1,720,987 research outputs found
Modellazione di ordin ridotto e decomposizione in sottodomini per l'analisi strutturale
No full textLAUREA MAGISTRALELa seguente tesi si pone l'obbiettivo di ottenere un algoritmo generale per lo sviluppo di modelli di ordine ridotto (Reduced Order Model, ROM) in abbinamento alla decomposizione in sottodomini (Domain Decomposition, DD). Allo stato attuale sono state proposte varie metodologie per l'uso combinato di queste due tecniche in problemi lineari e non; ciò è oggetto di un intensa ricerca.
Nel presente elaborato si proporranno soluzioni per l'uso della MOR e della DD, facendo riferimento ad esempi semplici, tuttavia assimilabili a casi realmente affrontati nella pratica. La semplicità degli esempi consentirà una migliore comprensione delle tecniche usate, mentre la somiglianza con i casi reali permetterà di esaltarne i vantaggi. Con l'uso delle tecniche sopra citate sarà possibile ridurre l'onere di calcolo fino a 300 volte rispetto a quello di un'analisi agli elementi finiti classica. In aggiunta alle problematiche legate all'uso della DD+MOR, nel presente lavoro si tratterà in modo esteso il problema della definizione delle basi di proiezione del modello ridotto mediante la decomposizione a valori singolari (SVD). L'elaborato mostrerà i risultati e le implicazioni teoriche derivanti dall'uso di sottospazi definiti da diversi tipi di basi; inoltre verrà mostrato il legame che intercorre tra la cosiddetta energia orientata e l'energia meccanica. La prima è il parametro che governa la riduzione del modello, mentre il legame con la seconda permetterà di comprendere meglio le conseguenze e le modalità della riduzione. La definizione del sottospazio richiede una fase preliminare di studio detta training. Di questa verrà studiato con particolare attenzione il problema della convergenza delle basi e il loro legame con la soluzione del ROM. Sulla base di esempi e osservazioni teoriche, si proporranno algoritmi finalizzati a controllare il training. Relativamente alla decomposizione in sottodomini si farà riferimento all'algoritmo di Gravouil-Combescure, che verrà modificato per adattarsi alla ROM. Si esporranno infine sia le modalità di impiego possibili sia il significato fisico degli algoritmi proposti.This thesis has the aim to develop a general algorithm for the application of the Model Order Reduction (MOR) combined with the Domain Decomposition (DD) method. Currently, the research related to the model order reduction and the domain decomposition (combined or not) is object of huge efforts.
In this thesis new algorithms will be proposed for the DD+MOR and they will be checked with simple examples; despite their simplicity, they have the structural behavior typical of many real problems. The simplicity of the models allows to look deeply into the meaning of the results; moreover their behaviors allow to highlight the advantages for real simulation cases. With the use of the DD+MOR, a speedup of the simulation up to 300, respect to the classical technique, will be shown. In addition of these topics, in this thesis the problems related to the definition of the subspace of the solution used for the MOR will be treated. This work will show the results and the theoretical implications coming from different types of subspace's basis. Furthermore a physical relationship between the so called oriented energy and the mechanical energy will be proposed. The former is the parameter governing the model order reduction, whereas the relationship with the latter allows to show how the reduction is imposed and its consequences. The definition of the subspace requires a preliminary analysis, as known as training. About this, the problem related to the convergence of the basis and their relationships with the ROM solution will be studied in detail. On the base of examples and theoretical observations, algorithms for the control of the training will be proposed. The reference for the DD will be the Gravouil-Combescure method; this will be modified in order to work with the ROM. For all the proposed algorithms, a mathematical and a physical interpretation will be shown
Leveraging Deep Learning for Robust Structural Damage Detection and Classification: A Transfer Learning Approach via CNN
No full textTransfer learning techniques for structural health monitoring in bridge-type structures are investigated, focusing on model generalizability and domain adaptation challenges. Finite element models of bridge-type structures with varying geometry were simulated using the OpenSeesPy platform. Different levels of damage states were introduced at the midspans of these models, and Gaussian-based load time histories were applied at mid-span for dynamic time-history analysis to calculate acceleration data. Then, this acceleration time-history series was transformed into grayscale images, serving as inputs for a Convolutional Neural Network developed to detect and classify structural damage states. Initially, it was trained and tested on datasets derived from a Single-Source Domain structure, achieving perfect accuracy (1.0) in a ten-label multi-class classification task. However, this accuracy significantly decreased when the model was sequentially tested on structures with different geometry without retraining. To address this challenge, it is proposed that transfer learning be employed via feature extraction and joint training. The model showed a reduction in accuracy percentage when adapting from a Single-Source Domain to Multiple-Target Domains, revealing potential issues with non-homogeneous data distribution and catastrophic forgetting. Conversely, joint training, which involves training on all datasets except the specific Target Domain, generated a generalized network that effectively mitigated these issues and maintained high accuracy in predicting unseen class labels. This study highlights the integration of simulation data into the Deep Learning-based SHM framework, demonstrating that a generalized model created via Joint Learning utilizing FEM can potentially reduce the consequences of modeling errors and operational uncertainties unavoidable in real-world applications
- …
