Rozprawy doktorskie na temat „Multi-fidelity models”

L’optimisation des différents composants d’une turbomachine reste encore un sujet épineux, malgré les récentes avancées théoriques, expérimentales ou informatiques. Cette thèse propose et investigue des techniques d’optimisation assistées par méta-modèles vectoriels multi-fidélité basés sur la Décomposition aux Valeurs Propres (POD). Le couplage de la POD à des techniques de modélisation multifidélité permet de suivre l’évolution des structures dominantes de l’écoulement en réponse à des déformations géométriques. Deux méthodes d’optimisation multi-fidélité basées sur la POD sont ici proposées. La première consiste en une stratégie d’enrichissement adaptée aux modèles multi-fidelité par Gappy-POD (GPOD). Celle-ci vise surtout des problèmes associés à des simulations basse-fidélité à coût de restitution négligeable, ce qui la rend difficilement utilisable pour la conception aérodynamique de turbomachines. Elle est néanmoins validée sur une étude du domaine de vol d’une aile 2D issue de la littérature. La seconde méthodologie est basée sur une extension multi-fidèle des modèles par POD Non-Intrusive (NIPOD). Cette extension naît de la ré-interprétation du concept de POD Contrainte (CPOD) et permet l’enrichissement de l’espace réduit par ajout important d’information basse-fidélité approximative. En seconde partie de cette thèse, un cas de validation est introduit pour valider les méthodologies d’optimisation vectorielle multi-fidélité. Cet exemple présente des caractéristiques représentatives des problèmes d’optimisation de turbomachines. La capacité de généralisation des méta-modèles par NIPOD multifidélité proposés est comparée, aussi bien sur cas analytique qu’industriel, à des techniques de méta-modélisation issues de la littérature. Enfin, nous utilisons la méthode développée au cours de cette thèse pour l’optimisation d’un étage et demi d’un compresseur basse-pression et comparons les résultats obtenus à des approches à l’état de l’art
Optimizing turbomachinery components stands as a real challenge despite recent advances in theoretical, experimental and High-Performance Computing (HPC) domains. This thesis introduces and validates optimization techniques assisted by full-field Multi-Fidelity Surrogate Models (MFSMs) based on Proper Orthogonal Decomposition (POD). The combination of POD and Multi-Fidelity Modeling (MFM) techniques allows to capture the evolution of dominant flow features with geometry modifications. Two POD based multi-fidelity optimization methods are proposed. Thefirst one consists in an enrichment strategy dedicated to Gappy-POD (GPOD)models. It is more suitable for instantaneous low-fidelity computations whichmakes it hardly tractable for aerodynamic design of turbomachines. This methodis demonstrated on the flight domain study of a 2D airfoil from the literature. The second methodology is based on a multi-fidelity extension to Non-IntrusivePOD (NIPOD) models. This extension starts with a re-interpretation of theConstrained POD (CPOD) concept and allows to enrich the reduced spacedefinition with abondant, albeit inaccurate, low-fidelity information. In the second part of the thesis, a benchmark test case is introduced to test fullfield multi-fidelity optimization methodologies on an example presenting featuresrepresentative of turbomachinery problems. The predictability of the proposedMulti-Fidelity NIPOD (MFNIPOD) surrogate models is compared to classical surrogates from the literature on both analytical and industrial-scale applications. Finally, we employ the proposed tool to the shape optimization of a 1.5-stage boosterand we compare the obtained results with standard state of the art approaches

The purpose of this work is to help bridge the gap between aircraft conceptual design and analysis. Much work is needed, but distilling essential characteristics from a design and collecting them in an easily accessible format that is amenable to use by inexpensive analysis tools is a significant contribution to this goal. Toward that end, four types of reduced-fidelity or degenerate geometric representations have been defined and implemented in VSP, a parametric geometry modeler. The four types are degenerate surface, degenerate plate, degenerate stick, and degenerate point, corresponding to three-, two-, one-, and zero- dimensional representations of underlying geometry, respectively. The information contained in these representations was targeted specifically at lifting line, vortex lattice, equivalent beam, and equivalent plate theories, with the idea that suitability for interface with these methods would imply suitability for use with many other analysis techniques. The ability to output this information in two plain text formats— comma separated value and Matlab script—has also been implemented in VSP, making it readily available for use. A modified Cessna 182 wing created in VSP was used to test the suitability of degenerate geometry to interface with the four target analysis techniques. All four test cases were easily completed using the information contained in the degenerate geometric types, and similar techniques utilizing different degenerate geometries produced similar results. The following work outlines the theoretical underpinnings of degenerate geometry and the fidelity-reduction process. It also describes in detail how the routines that create degenerate geometry were implemented in VSP and concludes with the analysis test cases, stating their results and comparing results among different techniques.

