In recent years, Artificial Intelligence (AI) has begun to profoundly transform the Architecture, Engineering, and Construction sector, introducing new forms of automation and human-machine interaction. However, in the field of Heritage Building Information Modeling (HBIM), where informational complexity is high and parameter management remains largely manual, the need emerges to define new methodologies for integrating AI into BIM workflows. Currently available automation solutions present numerous structural limitations. "AI-powered" plugins, while offering operational shortcuts, are mostly vertical, with limited extensibility and rigid functions. Visual programming (e.g. Dynamo) provides broad expressive power, but requires scripting skills that represent an entry barrier for most architects. The research question is therefore: how can AI be effectively integrated into BIM/HBIM workflows while simultaneously ensuring technical power, operational flexibility, accessibility for non-programmers, and professional control? The following thesis proposes the adoption of the Model Context Protocol (MCP) as an innovative approach that overcomes these limitations: the architect formulates requests in natural language, a Large Language Model (LLM) interprets the intent and, through an MCP server, uses specialized tools that invoke Revit APIs. This extensible "bridge" allows composing heterogeneous operations without writing any code, maintaining the flexibility and accessibility of conversational AI. The Model Context Protocol, developed by Anthropic, represents a standardized protocol for connecting conversational artificial intelligences to external technical tools, enabling natural language interactions. The methodological innovation emerges from the necessity to provide AI with domain knowledge, as access to specialized tools alone is not sufficient to obtain a reliable assistant. The use on real projects revealed that the AI, while technically capable of "calling" the tools, lacked procedural knowledge. The solution was the formalization of operational workflows into a structured System Prompt. The designed workflows integrate preventive validation protocols, step-by-step confirmation mechanisms, and complete traceability of modifications for professional accountability. This structured interaction methodology was subsequently formalized in a system prompt that encodes the conceptualized workflows, transforming generic conversational artificial intelligence - characterized by unpredictability and lack of control - into a controllable specialized agent, ensuring predictability of operations and reliability of results in professional contexts. The original open source MCP-Revit project constituted the foundation of the integration that is the subject of this thesis, offering element reading functionalities and simple BIM model querying, without specific tools dedicated to advanced parametric management. The technical implementation therefore required the extension of the project through custom tools specific for parametric management (reading type/instance parameters, creating parameters, updating values, deleting parameters, batch operations with conditional filters, advanced parameter-based queries), validated on the case study of the Mulini San Giorgio in Parco di Monza, a historic complex documented through laser scanner survey and modeled in Autodesk Revit 2026 with structured parameterization. The agent also integrates automatic parameter extraction capabilities from technical datasheets with human validation before BIM population. The critical analysis examines the advantages of the extended system in terms of operational efficiency, professional accessibility, and supervised control, identifying limitations related to cloud service dependency, usage costs, and implications for privacy and professional accountability. The structured comparison with alternatives (Dynamo, manual management, dedicated plugins) highlights the system's distinctive positioning in terms of balance between accessibility, control, and reliability.
Negli ultimi anni l’Intelligenza Artificiale (AI) ha iniziato a trasformare profondamente il settore dell’Architettura, dell’Ingegneria e delle Costruzioni, introducendo nuove forme di automazione e interazione uomo-macchina. Tuttavia, nel campo dell’Heritage Building Information Modeling (HBIM), dove la complessità informativa è elevata e la gestione dei parametri rimane in gran parte manuale, emerge la necessità di definire nuove metodologie per integrare l’AI nei flussi di lavoro BIM. Le soluzioni di automazione oggi disponibili presentano numerosi limiti strutturali. I plugin “AI-powered”, pur offrendo scorciatoie operative, sono per lo più verticali, con funzioni poco estensibili e rigide. Il visual programming (ad es. Dynamo) garantisce invece un’ampia potenza espressiva, ma richiede competenze di scripting che rappresentano una barriera d’ingresso per la maggior parte degli architetti. La domanda di ricerca è quindi: come integrare efficacemente l'AI nei workflow BIM/HBIM garantendo simultaneamente potenza tecnica, flessibilità operativa, accessibilità per non-programmatori e controllo professionale? Il seguente lavoro di tesi propone l'adozione del Model Context Protocol (MCP) come approccio innovativo che supera questi limiti: l'architetto formula richieste in linguaggio naturale, un Large Language Model (LLM) interpreta l'intento e, tramite un server MCP, utilizza dei tool specializzati che invocano le API di Revit. Questo "ponte" estensibile consente di comporre operazioni eterogenee senza scrivere alcun codice, mantenendo la flessibilità e l’accessibilità dell’AI conversazionale. Il Model Context Protocol, sviluppato da Anthropic, rappresenta un protocollo standardizzato per collegare intelligenze artificiali conversazionali a strumenti tecnici esterni, consentendo interazioni in linguaggio naturale. L'innovazione metodologica emerge dalla necessità di fornire all’AI una conoscenza di dominio in quanto il solo accesso a tool specializzati non è sufficiente per ottenere un assistente affidabile. L'uso su progetti reali ha rivelato che l'AI, pur tecnicamente capace di “chiamare” i tool, mancava di conoscenza procedurale. La soluzione è stata la formalizzazione dei workflow operativi in un System Prompt strutturato. I workflow progettati integrano protocolli di validazione preventiva, meccanismi di conferma step-by-step e tracciabilità completa delle modifiche per responsabilità professionale. Questa