A Multilevel Gradient Method for Optimization Problems

I problemi di ottimizzazione di dimensione finita di larga scala spesso derivano dalla discretizzazione di problemi di dimensione infinita. È perciò possibile descrivere il problema di ottimizzazione su più livelli discreti. Lavorando su un livello più basso di quello del problema considerato, si possono calcolare soluzioni approssimate che saranno poi punti di partenza per il problema di ottimizzazione al livello più fine. I metodi multilivello, già ampiamente presenti in letteratura a partire dagli anni Novanta, sfruttano tale caratteristica dei problemi di ottimizzazione per migliorare le prestazioni dei metodi di ottimizzazione standard. L’obiettivo di questa tesi è quello di implementare una variante multilivello del metodo del gradiente (MGM) e di testarlo su due diversi campi: la risoluzione delle Equazioni alle Derivate Parziali la ricostruzione di immagini. In questo elaborato viene illustrata la teoria dello schema multilivello e presentato l’algoritmo di MGM utilizzato nei nostri esperimenti. Sono poi discusse le modalità di utilizzo di MGM per i due problemi sopra presentati. Per il problema PDE, i risultati ottenuti mostrano un ottimo comportamento di MGM rispetto alla implementazione classica ad un livello. I risultati ottenuti per il problema di ricostruzione di immagini, al contrario delle PDEs, evidenziano come MGM sia efficace solo in determinate condizioni.

Exact Combinatorial Optimization with Graph Convolutional Neural Networks

Combinatorial optimization problems are typically tackled by the branch-and-bound paradigm. We propose to learn a variable selection policy for branch-and-bound in mixed-integer linear programming, by imitation learning on a diversified variant of the strong branching expert rule. We encode states as bipartite graphs and parameterize the policy as a graph convolutional neural network. Experiments on a series of synthetic problems demonstrate that our approach produces policies that can improve upon expert-designed branching rules on large problems, and generalize to instances significantly larger than seen during training.

Shear behavior of reinforced concrete slabs under concentrated load: an investigation through Sequentially Linear Analysis

ABSTRACT (italiano) Con crescente attenzione riguardo al problema della sicurezza di ponti e viadotti esistenti nei Paesi Bassi, lo scopo della presente tesi è quello di studiare, mediante la modellazione con Elementi Finiti ed il continuo confronto con risultati sperimentali, la risposta in esercizio di elementi che compongono infrastrutture del genere, ovvero lastre in calcestruzzo armato sollecitate da carichi concentrati. Tali elementi sono caratterizzati da un comportamento ed una crisi per taglio, la cui modellazione è, da un punto di vista computazionale, una sfida piuttosto ardua, a causa del loro comportamento fragile combinato a vari effetti tridimensionali. La tesi è incentrata sull'utilizzo della Sequentially Linear Analysis (SLA), un metodo di soluzione agli Elementi Finiti alternativo rispetto ai classici approcci incrementali e iterativi. Il vantaggio della SLA è quello di evitare i ben noti problemi di convergenza tipici delle analisi non lineari, specificando direttamente l'incremento di danno sull'elemento finito, attraverso la riduzione di rigidezze e resistenze nel particolare elemento finito, invece dell'incremento di carico o di spostamento. Il confronto tra i risultati di due prove di laboratorio su lastre in calcestruzzo armato e quelli della SLA ha dimostrato in entrambi i casi la robustezza del metodo, in termini di accuratezza dei diagrammi carico-spostamento, di distribuzione di tensioni e deformazioni e di rappresentazione del quadro fessurativo e dei meccanismi di crisi per taglio. Diverse variazioni dei più importanti parametri del modello sono state eseguite, evidenziando la forte incidenza sulle soluzioni dell'energia di frattura e del modello scelto per la riduzione del modulo elastico trasversale. Infine è stato effettuato un paragone tra la SLA ed il metodo non lineare di Newton-Raphson, il quale mostra la maggiore affidabilità della SLA nella valutazione di carichi e spostamenti ultimi insieme ad una significativa riduzione dei tempi computazionali. ABSTRACT (english) With increasing attention to the assessment of safety in existing dutch bridges and viaducts, the aim of the present thesis is to study, through the Finite Element modeling method and the continuous comparison with experimental results, the real response of elements that compose these infrastructures, i.e. reinforced concrete slabs subjected to concentrated loads. These elements are characterized by shear behavior and crisis, whose modeling is, from a computational point of view, a hard challenge, due to their brittle behavior combined with various 3D effects. The thesis is focused on the use of Sequentially Linear Analysis (SLA), an alternative solution technique to classical non linear Finite Element analyses that are based on incremental and iterative approaches. The advantage of SLA is to avoid the well-known convergence problems of non linear analyses by directly specifying a damage increment, in terms of a reduction of stiffness and strength in the particular finite element, instead of a load or displacement increment. The comparison between the results of two laboratory tests on reinforced concrete slabs and those obtained by SLA has shown in both the cases the robustness of the method, in terms of accuracy of load-displacements diagrams, of the distribution of stress and strain and of the representation of the cracking pattern and of the shear failure mechanisms. Different variations of the most important parameters have been performed, pointing out the strong incidence on the solutions of the fracture energy and of the chosen shear retention model. At last a confrontation between SLA and the non linear Newton-Raphson method has been executed, showing the better reliability of the SLA in the evaluation of the ultimate loads and displacements, together with a significant reduction of computational times.

