In modern industrial manufacturing, unexpected machine breakdowns cause significant production delays and high financial losses. Although manufacturing plants regularly record maintenance activities in digital databases, the practical experience needed to diagnose and solve complex technical problems is often fragmented across different factories and concentrated in the minds of a few senior technicians. When a breakdown occurs, technicians often have to diagnose the issue from scratch, even if an identical problem has already been solved in another company facility. Sharing know-how properly across all the plants of the company would bring a lot of benefits. This thesis investigates how artificial intelligence (AI) can help bridge this gap by transforming historical maintenance archives into a shared, centralised knowledge tool. Specifically, the research develops and evaluates a system that combines historical repair records with modern language models to automatically provide technicians with clear, step-by-step troubleshooting procedures whenever a fault occurs, with the goal of decreasing line downtime. The proposed approach is evaluated using realistic maintenance records from multiple manufacturing plants, the experimental validation of the approach is carried out using a set of 500 maintenance tickets collected from three different plants. The analysis focuses on verifying whether the system can reliably identify the correct historical precedents, assist technicians during emergencies and operate safely without inventing incorrect advice. To evaluate knowledge fragmentation across different plants, different analyses are carried out in order to establish the proportion of recurring failures for which there are known solutions in other company locations. The findings indicate that knowledge retrieval from tickets can effectively unlock distributed maintenance experience, significantly reduce diagnostic time and decrease machine downtime. Furthermore, an economic model evaluates the business feasibility and financial return of adopting this technology. The results show a strong economic return, confirming that this solution represents a viable decision-support tool to standardise technical knowledge, improve operational resilience and preserve company expertise over time.

In modern industrial manufacturing, unexpected machine breakdowns cause significant production delays and high financial losses. Although manufacturing plants regularly record maintenance activities in digital databases, the practical experience needed to diagnose and solve complex technical problems is often fragmented across different factories and concentrated in the minds of a few senior technicians. When a breakdown occurs, technicians often have to diagnose the issue from scratch, even if an identical problem has already been solved in another company facility. Sharing know-how properly across all the plants of the company would bring a lot of benefits. This thesis investigates how artificial intelligence (AI) can help bridge this gap by transforming historical maintenance archives into a shared, centralised knowledge tool. Specifically, the research develops and evaluates a system that combines historical repair records with modern language models to automatically provide technicians with clear, step-by-step troubleshooting procedures whenever a fault occurs, with the goal of decreasing line downtime. The proposed approach is evaluated using realistic maintenance records from multiple manufacturing plants, the experimental validation of the approach is carried out using a set of 500 maintenance tickets collected from three different plants. The analysis focuses on verifying whether the system can reliably identify the correct historical precedents, assist technicians during emergencies and operate safely without inventing incorrect advice. To evaluate knowledge fragmentation across different plants, different analyses are carried out in order to establish the proportion of recurring failures for which there are known solutions in other company locations. The findings indicate that knowledge retrieval from tickets can effectively unlock distributed maintenance experience, significantly reduce diagnostic time and decrease machine downtime. Furthermore, an economic model evaluates the business feasibility and financial return of adopting this technology. The results show a strong economic return, confirming that this solution represents a viable decision-support tool to standardise technical knowledge, improve operational resilience and preserve company expertise over time.

From Site-Specific Technical Records to Distributed Expertise: Enhancing Industrial Knowledge Management through Generative AI

BORDIGNON, THOMAS
2025/2026

Abstract

In modern industrial manufacturing, unexpected machine breakdowns cause significant production delays and high financial losses. Although manufacturing plants regularly record maintenance activities in digital databases, the practical experience needed to diagnose and solve complex technical problems is often fragmented across different factories and concentrated in the minds of a few senior technicians. When a breakdown occurs, technicians often have to diagnose the issue from scratch, even if an identical problem has already been solved in another company facility. Sharing know-how properly across all the plants of the company would bring a lot of benefits. This thesis investigates how artificial intelligence (AI) can help bridge this gap by transforming historical maintenance archives into a shared, centralised knowledge tool. Specifically, the research develops and evaluates a system that combines historical repair records with modern language models to automatically provide technicians with clear, step-by-step troubleshooting procedures whenever a fault occurs, with the goal of decreasing line downtime. The proposed approach is evaluated using realistic maintenance records from multiple manufacturing plants, the experimental validation of the approach is carried out using a set of 500 maintenance tickets collected from three different plants. The analysis focuses on verifying whether the system can reliably identify the correct historical precedents, assist technicians during emergencies and operate safely without inventing incorrect advice. To evaluate knowledge fragmentation across different plants, different analyses are carried out in order to establish the proportion of recurring failures for which there are known solutions in other company locations. The findings indicate that knowledge retrieval from tickets can effectively unlock distributed maintenance experience, significantly reduce diagnostic time and decrease machine downtime. Furthermore, an economic model evaluates the business feasibility and financial return of adopting this technology. The results show a strong economic return, confirming that this solution represents a viable decision-support tool to standardise technical knowledge, improve operational resilience and preserve company expertise over time.
2025
From Site-Specific Technical Records to Distributed Expertise: Enhancing Industrial Knowledge Management through Generative AI
In modern industrial manufacturing, unexpected machine breakdowns cause significant production delays and high financial losses. Although manufacturing plants regularly record maintenance activities in digital databases, the practical experience needed to diagnose and solve complex technical problems is often fragmented across different factories and concentrated in the minds of a few senior technicians. When a breakdown occurs, technicians often have to diagnose the issue from scratch, even if an identical problem has already been solved in another company facility. Sharing know-how properly across all the plants of the company would bring a lot of benefits. This thesis investigates how artificial intelligence (AI) can help bridge this gap by transforming historical maintenance archives into a shared, centralised knowledge tool. Specifically, the research develops and evaluates a system that combines historical repair records with modern language models to automatically provide technicians with clear, step-by-step troubleshooting procedures whenever a fault occurs, with the goal of decreasing line downtime. The proposed approach is evaluated using realistic maintenance records from multiple manufacturing plants, the experimental validation of the approach is carried out using a set of 500 maintenance tickets collected from three different plants. The analysis focuses on verifying whether the system can reliably identify the correct historical precedents, assist technicians during emergencies and operate safely without inventing incorrect advice. To evaluate knowledge fragmentation across different plants, different analyses are carried out in order to establish the proportion of recurring failures for which there are known solutions in other company locations. The findings indicate that knowledge retrieval from tickets can effectively unlock distributed maintenance experience, significantly reduce diagnostic time and decrease machine downtime. Furthermore, an economic model evaluates the business feasibility and financial return of adopting this technology. The results show a strong economic return, confirming that this solution represents a viable decision-support tool to standardise technical knowledge, improve operational resilience and preserve company expertise over time.
Industrial Knowledge
Knowledge Management
Distributed Expertis
Generative AI
Manufacturing
File in questo prodotto:
File Dimensione Formato  
thomas_bordignon_2157907.pdf

embargo fino al 08/10/2029

Dimensione 3.94 MB
Formato Adobe PDF
3.94 MB Adobe PDF

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

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/116729