Building complex physics-based systems in a timely cost-effective manner, that perform well, meet diverse user needs, and have no bad emergent behaviors is a challenge. To meet these requirements the solution is to model the physics-based system before building it. Modeling and Simulation capabilities for these type systems have advanced continuously during the past 20 years thanks to progress in the application of high fidelity computational codes that are able to model the real physical performance of system components. The problem is that it is often too time consuming and costly to model complex systems, end-to-end, using these high fidelity computational models alone. Missing are good approaches to segment the modeling of complex systems performance and behaviors, keep the model chain coherent and only model what is necessary. Current research efforts have shown that using multi-fidelity and/or surrogate models might offer alternative methods of performing the modeling and simulations needed to design and develop physics-based systems more efficiently. This study demonstrates that it is possible reduce the number of high fidelity runs allowing the use of classical systems engineering analysis and tools that would not be possible if only high fidelity codes were employed. This study advances the systems engineering of physics-based systems by reducing the number of time consuming high fidelity models and simulations that must be used to design and develop the systems. The study produced a novel approach to the design and development of complex physics-based systems by using a mix of variable fidelity physics-based models and surrogate models. It shows that this combination of increasing fidelity models enables the computationally and cost efficient modeling and simulation of these complex systems and their components. The study presents an example of the methodology for the analysis and design of two physics-based systems: a Ground Penetrating Radar (GPR) and a Nuclear Electromagnetic Pulse Bounded Wave System.

