Numerical simulation is a cornerstone of modern engineering, yet the computational cost of high-fidelity, multi-query analyses, such as parametric studies and bifurcation tracking, remains prohibitive for complex nonlinear systems. While the Direct Parametrisation method for Invariant Manifolds (DPIM) has emerged as a rigorous alternative to traditional projection-based reduction, its application has historically been centered on mechanical structures with polynomial nonlinearities. This Thesis significantly expands the scope and utility of the DPIM by developing novel formulations for two complex domains: strongly coupled multiphysics systems and parameter-dependent fluid instabilities. First, the framework is extended to address coupled multiphysics systems, with a focus on electrostatically and piezoelectrically actuated Micro-Electro-Mechanical Systems (MEMS). In the electrostatic case, where forces introduce non-polynomial couplings in the reference configuration, a novel mixed formulation is developed. By introducing auxiliary potential gradient fields, the coupled problem is recast into a quadratic-cubic differential-algebraic system compatible with the DPIM. This enables the direct parametrisation of a 1.3-million-degree-of-freedom industrial MEMS gyroscope. The resulting reduced-order model accurately predicts a subtle three-mode internal combination resonance, showing excellent quantitative agreement with both theoretical normal-form derivations and experimental measurements. Furthermore, the framework is extended to piezoelectric systems; unlike the electrostatic case, the Saint Venant-Kirchhoff extension for piezoelectricity naturally yields the required polynomial form in the Lagrangian configuration, allowing for efficient reduction of high-dimensional energy-harvesting and sensing structures while maintaining full coupling. Second, the Thesis advances the application of the DPIM to fluid systems, specifically the incompressible Navier-Stokes equations. Whereas previous studies have applied invariant manifold reduction to simple flows, this work develops a parameter-dependent formulation by incorporating the Reynolds number as an auxiliary state variable. This approach generates a single reduced-order model valid across a wide range of operating conditions. Applied to the Turek-Sch\"afer cylinder wake benchmark, the model successfully predicts the Hopf bifurcation point and post-bifurcation limit cycles. Furthermore, the work introduces an \textit{a priori} error estimator that quantifies model validity without requiring full-order validation. By extending the DPIM beyond its traditional domain of polynomial mechanical structures to multiphysics and fluid systems, this Thesis demonstrates that the DPIM can serve as a robust, computationally efficient, and experimentally validated tool for realistic engineering design and analysis.
La simulazione numerica è un pilastro dell'ingegneria moderna, tuttavia il costo computazionale di analisi ad alta fedeltà e multi-query, come studi parametrici e tracciamento di biforcazioni, rimane proibitivo per sistemi nonlineari complessi. Sebbene il metodo di Parametrizzazione Diretta delle Varietà Invarianti (DPIM) sia emerso come un'alternativa rigorosa alla riduzione tradizionale basata su proiezione, la sua applicazione è storicamente stata incentrata su strutture meccaniche con nonlinearità polinomiali. Questa Tesi espande significativamente l'ambito e l'utilità della DPIM sviluppando nuove formulazioni per due domini complessi: sistemi multifisici accoppiati e instabilità dipendenti da parametri. In primo luogo, il framework viene esteso per affrontare sistemi multifisici accoppiati, con particolare attenzione ai Sistemi Micro-Elettro-Meccanici (MEMS) attuati elettrostaticamente e piezoelettricamente. Nel caso elettrostatico, in cui le forze introducono accoppiamenti non polinomiali nella configurazione di riferimento, viene sviluppata una nuova formulazione mista. Introducendo campi ausiliari del gradiente di potenziale, il problema accoppiato viene riformulato in un sistema differenziale-algebrico compatibile con la DPIM. Ciò consente la parametrizzazione diretta di un giroscopio MEMS industriale con 1,3 milioni di gradi di libertà. Il modello risultante predice accuratamente una sottile risonanza di combinazione interna a tre modi, mostrando un eccellente accordo con le derivazioni teoriche in forma normale e con le misure sperimentali. Inoltre, il framework viene esteso ai sistemi piezoelettrici; a differenza del caso elettrostatico, l'estensione di Saint Venant-Kirchhoff per