Model Predictive Control (MPC) is a class of controllers that has attracted significant interest due to its ability to explicitly handle system constraints while optimizing performance. However, MPC is inherently subject to stringent time requirements. In real-time applications, control actions must be computed within tight computational deadlines, making execution time a critical factor. This thesis aims to reduce the computational load required to construct the control problem associated with an existing MPC algorithm. The objective is not to redesign the control strategy itself, but to improve the computational efficiency of the problem construction at each control iteration. Particular attention is devoted to identifying the most time-consuming operations within the controller. The analysis highlights that the linearization of the system dynamics and the associated numerical integration along the prediction horizon represent the dominant sources of computational cost. Alternative formulations capable of reducing their impact are therefore investigated. Several strategies are analyzed and assessed through a structured computational study. Based on this analysis, a polynomial interpolation-based approach is selected as the most suitable solution for achieving the intended objective. The proposed framework is developed and tested within the OVERBORE project, focusing on the landing burn phase of the booster of a reusable launch vehicle. The required modifications are introduced in a manner that preserves the original control structure and are implemented within the existing simulation environment. The modified algorithm is subsequently evaluated through both offline analyses and closed-loop simulations to quantify the achieved computational savings and to verify that tracking accuracy, stability, and overall control performance are not penalized by the proposed solution.
Il Model Predictive Control (MPC) è una classe di controllori che ha suscitato notevole interesse grazie alla sua capacità di gestire esplicitamente i vincoli di sistema ottimizzando al contempo le prestazioni. Tuttavia, l’MPC è intrinsecamente soggetto a stringenti requisiti temporali. Nelle applicazioni real-time, le azioni di controllo devono essere calcolate entro scadenze computazionali molto ristrette, rendendo il tempo di esecuzione un fattore critico. Questa tesi si propone di ridurre il carico computazionale necessario per la costruzione del problema di controllo associato a un algoritmo MPC già esistente. L’obiettivo non è riprogettare la strategia di controllo, bensì migliorare l’efficienza computazionale della fase di costruzione del problema a ogni iterazione di controllo. Particolare attenzione è dedicata all’identificazione delle operazioni più onerose in termini di tempo all’interno del controllore. L’analisi evidenzia che la linearizzazione della dinamica del sistema e le operazioni di integrazione numerica lungo l’orizzonte di predizione rappresentano le principali fonti di costo computazionale. Vengono pertanto indagate formulazioni alternative in grado di ridurne l’impatto. Diverse strategie vengono analizzate e valutate attraverso uno studio computazionale strutturato. Sulla base di tale analisi, viene selezionato un approccio basato su interpolazione polinomiale come soluzione più idonea al raggiungimento dell’obiettivo prefissato. Il framework proposto è sviluppato e testato nell’ambito del progetto OVERBORE, con riferimento alla fase di landing burn del booster di un reusable launch vehicle. Le modifiche necessarie al framework esistente vengono introdotte in modo da preservare la struttura originale del controllore e sono implementate all’interno dell’ambiente di simulazione. L’algoritmo modificato viene infine valutato mediante analisi offline e simulazioni in anello chiuso, al fine di quantificare i risparmi computazionali ottenuti e verificare che accuratezza di inseguimento, stabilità e prestazioni complessive del controllo non risultino penalizzate dalla soluzione proposta.
Techniques for efficient MPC implementation for the powered landing phase of a reusable launch vehicle
SAVINO, ANDREA
2025/2026
Abstract
Model Predictive Control (MPC) is a class of controllers that has attracted significant interest due to its ability to explicitly handle system constraints while optimizing performance. However, MPC is inherently subject to stringent time requirements. In real-time applications, control actions must be computed within tight computational deadlines, making execution time a critical factor. This thesis aims to reduce the computational load required to construct the control problem associated with an existing MPC algorithm. The objective is not to redesign the control strategy itself, but to improve the computational efficiency of the problem construction at each control iteration. Particular attention is devoted to identifying the most time-consuming operations within the controller. The analysis highlights that the linearization of the system dynamics and the associated numerical integration along the prediction horizon represent the dominant sources of computational cost. Alternative formulations capable of reducing their impact are therefore investigated. Several strategies are analyzed and assessed through a structured computational study. Based on this analysis, a polynomial interpolation-based approach is selected as the most suitable solution for achieving the intended objective. The proposed framework is developed and tested within the OVERBORE project, focusing on the landing burn phase of the booster of a reusable launch vehicle. The required modifications are introduced in a manner that preserves the original control structure and are implemented within the existing simulation environment. The modified algorithm is subsequently evaluated through both offline analyses and closed-loop simulations to quantify the achieved computational savings and to verify that tracking accuracy, stability, and overall control performance are not penalized by the proposed solution.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Savino_Tesi.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
5.51 MB
Formato
Adobe PDF
|
5.51 MB | Adobe PDF | Visualizza/Apri |
|
2026_03_Savino_Executive_Summary.pdf
non accessibile
Descrizione: Executive summary
Dimensione
721.6 kB
Formato
Adobe PDF
|
721.6 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/10589/252079