Einleitung Studierende der Veterinärmedizin müssen neben umfangreichem theoretischem Wissen zahlreiche praktische Fertigkeiten erlernen. Da jeder Einzelne in seinem eigenen Tempo lernt, besteht ein großer Bedarf an Trainingsmöglichkeiten. Kadaver und lebende Tiere sind selten in ausreichender Menge verfügbar und lebende Tiere sind zudem aus Gründen des Tierwohls nur eingeschränkt zu verwenden. Simulationsmodelle (Modelle von Organismen / Körperteilen) können hier Abhilfe schaffen. Kommerziell erhältliche Modelle sind sehr kostenintensiv und für die Tiermedizin noch nicht flächendeckend erhältlich. Zunehmend werden selbst entwickelte low-fidelity Modelle in der Tiermedizin verwendet. Aufgrund des Mangels an publizierten Daten zu ihrem Einsatz besteht intensiver Forschungsbedarf. Ziele der Untersuchungen In dieser Arbeit sollte untersucht werden, ob einfache, selbst entwickelte Simulationsmodelle (low-fidelity Modelle) erfolgreich in der Lehre eingesetzt werden können. Dazu wurden zwei selbst entwickelte und gebaute Simulationsmodelle evaluiert (Studie 1) und ihr Einsatz in Kombination mit anderen Lehrmedien untersucht (Studie 2). Materialien und Methoden In Studie 1 wurden zwei low-fidelity Modelle zur kaninen Intubation und Katheterisierung entwickelt und evaluiert. Es wurde ein Studiendesign genutzt, das die erworbenen Fertigkeiten zweier Übungsgruppen und einer Kontrollgruppe in einer praktischen Prüfung (OSCE = objective structured clinical examination) am toten Hund vergleicht. Achtundfünfzig Studierende (4. FS) erhielten eine theoretische Einführung zur Intubation und wurden randomisiert auf drei Gruppen aufgeteilt. Gruppe 1 (high-fidelity) übte am kommerziell erhältlichen Intubation Training Manikin, Gruppe 2 (low-fidelity) am entwickelten low-fidelity Modell und die Textgruppe las einen Text, der die Intubation beim Hund beschreibt. Siebenundvierzig Studierende (10. FS) durchliefen dasselbe Studiendesign zum Thema Katheterisierung der Hündin. Sie nutzten das kommerziell erhältliche Female Urinary Catheter Training Manikin, das selbst entwickelte low-fidelity Modell und Lehrtexte. In Studie 2 wurde die Vermittlung zweier spezifischer Fertigkeiten mit Hilfe von Potcasts und Simulationstraining evaluiert. Zwei anleitende Potcasts zu Intubation und Katheterisierung und die oben beschriebenen Modelle wurden innerhalb eines crossover-Studiendesigns genutzt. In dieser Studie sind Potcasts audio-visuell aufbereitete Animationen mit Schritt für Schritt – Anleitungen und Informationen. Die erworbenen praktischen Fertigkeiten zweier Übungsgruppen, die sich in der Art der theoretischen Vorbereitung unterschieden, wurden in einer praktischen Prüfung (OSCE) am toten Hund verglichen. Ein Fragebogen erfasste das Feedback der Teilnehmer. Sechzig Studierende (2. FS) wurden randomisiert auf eine Potcast- und eine Textgruppe aufgeteilt. Die Potcastgruppe sah sich das anleitende Potcast an, die Textgruppe bereitete sich anhand eines Lehrtextes vor. Im Anschluss hatten beide Gruppen separate Übungseinheiten an den low-fidelity Modellen ohne Betreuung durch Lehrende. Ergebnisse In Studie 1 schnitten alle Übungsgruppen signifikant besser ab als die Textgruppen. Gruppe 1 (high-fidelity) und Gruppe 2 (low-fidelity) unterschieden sich weder bei der Intubation noch bei der Katheterisierung signifikant in ihren Leistungen. In Studie 2 schnitt die Potcastgruppe beim Thema Intubation signifikant besser ab als die Textgruppe, beim Thema Katheterisierung ergaben sich keine signifikanten Unterschiede. Insgesamt hatte das Simulationstraining den Studierenden Spaß gemacht, das Lernen ohne Betreuer wurde jedoch als Herausforderung empfunden. Schlussfolgerungen Es ist davon auszugehen, dass low-fidelity Modelle genauso geeignet für das Training klinischer Fertigkeiten sein können wie high-fidelity Modelle. Das Training klinischer Fertigkeiten mit Hilfe von Potcasts und low-fidelity Modellen sollte durch Betreuer ergänzt werden, anstatt als alleiniges Lehrmedium für Studierende des ersten Studienjahres Verwendung zu finden. Eigenständiges Lernen klinischer Fertigkeiten, angeleitet durch Potcasts bietet eine Möglichkeit für vertiefendes und wiederholendes Training höherer Semester. Der Einsatz von Simulationsmodellen in der veterinärmedizinischen Ausbildung wächst seit wenigen Jahren stetig. Diese Arbeit leistet einen zeitgerechten Beitrag bei der Evaluierung von Simulationstraining
Introduction Students of veterinary medicine are expected to acquire various practical skills in addition to a wide range of theoretical knowledge. There is a strong demand for training opportunities, as every individual learns and acquires practical skills at individual pace. For reasons of animal welfare concerns and availability, live animals and cadavers cannot always be used for clinical skills training. Simulation models, which are models of organisms or body parts can be a considerable alternative for clinical skills training. Models that are commercially produced often have a high price and are not available for all skills. Self-made models are increasingly used in veterinary education. Because there is few published data regarding their use, more scientific research is required. Aims of the Investigation The objective of this study was to determine, if self-made low-fidelity models can be successfully used in veterinary medical education. For this purpose, two self-made low-fidelity models were evaluated (study 1) and their use in combination with other teaching tools was analyzed (study 2). Materials and Methods In study 1, two self-made low-fidelity models for simulation of canine intubation and canine female urinary catheterization were developed and evaluated. We used a study design that compares acquired skills of two intervention groups and one control group in a practical examination (OSCE = objective structured clinical examination). Fifty-eight second-year veterinary medicine students received a theoretical introduction to intubation and were randomly divided into three groups. Group 1 (high-fidelity) was then trained on a commercially available Intubation Training Manikin, group 2 (low-fidelity) was trained on our low-fidelity model, and the text group read a text describing intubation of the dog. Forty-seven fifth-year veterinary medicine students followed the same procedure for training urinary catheterization using the commercially available Female Urinary Catheter Training Manikin, our self-made model, and text. Outcomes were assessed in a practical examination on a cadaver using an OSCE checklist. In study 2 we evaluated the teaching of two specific clinical skills using potcasts and low-fidelity simulation training. Two instructional potcasts describing intubation and catheterization and both low-fidelity models described above were used. In our study, potcasts are audio-visual animations that provide the learner with step by step information and instruction on a clinical skill. We used a crossover study design and compared the acquired practical skills of two intervention groups after a different theoretical preparation. A survey captured the participants’ feedback. Sixty first year veterinary medicine students were randomly allocated to two groups, a potcast group and a text group. The potcast group watched a potcast while the text group read an instructional text for preparation. Then both groups had separate self-directed training sessions on low-fidelity models. Outcomes were assessed in practical examinations on a cadaver using an objective structured clinical examination (OSCE) checklist. Results In study 1 all intervention groups performed significantly better than the text groups. Group I (high-fidelity) and group II (low-fidelity) for both intubation and catheterization showed no significant differences. In study 2 the potcast group performed significantly better than the text group in study intubation but no significant differences were observed in study catheterization. Overall, participants enjoyed clinical skills training but experienced self-directed learning as challenging. Conclusion Low-fidelity models can be as effective as high-fidelity models for clinical skills training. Clinical skills training using potcasts and self-directed low-fidelity simulation training should be complemented by supervisor or peer instruction rather than used as exclusive tool for teaching first year veterinary students. We assume though, that self-directed learning instructed by our potcasts can be a valuable chance for deepening and repetitive training of higher semesters. The use of simulation models in veterinary education has been consistently increasing in the past few years. This study is an important, timely contribution to the evaluation of simulation based education

