Building construction constitutes a significant portion of the construction industry. It accounts, for example, approximately 25% of the value added in the EU construction sector and more than 50% in Australia. Nevertheless, building non-conformities, more simply, “building defects” remain a primary cause of delays, cost overruns, and litigation, to the extent that the management of contractors’ claims and clients’ complaints, at the end of works, has become an accepted part of the building process. This suggests that the Architecture, Engineering, and Construction (AEC) sector is unable to implement simple procedures allowing stakeholders to learn from the past and prevent defect recurrence. Building defects directly or indirectly caused by design choices have, as everybody knows, a higher potential for rectification than defects caused by construction errors since their early stage correction requires only the updating of the design model, which is significantly less costly than rework during building construction or retrofitting during use. In this regard, the disciplines of project, construction, and quality management applied to a building project provide several useful approaches, including risk analysis and design review, to validate design choices in terms of their effectiveness in satisfying established requirements, the constructability and the maintainability of the building, and the environmental, economic, and social sustainability of their impacts, in every phase of its life cycle. However, this is not sufficient. The life cycle of a building asset is too long, its delivery chain too complex, and privacy requirements too sensitive. As a result, negative experiences from one project are not transferred as a normal practice to subsequent projects through the systematic extraction, sharing, and reuse of knowledge derived from adverse events. Nevertheless, such knowledge could make future projects more effective and efficient and could contribute to improving productivity in the construction sector. This research addresses one of the possible sources of such knowledge, which has been substantially undervalued and underexploited to date, namely the design inspection reports documenting the results of the preventive design verification process that has been mandatory for Italian public works projects for more than thirty years. Since 1994, following a practice that reflects the principles of quality management systems, it has been mandatory to verify the adopted design choices and the completeness of the project documentation before it is used as the basis for tendering. This verification must be carried out by an independent body and must be systematically organized and reported to the stakeholders involved. To exploit the knowledge embedded in the design inspection reports of individual projects, this thesis proposes a methodology structured around four main components: (1) an information model for representing unstructured design defect data in a structured and standardized format; (2) an LLM-based approach for automating information extraction and classification; (3) statistical analysis to measure design quality and extract knowledge on building design defect patterns across design documentation and building elements; and (4) the transformation of structured design defect information into verification actions to support improvements in design and design review, together with their integration into BIM environments and the automation of their execution.
L'edilizia costituisce una quota significativa dell’industria delle costruzioni. Per esempio, rappresenta circa il 25% del valore aggiunto del settore, nell’Unione Europea, e oltre il 50% in Australia. Nonostante ciò, la non conformità, più semplicemente i “difetti” delle realizzazioni rimane una delle principali cause di ritardo, superamento dei budget e comunque di contenzioso, al punto che la gestione delle riserve e delle contestazioni a fine lavori è diventata una componente accettata del processo edilizio che affianca il processo di loro consegna all’utente. Si può concludere che il settore dell’architettura, dell’ingegneria e delle costruzioni (Architecture, Engineering, and Construction, AEC), non sia in grado di attuare normali procedure che, apprendendo dagli errori del passato, consentirebbero a tutti gli operatori di prevenire i difetti che essi causano. I difetti direttamente o indirettamente legati a scelte progettuali presentano, come tutti sanno, un potenziale di correzione molto alto, rispetto a quelli causati da errori di realizzazione, in quanto la loro correzione anticipata comporta solo il costo dell’aggiornamento delle scelte progettuali, di molto inferiore ai costi associati alle rilavorazioni durante la costruzione o all’adeguamento in fase d'uso. A questo proposito, le discipline del project, construction e quality management applicate a un progetto edilizio, forniscono diversi spunti, tra i quali quelli dell’analisi dei rischi e della revisione delle scelte progettuali, ovvero della sua validazione sistematica in termini di efficacia nel soddisfacimento dei requisiti stabiliti, la costruibilità e la manutenibilità dell’edificio e la sostenibilità complessiva: ambientale, economica e sociale dei suoi impatti, in tutte le fasi del suo ciclo di vita. Ciò non basta, tuttavia: il ciclo di vita di un manufatto edilizio è troppo lungo, la catena della sua realizzazione troppo complessa e le esigenze di privacy troppo delicate, perché le esperienze negative nella realizzazione di un progetto vengano naturalmente messe a servizio dei successivi, estraendo, condividendo e riutilizzando sistematicamente la conoscenza derivata da eventi negativi, nonostante ciò, renderebbero