metodologia di interazione strutturata è stata successivamente formalizzata in un system prompt che codifica i workflow concettualizzati, trasformando l'intelligenza artificiale conversazionale generica — caratterizzata da imprevedibilità e mancanza di controllo — in un agente specializzato controllabile, garantendo prevedibilità delle operazioni e affidabilità dei risultati in contesti professionali. Il progetto originale open source MCP-Revit ha costituito la base dell’integrazione oggetto di questa tesi, offrendo funzionalità di lettura elementi e interrogazione semplice del modello BIM, senza specifici strumenti dedicati alla gestione parametrica avanzata. L'implementazione tecnica ha quindi richiesto l'estensione del progetto mediante sette tools custom specifici per gestione parametrica completa (lettura parametri tipo/istanza, creazione parameteri, aggiornamento valori, eliminazione parametri, operazioni multiple con filtri condizionali, query avanzate su base parametrica), validati sul caso studio dei Mulini San Giorgio nel Parco di Monza, complesso storico documentato mediante rilievo laser scanner e modellato in Autodesk Revit 2026 con parametrizzazione strutturata. L'agente integra inoltre capacità di estrazione automatica di parametri da schede tecniche con validazione umana prima del popolamento BIM. L'analisi critica esamina i vantaggi del sistema esteso in termini di efficienza operativa, accessibilità professionale e controllo supervisionato, identificando i limiti relativi alla dipendenza da servizi cloud, ai costi di utilizzo e alle implicazioni per privacy e responsabilità professionale. Il confronto strutturato con alternative (Dynamo, gestione manuale, plugin dedicati) evidenzia il posizionamento distintivo del sistema in termini di equilibrio tra accessibilità, controllo e affidabilità.
Intelligenza Artificiale e HBIM: integrazione del Model Context Protocol nella gestione parametrica avanzata: il caso studio dei Mulini San Giorgio
Scognamiglio, Valeria;Napoli, Giuseppe
2024/2025
Abstract
In recent years, Artificial Intelligence (AI) has begun to profoundly transform the Architecture, Engineering, and Construction sector, introducing new forms of automation and human-machine interaction. However, in the field of Heritage Building Information Modeling (HBIM), where informational complexity is high and parameter management remains largely manual, the need emerges to define new methodologies for integrating AI into BIM workflows. Currently available automation solutions present numerous structural limitations. "AI-powered" plugins, while offering operational shortcuts, are mostly vertical, with limited extensibility and rigid functions. Visual programming (e.g. Dynamo) provides broad expressive power, but requires scripting skills that represent an entry barrier for most architects. The research question is therefore: how can AI be effectively integrated into BIM/HBIM workflows while simultaneously ensuring technical power, operational flexibility, accessibility for non-programmers, and professional control? The following thesis proposes the adoption of the Model Context Protocol (MCP) as an innovative approach that overcomes these limitations: the architect formulates requests in natural language, a Large Language Model (LLM) interprets the intent and, through an MCP server, uses specialized tools that invoke Revit APIs. This extensible "bridge" allows composing heterogeneous operations without writing any code, maintaining the flexibility and accessibility of conversational AI. The Model Context Protocol, developed by Anthropic, represents a standardized protocol for connecting conversational artificial intelligences to external technical tools, enabling natural language interactions. The methodological innovation emerges from the necessity to provide AI with domain knowledge, as access to specialized tools alone is not sufficient to obtain a reliable assistant. The use on real projects revealed that the AI, while technically capable of "calling" the tools, lacked procedural knowledge. The solution was the formalization of operational workflows into a structured System Prompt. The designed workflows integrate preventive validation protocols, step-by-step confirmation mechanisms, and complete traceability of modifications for professional accountability. This structured interaction methodology was subsequently formalized in a system prompt that encodes the conceptualized workflows, transforming generic conversational artificial intelligence - characterized by unpredictability and lack of control - into a controllable specialized agent, ensuring predictability of operations and reliability of results in professional contexts. The original open source MCP-Revit project constituted the foundation of the integration that is the subject of this thesis, offering element reading functionalities and simple BIM model querying, without specific tools dedicated to advanced parametric management. The technical implementation therefore required the extension of the project through custom tools specific for parametric management (reading type/instance parameters, creating parameters, updating values, deleting parameters, batch operations with conditional filters, advanced parameter-based queries), validated on the case study of the Mulini San Giorgio in Parco di Monza, a historic complex documented through laser scanner survey and modeled in Autodesk Revit 2026 with structured parameterization. The agent also integrates automatic parameter extraction capabilities from technical datasheets with human validation before BIM population. The critical analysis examines the advantages of the extended system in terms of operational efficiency, professional accessibility, and supervised control, identifying limitations related to cloud service dependency, usage costs, and implications for privacy and professional accountability. The structured comparison with alternatives (Dynamo, manual management, dedicated plugins) highlights the system's distinctive positioning in terms of balance between accessibility, control, and reliability.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tavole di progetto.pdf
accessibile in internet per tutti
Descrizione: Tavole di progetto
Dimensione
889.74 MB
Formato
Adobe PDF
|
889.74 MB | Adobe PDF | Visualizza/Apri |
|
Tesi.pdf
accessibile in internet per tutti
Descrizione: Relazione
Dimensione
1.04 GB
Formato
Adobe PDF
|
1.04 GB | 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/10589/247558