Industrial Systems (ISs), ranging from manufacturing lines to more advanced networks, increasingly require smarter and smarter control mechanisms due to the complexities of modern industrial automation. However, traditional model-based techniques often fall short in capturing the complexities of these systems due to the inherent system nonlinearities and the exponential growth of process data. As a result, Data-Driven (DD) technology has emerged as a pivotal enabler for the digital transformation of modern ISs, promoting integrated and adaptive control solutions compliant to the so-called Industry 4.0. This evolution introduces advanced technologies, including Internet of Things, Machine Learning, and cyber-physical systems --especially Digital Twin--, which together allow systems to adaptively respond to real-time production demands and environmental changes. Despite DD methodologies represent a promising alternative for improving system control and diagnostics, as well preserving process performance by leveraging real-time data, several technical limitations hindering their widespread application in real-world systems still exist. Within such a context, this thesis investigates the potential of DD methods for monitoring and controlling ISs, addressing their most modern challenges. To this aim, two primary research directions are pursued. The first part of the thesis explores DD dynamic modeling for process control, developing indirect DD control methods for optimizing the process performance. Contributions in this direction include (i) the design of Model Predictive Control techniques for the deep drawing process, which significantly enhances process efficiency and product quality, and (ii) a robust control scheme for input-affine nonlinear systems using Subspace Identification of Nonlinear Dynamics and online Semi-Definite Programming. This control approach offers superior stability and performance for such a class of nonlinear models without approximating their dynamics. The second part of the thesis investigates DD adaptive control and fault detection methods for preserving the performance of industrial processes. This includes (i) an Adaptive Model Predictive Control for Hydraulic Servo Actuators (HSAs) based on flow control valves, and (ii) a novel Model Reference Adaptive Control framework for multi-chamber HSAs based on pressure control valves. Additionally, (iii) a new Adaptive Constrained Clustering algorithm is introduced for the real-time fault detection of ISs, effectively distinguishing between nominal and non-nominal working conditions in a dynamic operational context.
I sistemi industriali (SI), che vanno dalle linee di produzione alle reti più avanzate, richiedono sempre più meccanismi di controllo più intelligenti e intelligenti a causa delle complessità della moderna automazione industriale. Tuttavia, le tecniche tradizionali basate su modelli spesso non riescono a catturare la complessità di questi sistemi a causa delle non linearità intrinseche del sistema e della crescita esponenziale dei dati di processo. Di conseguenza, la tecnologia data-driven (DD) si è affermata come un enabler fondamentale per la trasformazione digitale dei moderni ISs, promuovendo soluzioni di controllo integrate e adattative conformi alla cosiddetta Industria 4.0. Questa evoluzione introduce tecnologie avanzate, tra cui l'Internet delle cose, il machine learning e i sistemi cyber-fisici -in particolare il Digital Twin -, che insieme consentono ai sistemi di rispondere in modo adattativo alle richieste di produzione in tempo reale e ai cambiamenti ambientali. Nonostante le metodologie DD rappresentino un'alternativa promettente per migliorare il controllo e la diagnostica dei sistemi, nonché per preservare le prestazioni del processo sfruttando i dati in tempo reale, esistono ancora diverse limitazioni tecniche che ne impediscono l'applicazione diffusa nei sistemi del mondo reale. In tale contesto, questa tesi indaga il potenziale dei metodi DD per monitorare e controllare gli IS, affrontando le loro sfide più moderne. Per raggiungere questo obiettivo, vengono perseguite due direzioni principali di ricerca. La prima parte della tesi esplora la modellazione dinamica DD per il controllo di processo, sviluppando metodi di controllo indiretto DD per ottimizzare le prestazioni del processo. Tra i contributi in tal senso figurano (i) la progettazione di tecniche di controllo predittivo del modello per il processo di imbutitura, che migliora in modo significativo l'efficienza del processo e la qualità del prodotto, e (ii) un solido schema di controllo per lesistemi non lineari affini che utilizzano l'identificazione subspaziale della dinamica non lineare e la programmazione semidefinita on-li ne. Questo approccio di controllo offre stabilità e prestazioni superiori per una tale classe di modelli non lineari senza approssimarne la dinamica. La seconda parte della tesi indaga i metodi di controllo adattivo e di rilevamento dei guasti DD per preservare le prestazioni dei processi industriali. Ciò include (i) un controllo predittivo adattivo del modello per attuatori servo idraulici (HSA) basato su valvole di controllo del flusso e (ii) una nuova struttura di controllo adattivo di riferimento del modello per HSA multicamera basata su valvole di controllo della pressione. Inoltre, (iii) un nuovo algoritmo di clustering limitato adattivo è introdotto per il rilevamento in tempo reale dei guasti degli IS, distinguendo efficacemente tra condizioni di lavoro nominali e non nominali in un contesto operativo dinamico.
Digital twin for industrial systems: data-driven approaches for monitoring and control / Bozza, Augusto. - ELETTRONICO. - (2025).
Digital twin for industrial systems: data-driven approaches for monitoring and control
Bozza, Augusto
2025-01-01
Abstract
Industrial Systems (ISs), ranging from manufacturing lines to more advanced networks, increasingly require smarter and smarter control mechanisms due to the complexities of modern industrial automation. However, traditional model-based techniques often fall short in capturing the complexities of these systems due to the inherent system nonlinearities and the exponential growth of process data. As a result, Data-Driven (DD) technology has emerged as a pivotal enabler for the digital transformation of modern ISs, promoting integrated and adaptive control solutions compliant to the so-called Industry 4.0. This evolution introduces advanced technologies, including Internet of Things, Machine Learning, and cyber-physical systems --especially Digital Twin--, which together allow systems to adaptively respond to real-time production demands and environmental changes. Despite DD methodologies represent a promising alternative for improving system control and diagnostics, as well preserving process performance by leveraging real-time data, several technical limitations hindering their widespread application in real-world systems still exist. Within such a context, this thesis investigates the potential of DD methods for monitoring and controlling ISs, addressing their most modern challenges. To this aim, two primary research directions are pursued. The first part of the thesis explores DD dynamic modeling for process control, developing indirect DD control methods for optimizing the process performance. Contributions in this direction include (i) the design of Model Predictive Control techniques for the deep drawing process, which significantly enhances process efficiency and product quality, and (ii) a robust control scheme for input-affine nonlinear systems using Subspace Identification of Nonlinear Dynamics and online Semi-Definite Programming. This control approach offers superior stability and performance for such a class of nonlinear models without approximating their dynamics. The second part of the thesis investigates DD adaptive control and fault detection methods for preserving the performance of industrial processes. This includes (i) an Adaptive Model Predictive Control for Hydraulic Servo Actuators (HSAs) based on flow control valves, and (ii) a novel Model Reference Adaptive Control framework for multi-chamber HSAs based on pressure control valves. Additionally, (iii) a new Adaptive Constrained Clustering algorithm is introduced for the real-time fault detection of ISs, effectively distinguishing between nominal and non-nominal working conditions in a dynamic operational context.File | Dimensione | Formato | |
---|---|---|---|
37 ciclo-BOZZA Augusto.pdf
accesso aperto
Tipologia:
Tesi di dottorato
Licenza:
Creative commons
Dimensione
20.51 MB
Formato
Adobe PDF
|
20.51 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.