la piezoelettricità fornisce naturalmente la forma polinomiale richiesta nella configurazione Lagrangiana, consentendo una riduzione efficiente di strutture ad alta dimensione per energy harvesting e sensing, mantenendo al contempo il pieno accoppiamento. In secondo luogo, la Tesi promuove l'applicazione della DPIM ai sistemi fluidodinamici, in particolare alle equazioni di Navier-Stokes incomprimibili. Mentre studi precedenti hanno applicato la riduzione a flussi semplici, questo lavoro sviluppa una formulazione dipendente da parametri incorporando il numero di Reynolds come variabile di stato ausiliaria. Questo approccio genera un unico modello ridotto valido su un ampio intervallo di condizioni operative. Applicato al benchmark di flusso attorno a un cilindro di Turek-Schäfer, il modello predice sia il punto di biforcazione di Hopf che i cicli limite post-biforcazione. Inoltre, viene introdotto uno stimatore di errore \textit{a priori} che quantifica la validità del modello senza richiedere validazione con il modello completo. Estendendo la DPIM oltre il suo dominio tradizionale di strutture meccaniche polinomiali a sistemi multifisici e fluidodinamici, questa Tesi dimostra che la DPIM può fungere da strumento robusto, computazionalmente efficiente e sperimentalmente validato per la progettazione e l'analisi ingegneristica realistica.
Invariant manifold reduction of coupled multiphysics and parametrised systems
COLOMBO, ALESSIO
2026
Abstract
Numerical simulation is a cornerstone of modern engineering, yet the computational cost of high-fidelity, multi-query analyses, such as parametric studies and bifurcation tracking, remains prohibitive for complex nonlinear systems. While the Direct Parametrisation method for Invariant Manifolds (DPIM) has emerged as a rigorous alternative to traditional projection-based reduction, its application has historically been centered on mechanical structures with polynomial nonlinearities. This Thesis significantly expands the scope and utility of the DPIM by developing novel formulations for two complex domains: strongly coupled multiphysics systems and parameter-dependent fluid instabilities. First, the framework is extended to address coupled multiphysics systems, with a focus on electrostatically and piezoelectrically actuated Micro-Electro-Mechanical Systems (MEMS). In the electrostatic case, where forces introduce non-polynomial couplings in the reference configuration, a novel mixed formulation is developed. By introducing auxiliary potential gradient fields, the coupled problem is recast into a quadratic-cubic differential-algebraic system compatible with the DPIM. This enables the direct parametrisation of a 1.3-million-degree-of-freedom industrial MEMS gyroscope. The resulting reduced-order model accurately predicts a subtle three-mode internal combination resonance, showing excellent quantitative agreement with both theoretical normal-form derivations and experimental measurements. Furthermore, the framework is extended to piezoelectric systems; unlike the electrostatic case, the Saint Venant-Kirchhoff extension for piezoelectricity naturally yields the required polynomial form in the Lagrangian configuration, allowing for efficient reduction of high-dimensional energy-harvesting and sensing structures while maintaining full coupling. Second, the Thesis advances the application of the DPIM to fluid systems, specifically the incompressible Navier-Stokes equations. Whereas previous studies have applied invariant manifold reduction to simple flows, this work develops a parameter-dependent formulation by incorporating the Reynolds number as an auxiliary state variable. This approach generates a single reduced-order model valid across a wide range of operating conditions. Applied to the Turek-Sch\"afer cylinder wake benchmark, the model successfully predicts the Hopf bifurcation point and post-bifurcation limit cycles. Furthermore, the work introduces an \textit{a priori} error estimator that quantifies model validity without requiring full-order validation. By extending the DPIM beyond its traditional domain of polynomial mechanical structures to multiphysics and fluid systems, this Thesis demonstrates that the DPIM can serve as a robust, computationally efficient, and experimentally validated tool for realistic engineering design and analysis.| File | Dimensione | Formato | |
|---|---|---|---|
|
PhD_Thesis.pdf
non disponibili
Licenza:
Tutti i diritti riservati
Dimensione
23.48 MB
Formato
Adobe PDF
|
23.48 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/384646
URN:NBN:IT:POLIMI-384646