensione | Formato | |
---|---|---|---|
Farhangian_Mohammadardalan.pdf
accesso aperto
Dimensione
1.87 MB
Formato
Adobe PDF
|
1.87 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/64608