Precision in laser-based manufacturing depends on several process parameters, so accurate predictive modeling is key to reducing errors and improving results. This thesis presents a method to build the foundation for a Digital Twin (DT) in laser-based manufacturing. The aim is to find the best way to estimate the engraving width produced by a low-power laser, based on controllable parameters. A virtual environment was created in Unity to allow real-time monitoring and virtual testing, helping reduce time and material use. The physical setup includes an educational robotic arm used to run controlled experiments under different conditions. To estimate engraving width, four different modelling strategies were developed. Three analytical models were based on physical laser theory, each combining equations with experimental data and differing in assumptions and required tests. The fourth model uses Extended Kernel Regression (EKR) to improve the predictions from the analytical models, increasing accuracy and reliability. This study explores different modeling approaches to assess their effectiveness in predicting engraving width under various experimental conditions. To build the analytical models and for the validation stage, a structured dataset was built with variations in laser power, speed, and defocusing distance. The models were tested in two stages: linear path experiments with combined parameter changes, and engraving tests with complex shapes like squares, triangles, and spirals. Results show that the analytical model accounting for beam shape produces highly accurate estimations, demonstrating how important this factor is. The inclusion of EKR in Model 4 further enhances prediction consistency, making it more adaptable to various engraving conditions. However, experimental data reveals that to further improve performance, additional training data is necessary, as certain variations in process parameters still present challenges in maintaining accuracy. This methodology contributes to the advancement of laser-based manufacturing within a Digital Twin environment, providing a foundation for integrating predictive modeling into real-time simulation.
La precisione nella produzione basata su laser dipende da diversi parametri di processo, per cui una modellazione predittiva accurata è essenziale per ridurre gli errori e migliorare i risultati. Questa tesi presenta un metodo per costruire le basi di un Digital Twin (DT) nella produzione laser. L’obiettivo è identificare l’approccio più adatto per stimare la larghezza dell’incisione generata da un laser a bassa potenza, in funzione di parametri controllabili. È stato sviluppato un ambiente virtuale in Unity per consentire il monitoraggio in tempo reale e la sperimentazione virtuale, riducendo tempi e consumo di materiali. Il sistema fisico include un braccio robotico educativo utilizzato per condurre esperimenti controllati in diverse condizioni operative. Per stimare la larghezza dell’incisione, sono state sviluppate quattro strategie di modellazione. Tre modelli analitici si basano sulla teoria fisica del laser, combinando equazioni con dati sperimentali e differenziandosi per ipotesi e test richiesti. Il quarto modello utilizza l’Extended Kernel Regression (EKR) per migliorare le previsioni dei modelli analitici, aumentando precisione e affidabilità. È stato costruito un dataset strutturato con variazioni di potenza, velocità e distanza di defocalizzazione. I modelli sono stati testati in due fasi: esperimenti su percorsi lineari con variazioni combinate e test di incisione con forme complesse come quadrati, triangoli e spirali. I risultati mostrano che il modello analitico che considera la forma del fascio produce stime molto accurate, evidenziando l’importanza di questo fattore. L’integrazione dell’EKR nel Modello 4 migliora ulteriormente la coerenza delle previsioni, rendendolo più adattabile a diverse condizioni di incisione. Tuttavia, i dati sperimentali indicano che sono necessari ulteriori dati di addestramento, poiché alcune variazioni dei parametri continuano a influenzare la precisione. Questa metodologia contribuisce allo sviluppo della produzione laser in un ambiente DT, fornendo una base per integrare la modellazione predittiva nella simulazione in tempo reale.
Towards a digital twin platform for laser engraving using an educational robot system
Seminario Gastelo, Javier Martin
2024/2025
Abstract
Precision in laser-based manufacturing depends on several process parameters, so accurate predictive modeling is key to reducing errors and improving results. This thesis presents a method to build the foundation for a Digital Twin (DT) in laser-based manufacturing. The aim is to find the best way to estimate the engraving width produced by a low-power laser, based on controllable parameters. A virtual environment was created in Unity to allow real-time monitoring and virtual testing, helping reduce time and material use. The physical setup includes an educational robotic arm used to run controlled experiments under different conditions. To estimate engraving width, four different modelling strategies were developed. Three analytical models were based on physical laser theory, each combining equations with experimental data and differing in assumptions and required tests. The fourth model uses Extended Kernel Regression (EKR) to improve the predictions from the analytical models, increasing accuracy and reliability. This study explores different modeling approaches to assess their effectiveness in predicting engraving width under various experimental conditions. To build the analytical models and for the validation stage, a structured dataset was built with variations in laser power, speed, and defocusing distance. The models were tested in two stages: linear path experiments with combined parameter changes, and engraving tests with complex shapes like squares, triangles, and spirals. Results show that the analytical model accounting for beam shape produces highly accurate estimations, demonstrating how important this factor is. The inclusion of EKR in Model 4 further enhances prediction consistency, making it more adaptable to various engraving conditions. However, experimental data reveals that to further improve performance, additional training data is necessary, as certain variations in process parameters still present challenges in maintaining accuracy. This methodology contributes to the advancement of laser-based manufacturing within a Digital Twin environment, providing a foundation for integrating predictive modeling into real-time simulation.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_07_Seminario_Tesi_01.pdf
accessibile in internet per tutti a partire dal 30/06/2026
Descrizione: Tesi
Dimensione
12.48 MB
Formato
Adobe PDF
|
12.48 MB | Adobe PDF | Visualizza/Apri |
|
2025_07_Seminario_Executive Summary_02.pdf
accessibile in internet per tutti a partire dal 30/06/2026
Descrizione: Executive summary
Dimensione
1.06 MB
Formato
Adobe PDF
|
1.06 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/10589/239839