L’optimisation de la performance des voiliers est un problème difficile en raison de la complexité du systèmemécanique (couplage aéro-élastique et hydrodynamique) et du nombre important de paramètres à optimiser (voiles, gréement,etc.). Malgré le fait que l’optimisation des voiliers est empirique dans la plupart des cas aujourd’hui, les approchesnumériques peuvent maintenant devenir envisageables grâce aux dernières améliorations des modèles physiques et despuissances de calcul. Les calculs aéro-hydrodynamiques restent cependant très coûteux car chaque évaluation demandegénéralement la résolution d’un problème non linéaire d’interaction fluide-structure. Ainsi, l’objectif central de cette thèseest de proposer et développer des méthodes originales dans le but de minimiser le coût numérique de l’optimisation dela performance des voiliers. L’optimisation globale par méta-modèles Gaussiens est utilisée pour résoudre différents problèmesd’optimisation. La méthode d’optimisation par méta-modèles est étendue aux cas d’optimisations sous contraintes,incluant de possibles points non évaluables, par une approche de type classification. L’utilisation de méta-modèles à fidélitésmultiples est également adaptée à la méthode d’optimisation globale. Les applications concernent des problèmesd’optimisation originaux où la performance est modélisée expérimentalement et/ou numériquement. Ces différentes applicationspermettent de valider les développements des méthodes d’optimisation sur des cas concrets et complexes, incluantdes phénomènes d’interaction fluide-structure
Sailing yacht performance optimization is a difficult problem due to the high complexity of the mechanicalsystem (aero-elastic and hydrodynamic coupling) and the large number of parameters to optimize (sails, rigs, etc.).Despite the fact that sailboats optimization is empirical in most cases today, the numerical optimization approach is nowconsidered as possible because of the latest advances in physical models and computing power. However, these numericaloptimizations remain very expensive as each simulation usually requires solving a non-linear fluid-structure interactionproblem. Thus, the central objective of this thesis is to propose and to develop original methods aiming at minimizing thenumerical cost of sailing yacht performance optimization. The Efficient Global Optimization (EGO) is therefore appliedto solve various optimization problems. The original EGO method is extended to cases of optimization under constraints,including possible non computable points, using a classification-based approach. The use of multi-fidelity surrogates isalso adapted to the EGO method. The applications treated in this thesis concern the original optimization problems inwhich the performance is modeled experimentally and/or numerically. These various applications allow for the validationof the developments in optimization methods on real and complex problems, including fluid-structure interactionphenomena

