Industry 5.0 promotes human-centric manufacturing, where operators play a key role in decision-making and process optimization. While Digital Twins are effective in modeling mechanical systems, their application to human behavior remains underexplored due to its complexity and non-determinism. This thesis addresses that gap by proposing an automated method to generate Human Digital Twins (HDTs) via event-based behavioral modeling in industrial workflows. The proposed framework integrates three core technologies: knowledge graphs for semantic data integration, process mining for pattern discovery, and automata learning for formal behavioral representation. The methodology leverages the L*_SHA algorithm, a variant of the L* algorithm tailored to Stochastic Hybrid Automata, to extract behavioral models directly from enterprise data without additional sensors or manual configuration. A case study at CROMA, a medical sterilization facility in the AUTO-TWIN project, validated the approach using real operational logs. The system extracted hundreds of distinct behavioral patterns and generated formal models reflecting both individual operator flexibility and system-wide constraints. Temporal inconsistencies and overlapping activities were addressed through innovative combined-event modeling. Model validation using UPPAAL Statistical Model Checking confirmed that the automata satisfy key temporal and probabilistic properties with high accuracy. The integration of timing constraints enabled realistic estimations of operation durations, supporting advanced behavioral simulation and scheduling. This work contributes: (1) a fully automated behavioral modeling methodology, (2) formal models suitable for verification, (3) effective reuse of historical data, and (4) time-aware analysis to support operational optimization. The results demonstrate that Human Digital Twins can achieve both theoretical rigor and practical impact in human-centered industrial systems.
L’Industria 5.0 promuove una manifattura centrata sull’uomo, dove gli operatori hanno un ruolo chiave nelle decisioni e nell’ottimizzazione. Sebbene i Digital Twin siano efficaci nella modellazione di sistemi meccanici, resta poco esplorata la loro estensione al comportamento umano, data la sua complessità e natura non deterministica. Questa tesi affronta tale lacuna proponendo un metodo automatico per generare Human Digital Twins (HDT) tramite modellazione comportamentale basata su eventi nei flussi industriali. Il framework integra tre tecnologie principali: knowledge graph per l’integrazione semantica dei dati, process mining per la scoperta di pattern ricorrenti, e automata learning per la rappresentazione formale dei comportamenti. La metodologia si basa sull’algoritmo L*_SHA, una variante del L* adattata agli Stochastic Hybrid Automata, per estrarre modelli direttamente dai dati aziendali, senza sensori aggiuntivi o configurazioni manuali. Uno studio di caso presso CROMA, impianto di sterilizzazione medicale nel progetto AUTO-TWIN, ha validato l’approccio su log reali. Il sistema ha estratto centinaia di pattern distinti e generato modelli formali che riflettono sia la flessibilità degli operatori sia i vincoli del sistema. Inconsistenze temporali e attività sovrapposte sono state gestite tramite una modellazione innovativa di eventi combinati. La validazione con UPPAAL Statistical Model Checking ha confermato che gli automi soddisfano proprietà temporali e probabilistiche critiche con alta accuratezza. L’integrazione dei vincoli temporali ha permesso stime realistiche delle durate operative, supportando simulazioni comportamentali e pianificazione avanzata. Questa tesi contribuisce con: (1) una metodologia completamente automatica per la modellazione comportamentale, (2) modelli formali adatti alla verifica, (3) riutilizzo efficace dei dati storici, e (4) analisi sensibili al tempo per l’ottimizzazione operativa. I risultati dimostrano che gli Human Digital Twin possono coniugare rigore teorico e impatto pratico nei sistemi industriali centrati sull’uomo.
Learning automata models of operator activity for human digital twin construction
CALTABIANO, FEDERICA
2024/2025
Abstract
Industry 5.0 promotes human-centric manufacturing, where operators play a key role in decision-making and process optimization. While Digital Twins are effective in modeling mechanical systems, their application to human behavior remains underexplored due to its complexity and non-determinism. This thesis addresses that gap by proposing an automated method to generate Human Digital Twins (HDTs) via event-based behavioral modeling in industrial workflows. The proposed framework integrates three core technologies: knowledge graphs for semantic data integration, process mining for pattern discovery, and automata learning for formal behavioral representation. The methodology leverages the L*_SHA algorithm, a variant of the L* algorithm tailored to Stochastic Hybrid Automata, to extract behavioral models directly from enterprise data without additional sensors or manual configuration. A case study at CROMA, a medical sterilization facility in the AUTO-TWIN project, validated the approach using real operational logs. The system extracted hundreds of distinct behavioral patterns and generated formal models reflecting both individual operator flexibility and system-wide constraints. Temporal inconsistencies and overlapping activities were addressed through innovative combined-event modeling. Model validation using UPPAAL Statistical Model Checking confirmed that the automata satisfy key temporal and probabilistic properties with high accuracy. The integration of timing constraints enabled realistic estimations of operation durations, supporting advanced behavioral simulation and scheduling. This work contributes: (1) a fully automated behavioral modeling methodology, (2) formal models suitable for verification, (3) effective reuse of historical data, and (4) time-aware analysis to support operational optimization. The results demonstrate that Human Digital Twins can achieve both theoretical rigor and practical impact in human-centered industrial systems.File | Dimensione | Formato | |
---|---|---|---|
2025_07_Caltabiano_Thesis_01.pdf
accessibile in internet per tutti
Descrizione: Thesis
Dimensione
1.16 MB
Formato
Adobe PDF
|
1.16 MB | Adobe PDF | Visualizza/Apri |
2025_07_Caltabiano_Executive_Summary_02.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
462.79 kB
Formato
Adobe PDF
|
462.79 kB | 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/10589/240696