A framework for risk analysis in automotive cybersecurity

We address the problem of automotive cybersecurity from the point of view of Threat Analysis and Risk Assessment (TARA). The central question that motivates the thesis is the one about the acceptability of risk, which is vital in taking a decision about the implementation of cybersecurity solutions. For this purpose, we develop a quantitative framework in which we take in input the results of risk assessment and define measures of various facets of a possible risk response; we then exploit the natural presence of trade-offs (cost versus effectiveness) to formulate the problem as a multi-objective optimization. Finally, we develop a stochastic model of the future evolution of the risk factors, by means of Markov chains; we adapt the formulations of the optimization problems to this non-deterministic context. The thesis is the result of a collaboration with the Vehicle Electrification division of Marelli, in particular with the Cybersecurity team based in Bologna; this allowed us to consider a particular instance of the problem, deriving from a real TARA, in order to test both the deterministic and the stochastic framework in a real world application. The collaboration also explains why in the work we often assume the point of view of a tier-1 supplier; however, the analyses performed can be adapted to any other level of the supply chain.

Development of the Simulation Model for the CoSES Laboratory Test Microgrid in Modelica

Evolution of the traditional consumer in a power system to a prosumer has posed many problems in the traditional uni-directional grid. This evolution in the grid model has made it important to study the behaviour of microgrids. This thesis deals with the laboratory microgrid setup at the Munich School of Engineering, built to assist researchers in studying microgrids. The model is built in Dymola which is a tool for the OpenModelica language. Models for the different components were derived, suiting the purpose of this study. The equivalent parameters were derived from data sheets and other simulation programs such as PSCAD. The parameters were entered into the model grid and tested at steady state, firstly. This yielded satisfactory results that were similar to the reference results from MATPOWER power flow. Furthermore, fault conditions at several buses were simulated to observe the behaviour of the grid under these conditions. Recommendations for further developing this model to include more detailed models for components, such as power electronic converters, were made at the end of the thesis.

Machine Learning in Automatic Tabulation for Constraint Model Reformulation

Combinatorial decision and optimization problems belong to numerous applications, such as logistics and scheduling, and can be solved with various approaches. Boolean Satisfiability and Constraint Programming solvers are some of the most used ones and their performance is significantly influenced by the model chosen to represent a given problem. This has led to the study of model reformulation methods, one of which is tabulation, that consists in rewriting the expression of a constraint in terms of a table constraint. To apply it, one should identify which constraints can help and which can hinder the solving process. So far this has been performed by hand, for example in MiniZinc, or automatically with manually designed heuristics, in Savile Row. Though, it has been shown that the performances of these heuristics differ across problems and solvers, in some cases helping and in others hindering the solving procedure. However, recent works in the field of combinatorial optimization have shown that Machine Learning (ML) can be increasingly useful in the model reformulation steps. This thesis aims to design a ML approach to identify the instances for which Savile Row’s heuristics should be activated. Additionally, it is possible that the heuristics miss some good tabulation opportunities, so we perform an exploratory analysis for the creation of a ML classifier able to predict whether or not a constraint should be tabulated. The results reached towards the first goal show that a random forest classifier leads to an increase in the performances of 4 different solvers. The experimental results in the second task show that a ML approach could improve the performance of a solver for some problem classes.