i futuri progetti più efficaci ed efficienti, e la produttività nel settore in crescita. Questa ricerca affronta una delle possibili fonti di questa conoscenza, sostanzialmente sottovalutata e non sfruttata, fino ad oggi, ovvero il risultato del complesso di verifiche preventive, obbligatorie per ogni progetto italiano di lavori pubblici da oltre trent’anni. Dal 1994, infatti, secondo una prassi che riecheggia le regole del sistema di gestione della qualità, è obbligatoria la verifica delle scelte progettuali adottate e della completezza prima che la documentazione di progetto sia utilizzata come base per un appalto, che tale operazione sia svolta da un ente indipendente, organizzata con sistematicità e in maniera altrettanto sistematico rendicontata agli operatori coinvolti. Per sfruttare ogni spunto derivante dalla verbalizzazione delle analisi relative a un singolo progetto, questa tesi propone una metodologia articolata nella messa a punto di: (1) un modello informativo per rappresentare dati non strutturati sui difetti di progettazione in un formato strutturato e standardizzato; (2) un approccio basato su LLM per automatizzare l’estrazione e la classificazione delle informazioni; (3) un’analisi statistica per misurare la qualità della progettazione ed estrarre conoscenza sui pattern dei difetti di progettazione edilizia attraverso la documentazione progettuale e gli elementi edilizi; e (4) la trasformazione delle informazioni strutturate sui difetti di progettazione in azioni di verifica, al fine di supportare il miglioramento della progettazione e della revisione progettuale, insieme alla loro integrazione negli ambienti BIM e all’automazione della loro esecuzione.
Design inspection reports as a source of quality metrics and defect prevention measures: Large Language Model (LLM) - based knowledge extraction to improve design quality and design review processes in the construction sector
Elshaboury, Hamada Ahmed Mohamed Kassem
2025/2026
Abstract
Building construction constitutes a significant portion of the construction industry. It accounts, for example, approximately 25% of the value added in the EU construction sector and more than 50% in Australia. Nevertheless, building non-conformities, more simply, “building defects” remain a primary cause of delays, cost overruns, and litigation, to the extent that the management of contractors’ claims and clients’ complaints, at the end of works, has become an accepted part of the building process. This suggests that the Architecture, Engineering, and Construction (AEC) sector is unable to implement simple procedures allowing stakeholders to learn from the past and prevent defect recurrence. Building defects directly or indirectly caused by design choices have, as everybody knows, a higher potential for rectification than defects caused by construction errors since their early stage correction requires only the updating of the design model, which is significantly less costly than rework during building construction or retrofitting during use. In this regard, the disciplines of project, construction, and quality management applied to a building project provide several useful approaches, including risk analysis and design review, to validate design choices in terms of their effectiveness in satisfying established requirements, the constructability and the maintainability of the building, and the environmental, economic, and social sustainability of their impacts, in every phase of its life cycle. However, this is not sufficient. The life cycle of a building asset is too long, its delivery chain too complex, and privacy requirements too sensitive. As a result, negative experiences from one project are not transferred as a normal practice to subsequent projects through the systematic extraction, sharing, and reuse of knowledge derived from adverse events. Nevertheless, such knowledge could make future projects more effective and efficient and could contribute to improving productivity in the construction sector. This research addresses one of the possible sources of such knowledge, which has been substantially undervalued and underexploited to date, namely the design inspection reports documenting the results of the preventive design verification process that has been mandatory for Italian public works projects for more than thirty years. Since 1994, following a practice that reflects the principles of quality management systems, it has been mandatory to verify the adopted design choices and the completeness of the project documentation before it is used as the basis for tendering. This verification must be carried out by an independent body and must be systematically organized and reported to the stakeholders involved. To exploit the knowledge embedded in the design inspection reports of individual projects, this thesis proposes a methodology structured around four main components: (1) an information model for representing unstructured design defect data in a structured and standardized format; (2) an LLM-based approach for automating information extraction and classification; (3) statistical analysis to measure design quality and extract knowledge on building design defect patterns across design documentation and building elements; and (4) the transformation of structured design defect information into verification actions to support improvements in design and design review, together with their integration into BIM environments and the automation of their execution.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_06_Elshaboury.pdf
accessibile in internet per tutti a partire dal 09/06/2029
Dimensione
19 MB
Formato
Adobe PDF
|
19 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/258738