Le sujet de cette thèse porte sur laréduction d'ordre de modèle (MOR) deproblèmes d'écoulement non-Newtonianparamétrés qui ont des applicationsindustrielles importantes. Les méthodestraditionnelles de réduction de l'ordre desmodèles limitent les performances decalcul de ces problèmes hautement nonlinéaires, nous suggérons donc une techniqued'hyper-réduction avancée basée sur uneapproximation sparse de l'évaluation destermes non linéaire à complexité reduite.Nous proposons également une stratégie destabilisation hors ligne pour stabiliser le modèleconstitutif dans le modèle d'ordre réduit quiest moins cher à calculer tout en maintenant laprécision du modèle d'ordre complet. Lacombinaison des deux réduit drastiquement lecoût du processeur, augmentantinévitablement les performances du MOR. Cetravail est validé sur deux problèmes debenchmark. En outre, une stratégied'échantillonnage adaptatif est égalementprésentée dans ce manuscrit, qui est réaliséeen tirant parti de l'approximation des modèlesmulti-fidélité. Vers la fin de la thèse, nousabordons un autre problème qui estgénéralement observé dans les cas où desmaillages d'éléments finis adaptatifs sontdéployés. Dans de tels cas, les méthodes MORne parviennent pas à produire unereprésentation de faible dimension car lessnapshots ne sont pas des vecteurs de mêmelongueur. Par conséquent, nous suggérons uneméthodologie qui peut générer des fonctionsde base réduites pour des snapshotsadaptative
The subject of this thesis concernsmodel-order reduction (MOR) of parameterizednon-Newtonian flow problems that havesignificant industrial applications. TraditionalMOR methods constrain the computationalperformance of such highly nonlinear problems,so we suggest a state-of-the-art hyper-reductiontechnique based on a sparse approximation totackle the evaluation of nonlinear terms at muchreduced complexity. We also provide offlinestabilization strategy for stabilizing theconstitutive model in the reduced order modelframework that is less expensive to computewhile maintaining the full order model's (FOM)accuracy. Combining the two significantlylowers the CPU cost as compared to the FOMevaluation which inevitably boosts MORperformance. This work is validated on twobenchmark flow problems. Additionally, anadaptive sampling strategy is also presented inthis manuscript which is achieved byleveraging multi-fidelity model approximation.Towards the end of the thesis, we addressanother issue that is typically observed forcases when adaptive finite element meshesare deployed. In such cases, MOR methods failto produce a low-dimensional representationsince the snapshots are not vectors of samelength. We therefore, suggest an alternatemethod that can generate reduced basisfunctions for database of space-adaptedsnapshots

Increasing worldwide demand for petroleum motivates greater efficiency, safety, and environmental responsibility in upstream oil and gas processes. The objective of this research is to improve these areas with advanced control methods. This work develops the integration of optimal control methods including model predictive control, moving horizon estimation, high fidelity simulators, and switched control techniques applied to subsea riser slugging and managed pressure drilling. A subsea riser slugging model predictive controller eliminates persistent offset and decreases settling time by 5% compared to a traditional PID controller. A sensitivity analysis shows the effect of riser base pressure sensor location on controller response. A review of current crude oil pipeline wax deposition prevention, monitoring, and remediation techniques is given. Also, industrially relevant control model parameter estimation techniques are reviewed and heuristics are developed for gain and time constant estimates for single input/single output systems. The analysis indicates that overestimated controller gain and underestimated controller time constant leads to better controller performance under model parameter uncertainty. An online method for giving statistical significance to control model parameter estimates is presented. Additionally, basic and advanced switched model predictive control schemes are presented. Both algorithms use control models of varying fidelity: a high fidelity process model, a reduced order nonlinear model, and a linear empirical model. The basic switched structure introduces a method for bumpless switching between control models in a predetermined switching order. The advanced switched controller builds on the basic controller; however, instead of a predetermined switching sequence, the advanced algorithm uses the linear empirical controller when possible. When controller performance becomes unacceptable, the algorithm implements the low order model to control the process while the high fidelity model generates simulated data which is used to estimate the empirical model parameters. Once this online model identification process is complete, the controller reinstates the empirical model to control the process. This control framework allows the more accurate, yet computationally expensive, predictive capabilities of the high fidelity simulator to be incorporated into the locally accurate linear empirical model while still maintaining convergence guarantees.

