Estimating and analyzing the manufacturing costs of mechanical components is now one of the key points for corporate sustainability in the manufacturing sector. The goal of any industrial setting is to reduce costs and minimize waste: for this reason, this thesis, carried out as part of an internship, aims to analyze, compare, and evaluate the tools used in Cost Engineering. Traditionally, this type of activity is entrusted to specialized software, but with the advent of artificial intelligence and the evolution of Large Language Models (LLMs), the challenge of this work lies in examining the performance, reliability, and effectiveness of models such as Gemini and ChatGPT by comparing them with the predictions made by expert software such as LeanCOST. The analysis was carried out through a case study involving the use of ten mechanical components of different nature, geometry, degree of precision, and required tolerances, subjected to the two different estimators and subsequently analyzed to verify the validity of the results obtained compared to LeanCOST's predictions. The results show overall a percentage mean absolute error of 17% for Gemini and 25% for ChatGPT against an average cost per component of 12.11€ (estimated by LeanCOST) (. By analyzing the “hallucinations” of LLMs, it is possible to deduce a tendency to overestimate costs, especially in the cost items of “setup” and “machining”. In conclusion, it can be stated that LLMs can currently be a significantly helpful support tool if you wish to obtain a preliminary and rapid estimate of the final manufacturing price of a given component; while for a ’more’ detailed and rigorous analysis of all cost items, it is best to rely on robust and advanced software such as LeanCOST.
La preventivazione e l’analisi dei costi di fabbricazione di componenti meccanici rappresentano ormai uno dei punti cardine per la sostenibilità aziendale nel settore manifatturiero. L’obiettivo di ogni realtà industriale è abbattere i costi e minimizzare gli sprechi: per questo motivo, il presente lavoro di tesi, svolto nell’ambito di un’attività di tirocinio, mira ad analizzare, confrontare e valutare gli strumenti utilizzati nell’ambito del Cost Engineering. Tradizionalmente questo tipo di attività viene affidato a software specializzati, ma con l’avvento dell’intelligenza artificiale e dell’evoluzione dei Large Language Models (LLM), la sfida di questo lavoro risiede nell’esaminare le prestazioni, l’affidabilità e l’efficacia di modelli quali Gemini e ChatGPT mettendoli a confronto con le preventivazioni effettuate da un software esperto come LeanCOST. L’analisi è stata effettuata attraverso un caso studio che prevedesse l’utilizzo di dieci componenti meccanici di diversa natura, geometria e grado di precisione e tolleranze richieste, sottoposti ai due diversi estimatori e successivamente analizzati per verificarne la bontà dei risultati ottenuti rispetto alle previsioni di LeanCOST. I risultati evidenziano nel complesso un errore medio assoluto percentuale del 17% per Gemini e del 25% per ChatGPT a fronte di un costo medio per componente di 12,11€ (stimato da LeanCOST) . Analizzando le “allucinazioni” degli LLM, è possibile dedurre una tendenza a sovrastimare i costi, soprattutto nelle voci di costo di “setup” e “machining”. In conclusione, è possibile affermare che gli LLM ad oggi possono essere uno strumento di supporto di notevole aiuto qualora si desideri ottenere una stima preliminare e rapida del prezzo finale di fabbricazione di un determinato componente; mentre per un’ analisi piu’ dettagliata e rigorosa di tutte le voci di costo è bene affidarsi a software robusti e avanzati come LeanCOST.
VALUTAZIONE E CONFRONTO DI STRUMENTI DI PREVENTIVAZIONE DEL COSTO: DEEP LEARNING VS EXPERT SYSTEMS
ALESSANDRINI, JACOPO
2025/2026
Abstract
Estimating and analyzing the manufacturing costs of mechanical components is now one of the key points for corporate sustainability in the manufacturing sector. The goal of any industrial setting is to reduce costs and minimize waste: for this reason, this thesis, carried out as part of an internship, aims to analyze, compare, and evaluate the tools used in Cost Engineering. Traditionally, this type of activity is entrusted to specialized software, but with the advent of artificial intelligence and the evolution of Large Language Models (LLMs), the challenge of this work lies in examining the performance, reliability, and effectiveness of models such as Gemini and ChatGPT by comparing them with the predictions made by expert software such as LeanCOST. The analysis was carried out through a case study involving the use of ten mechanical components of different nature, geometry, degree of precision, and required tolerances, subjected to the two different estimators and subsequently analyzed to verify the validity of the results obtained compared to LeanCOST's predictions. The results show overall a percentage mean absolute error of 17% for Gemini and 25% for ChatGPT against an average cost per component of 12.11€ (estimated by LeanCOST) (. By analyzing the “hallucinations” of LLMs, it is possible to deduce a tendency to overestimate costs, especially in the cost items of “setup” and “machining”. In conclusion, it can be stated that LLMs can currently be a significantly helpful support tool if you wish to obtain a preliminary and rapid estimate of the final manufacturing price of a given component; while for a ’more’ detailed and rigorous analysis of all cost items, it is best to rely on robust and advanced software such as LeanCOST.| File | Dimensione | Formato | |
|---|---|---|---|
|
tesi_finale_Alessandrini Jacopo.pdf
accesso aperto
Dimensione
1.64 MB
Formato
Adobe PDF
|
1.64 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.12075/27314