Laser cutting is nowadays a well established technology in the manufactoring field. It is able to guarantee at the same time high quality and productivity which makes it suitable for industrial applications. The main characteristic of the cutting process is the complexity, caused by the coexistence of many physical domains. This ensures high flexibility but, at the same time, an analytical description of the process is not trivial to define. Therefore, setup parameters cannot be found through mathematical optimization, but they have to be tune empirically by examining their behaviour experimentally, making the parameters choice sub-optimal. As consequence, many research works have been carried on in the recent years with the aim of developing a monitoring system able to self-optimize the process, allowing an improvement in the quality and production time. Different technologies and type of sensors have been use to address the problem, but the camera based monitoring system tuned out to be the most suitable. Such system can be used combined with data processing algorithms in order to provide an online estimation of some cutting quality indexes. At current state of the art, real time control systems of more than one cutting defect have been developed for monodirectional cuts. This thesis work aims of building an online estimation and control architecture for one defect (dross attachment), that guarantees robustness for different cutting geometries. A camera based system is used to capture information about the process and analysis about how it is influenced by the cutting geometry (direction and curvature). The collected data can be processed through a regression Machine Learning model that provides an online estimation of the cutting defect. The structure and complexity of such model is carefully chosen in order to make it able to handle the path geometry dependence. A closed-loop architecture can be designed considering the cutting speed as a control variable, after having identified the dynamical relationship between input (speed) and output (defect). Since the employed control variable is the speed, a detailed analysis on the cutting machine motion provided by electrical motors, have to be performed in order to find its limitations with respect to the path geometry. Therefore, operating regions of the control action can be defined, and the final architecture designed as consequence. Finally, the resulting real time estimation and closed-loop control logic is validated and tested in different working conditions, especially highlighting its robustness against cutting direction and curvature variation.
Il taglio laser è al giorno d'oggi, un'affermata tecnologia nel campo manufatturiero. Allo stesso tempo è in grado di garantire qualità e produttività, il che la rende adatta per molte applicazioni industriali. Ciò che caratterizza maggiormente il taglio laser è la sua complessità, dovuta alla coesistenza di grandezze fisiche di diversa natura. Ciò garantisce flessibilità ma, allo stesso tempo, rende non semplice la descrizione matematica del processo. Per questo motivo, i parametri di settaggio non possono essere ricavati da un'ottimizzazione matematica ma devono essere stabiliti empiricamente esaminando il loro comportamento sperimentalmente, ottenendo una soluzione sub-ottimale. Di conseguenza, negli anni recenti sono state svolte numerose ricerche con lo scopo di sviluppare un sistema di monitoraggio in grado di ottimizzare il processo, consentendo un miglioramento nella qualità del taglio e nella produttività. Differenti tipi di tecnologia e sensori sono stati adoperati per affrontare il tema, ma quello che risulta più idoneo allo scopo è il monitoraggio tramite videocamera in posizione coassiale. Tale sistema può essere combinato con algoritmi di processo dati per garantire una stima real time di alcuni indici di qualità. Allo stato attuale dell'arte, è stato sviluppato un controllo real time in grado di stimare e controllare più di un indice di difetto per il taglio monodirezionale. Questo lavoro di tesi ha lo scopo di costruire un architettura di stima e controllo online di un difetto (bava residua), che garantisca robustezza alle diverse direzioni di taglio. Un sistema basato su videocamera viene utilizzato per raccogliere informazioni sul processo ed analizzare l'influenza della geometria di taglio su di esso. I dati raccolti possono essere processati tramite un modello di Machine Learning che genera una stima real time del difetto. La struttura e la complessità del modello è appositamente scelta per far fronte alla dipendenza dalla direzione. Un'architettura in anello chiuso può essere costruita considerando la velocità come variabile di controllo, dopo aver identificato la relazione dinamica tra input (velocità) ed output (difetto). Essendo la variabile di controllo la velocità, viene svolta un'analisi sulla dinamica della testa di taglio, per identificare limitazioni nel movimento dovute alla geometria percorsa. Perciò, vengono definite le regioni di operatività della variabile di controllo e l'architettura di controllo progettata di conseguenza. Infine la stima real time ed il controllo in anello chiuso vengono testate in diverse condizioni di lavoro, evidenziando la robustezza alla alla direzione di taglio ed alla curvatura.
Real time vision-based estimation and control of dross attachment in fusion laser cutting for multi-diretional and curved geometries
VAZZOLA, LUCA
2022/2023
Abstract
Laser cutting is nowadays a well established technology in the manufactoring field. It is able to guarantee at the same time high quality and productivity which makes it suitable for industrial applications. The main characteristic of the cutting process is the complexity, caused by the coexistence of many physical domains. This ensures high flexibility but, at the same time, an analytical description of the process is not trivial to define. Therefore, setup parameters cannot be found through mathematical optimization, but they have to be tune empirically by examining their behaviour experimentally, making the parameters choice sub-optimal. As consequence, many research works have been carried on in the recent years with the aim of developing a monitoring system able to self-optimize the process, allowing an improvement in the quality and production time. Different technologies and type of sensors have been use to address the problem, but the camera based monitoring system tuned out to be the most suitable. Such system can be used combined with data processing algorithms in order to provide an online estimation of some cutting quality indexes. At current state of the art, real time control systems of more than one cutting defect have been developed for monodirectional cuts. This thesis work aims of building an online estimation and control architecture for one defect (dross attachment), that guarantees robustness for different cutting geometries. A camera based system is used to capture information about the process and analysis about how it is influenced by the cutting geometry (direction and curvature). The collected data can be processed through a regression Machine Learning model that provides an online estimation of the cutting defect. The structure and complexity of such model is carefully chosen in order to make it able to handle the path geometry dependence. A closed-loop architecture can be designed considering the cutting speed as a control variable, after having identified the dynamical relationship between input (speed) and output (defect). Since the employed control variable is the speed, a detailed analysis on the cutting machine motion provided by electrical motors, have to be performed in order to find its limitations with respect to the path geometry. Therefore, operating regions of the control action can be defined, and the final architecture designed as consequence. Finally, the resulting real time estimation and closed-loop control logic is validated and tested in different working conditions, especially highlighting its robustness against cutting direction and curvature variation.| File | Dimensione | Formato | |
|---|---|---|---|
|
Luca_Vazzola_Executive_Summary.pdf
non accessibile
Dimensione
2.2 MB
Formato
Adobe PDF
|
2.2 MB | Adobe PDF | Visualizza/Apri |
|
Luca_Vazzola_Master_Thesis.pdf
non accessibile
Dimensione
20.71 MB
Formato
Adobe PDF
|
20.71 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/10589/219816