The Ph.D. program has been focused on the development of a multidisciplinary integrated environment for the design of wing for which large changes in shape are expected to be allowed during the flight in order to be better adapted for the different flight segments. The first phase of study has been dedicated to the investigation of the proper Multidisciplinary Design Optimization (MDO) architecture for the integrated management of the design process and a multilevel solution has been proposed and implemented. Such framework involves several disciplinary analysis and optimization loops: in particular aerodynamic analysis, structural analysis, material optimization and mission and performance evaluation are the main components considered for the preliminary design development for such a “morphing” wing. This stage addressed basically the multidisciplinarity and interdisciplinarity issues. The second phase has been dedicated to the investigation of possible techniques for the reduction of the computational burden that characterizes typically this kind of integrated design processes. For this purpose multi-fidelity analysis techniques have been considered involving the use of surrogate models. In particular the attention has been focused on the study of a proper methodology to build an approximated model for the estimation of aerodynamic coefficients to be used for performance evaluation in the mission optimization stage. In this case a procedure involving variables screening phase, data-fit surrogate models evaluation and assessment phase and a final crucial global correction phase of the best surrogate model has been proposed.

The work presented in this thesis is dedicated to the study of vehicle rollover and the tyre and suspension characteristics influencing it. Recent data shows that 35.4% of recorded fatal crashes in Sports Utility Vehicles (SUVs) included vehicle rollover. The effect of rollover on an SUV tends to be more severe than for other types of passenger vehicle. Additionally, the number of SUVs on the roads is rising. Therefore, a thorough understanding of factors affecting the rollover resistance of SUVs is needed. The majority of previous research work on rollover dynamics has been based on low fidelity models. However, vehicle rollover is a highly non-linear event due to the large angles in vehicle body motion, extreme suspension travel, tyre non-linearities and large forces acting on the wheel, resulting in suspension spring-aids, rebound stops and bushings operating in the non-linear region. This work investigates vehicle rollover using a complex and highly non-linear multi-body validated model with 165 degrees of freedom. The vehicle model is complemented by a Magic Formula tyre model. Design of experiment methodology is used to identify tyre properties affecting vehicle rollover. A novel, statistical approach is used to systematically identify the sensitivity of rollover propensity to suspension kinematic and compliance characteristics. In this process, several rollover metrics are examined together with stability considerations and an appropriate rollover metric is devised. Research so far reveals that the tyre properties having the greatest influence on vehicle rollover are friction coefficient, friction variation with load, camber stiffness, and tyre vertical stiffness. Key kinematic and compliance characteristics affecting rollover propensity are front and rear suspension rate, front roll stiffness, front camber gain, front and rear camber compliance and rear jacking force. The study of suspension and tyre parameters affecting rollover is supplemented by an investigation of a novel anti-rollover control scheme based on a reaction wheel actuator. The simulations performed so far show promising results. Even with a very simple and conservative control scheme the reaction wheel, with actuator torque limited to 100Nm, power limited to 5kW and total energy consumption of less than 3kJ, increases the critical manoeuvre velocity by over 9%. The main advantage of the proposed control scheme, as opposed to other known anti-rollover control schemes, is that it prevents rollover whilst allowing the driver to maintain the desired vehicle path.

Thesis (Ph.D)--Aerospace Engineering, Georgia Institute of Technology, 2010.
Committee Chair: Mavris, Dimitri; Committee Member: Beeson, Don; Committee Member: Duncan, Scott; Committee Member: German, Brian; Committee Member: Kumar, Viren. Part of the SMARTech Electronic Thesis and Dissertation Collection.

Thesis: S.M., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 2017.
Cataloged from PDF version of thesis.
Includes bibliographical references (pages 177-180).
Many fluid flows in engineering are turbulent and require the use of computational fluid dynamics (CFD) for design purposes. Optimization with CFD has largely been limited to low-fidelity simulation methods, such as Reynolds Averaged Navier Stokes (RANS), due to current computational capabilities. However, RANS has been shown to lack sufficient accuracy for certain flows. This thesis presents CFD simulation of a 180 degree U-bend square duct using low-fidelity steady RANS and high-fidelity wall-resolved Large Eddy Simulation (LES) models. The LES solution is shown to match experimental results, whereas the RANS solution is not sufficiently accurate. A process for training a RANS eddy viscosity field using the LES solution is provided. This approach is based on solving an inference problem by comparing the RANS calculations to the LES solution and tuning cell-based turbulent viscosity values. This multi-fidelity framework is intended to highlight that high-fidelity solutions can be used to improve even the simplest RANS turbulence models. The adjoint method is used for efficient gradient-based optimization of the turbulent viscosity on a U-bend channel to minimize the velocity solution error. Other objective functions are explored to check the uniqueness of the optimized turbulent viscosity. Sensitivity of the optimized result to the numerical convection scheme is presented to help provide insight for future optimization of turbulence models. The optimized turbulent viscosity is also used on a modified U-bend channel to demonstrate the applicability of the method on new geometries.
by Michael Elia Hayek.
S.M.

To decrease the computational cost of genetic algorithm optimizations, surrogate models are used during optimization. Online update of surrogate models and repeated exchange of surrogate models with exact model during genetic optimization converts static optimization problems to dynamic ones. However, genetic algorithms fail to converge to the global optimum in dynamic optimization problems. To address these problems, a multiploid genetic algorithm optimization method is proposed. Multi-fidelity surrogate models are assigned to corresponding levels of fitness values to sustain the static optimization problem. Low fidelity fitness values are used to decrease the computational cost. The exact/highest-fidelity model fitness value is used for converging to the global optimum. The algorithm is applied to single and multi-objective turbine blade aerodynamic optimization problems. The design objectives are selected as maximizing the adiabatic efficiency and torque so as to reduce the weight, size and the cost of the gas turbine engine. A 3-D steady Reynolds-Averaged Navier-Stokes solver is coupled with an automated unstructured grid generation tool. The solver is validated by using two well known test cases. Blade geometry is modelled by 37 design variables. Fine and coarse grid solutions are respected as high and low fidelity surrogate models, respectively. One of the test cases is selected as the baseline and is modified in the design process. The effects of input parameters on the performance of the multiploid genetic algorithm are studied. It is demonstrated that the proposed algorithm accelerates the optimization cycle while providing convergence to the global optimum for single and multi-objective problems.

During the design process for an aerospace vehicle, decision-makers must have an accurate understanding of how each choice will affect the vehicle and its performance. This understanding is based on experiments and, increasingly often, computer models. In general, as a computer model captures a greater number of phenomena, its results become more accurate for a broader range of problems. This improved accuracy typically comes at the cost of significantly increased computational expense per analysis. Although rapid analysis tools have been developed that are sufficient for many design efforts, those tools may not be accurate enough for revolutionary concepts subject to grueling flight conditions such as transonic or supersonic flight and extreme angles of attack. At such conditions, the simplifying assumptions of the rapid tools no longer hold. Accurate analysis of such concepts would require models that do not make those simplifying assumptions, with the corresponding increases in computational effort per analysis. As computational costs rise, exploration of the design space can become exceedingly expensive. If this expense cannot be reduced, decision-makers would be forced to choose between a thorough exploration of the design space using inaccurate models, or the analysis of a sparse set of options using accurate models. This problem is exacerbated as the number of free parameters increases, limiting the number of trades that can be investigated in a given time. In the face of limited resources, it can become critically important that only the most useful experiments be performed, which raises multiple questions: how can the most useful experiments be identified, and how can experimental results be used in the most effective manner? This research effort focuses on identifying and applying techniques which could address these questions. The demonstration problem for this effort was the modeling of a reusable booster vehicle, which would be subject to a wide range of flight conditions while returning to its launch site after staging. Contour-based sampling, an adaptive sampling technique, seeks cases that will improve the prediction accuracy of surrogate models for particular ranges of the responses of interest. In the case of the reusable booster, contour-based sampling was used to emphasize configurations with small pitching moments; the broad design space included many configurations which produced uncontrollable aerodynamic moments for at least one flight condition. By emphasizing designs that were likely to trim over the entire trajectory, contour-based sampling improves the predictive accuracy of surrogate models for such designs while minimizing the number of analyses required. The simplified models mentioned above, although less accurate for extreme flight conditions, can still be useful for analyzing performance at more common flight conditions. The simplified models may also offer insight into trends in the response behavior. Data from these simplified models can be combined with more accurate results to produce useful surrogate models with better accuracy than the simplified models but at less cost than if only expensive analyses were used. Of the data fusion techniques evaluated, Ghoreyshi cokriging was found to be the most effective for the problem at hand. Lastly, uncertainty present in the data was found to negatively affect predictive accuracy of surrogate models. Most surrogate modeling techniques neglect uncertainty in the data and treat all cases as deterministic. This is plausible, especially for data produced by computer analyses which are assumed to be perfectly repeatable and thus truly deterministic. However, a number of sources of uncertainty, such as solver iteration or surrogate model prediction accuracy, can introduce noise to the data. If these sources of uncertainty could be captured and incorporated when surrogate models are trained, the resulting surrogate models would be less susceptible to that noise and correspondingly have better predictive accuracy. This was accomplished in the present effort by capturing the uncertainty information via nuggets added to the Kriging model. By combining these techniques, surrogate models could be created which exhibited better predictive accuracy while selecting the most informative experiments possible. This significantly reduced the computational effort expended compared to a more standard approach using space-filling samples and data from a single source. The relative contributions of each technique were identified, and observations were made pertaining to the most effective way to apply the separate and combined methods.

In this thesis, we look for multi-fidelity surrogate method to provide an approximation of the output of an input-output relationship using as few model evaluations as possible for uncertainty quantification purpose. We explore the use of multi-fidelity modelling within uncertainty quantification, for which we combine information from simulation models within a hierarchy of fidelity, in seeking accurate and efficient approximation of multi-fidelity model at reduced computational cost. Multi-fidelity methods have been developed on multi-fidelity Monte Carlo, multi-fidelity Kriging and multi-fidelity sparse grids to accelerate the estimation of outputs of high-fidelity models. At first, our approach provides a framework to sample the input parameter space on a sparse grid, and then use sparse grid interpolation as our surrogate. For accurate but expensive high-fidelity models, however, even performing the number of simulations needed for fitting a surrogate may be too expensive. Thus, multi-fidelity is an useful approach to reduce the number of high-fidelity simulations. To further increase the efficiency, a multi-fidelity approach combine results from accurate and expensive high-fidelity simulations with inexpensive but less accurate low-fidelity simulations via the combination technique in order to achieve accuracy at a reasonable cost. We focus attention on multi-fidelity sparse grids in which fidelities are combined inside the surrogate model. We present the multi-fidelity approach via a careful study and generalisation of the sparse grid combination technique and give a general formula for the multi-fidelity approach. The combination technique as a method to achieve a function representation on a sparse grid without having to work with a hierarchical basis. Study of the combination technique often assumes that approximations satisfy an error splitting model. The combination technique is competitive to the classical sparse grid approach with respect to quality and run time and give proof that the combined interpolant is equivalent to the hierarchical sparse grid interpolant.\\ Instead of relying on error and cost rates, an optimisation problem with an analytical result optimal balance the model evaluations across the high-fidelity model and an arbitrary number of low-fidelity models with respect to error and costs. The error analysis of convergence on arbitrary dimensions is applicable to general use cases of the multi-fidelity approach.\\ Our results consists of both theoretical and numerical parts. In the numerical part, we look at the application of the multi-fidelity sparse grid model to the solution of several test functions and partial differential equation (PDE) models particularly focusing on numerical aspects like stability, order of approximation and error convergence. Comparing the mathematical results and the numerical results, some of numerical test cases can obtain the same results as the theoretical analysis. Lastly, a multi-fidelity sparse grid surrogate model was constructed for the Hokkaido-Nansei-Oki tsunami modelling for uncertainty quantification. We demonstrate the experimental results in Okushiri wave flume, which reproduce the maximum value of the time-dependent average tsunami height on top of area of interest. We illustrate the multi-fidelity approach with the number of uncertain input parameters to quantify the uncertainty in the output of tsunami wave shape and properly report the achieved savings. The numerical results can be up to given accuracy with a reduced cost.

Einleitung Studierende der Veterinärmedizin müssen neben umfangreichem theoretischem Wissen zahlreiche praktische Fertigkeiten erlernen. Da jeder Einzelne in seinem eigenen Tempo lernt, besteht ein großer Bedarf an Trainingsmöglichkeiten. Kadaver und lebende Tiere sind selten in ausreichender Menge verfügbar und lebende Tiere sind zudem aus Gründen des Tierwohls nur eingeschränkt zu verwenden. Simulationsmodelle (Modelle von Organismen / Körperteilen) können hier Abhilfe schaffen. Kommerziell erhältliche Modelle sind sehr kostenintensiv und für die Tiermedizin noch nicht flächendeckend erhältlich. Zunehmend werden selbst entwickelte low-fidelity Modelle in der Tiermedizin verwendet. Aufgrund des Mangels an publizierten Daten zu ihrem Einsatz besteht intensiver Forschungsbedarf. Ziele der Untersuchungen In dieser Arbeit sollte untersucht werden, ob einfache, selbst entwickelte Simulationsmodelle (low-fidelity Modelle) erfolgreich in der Lehre eingesetzt werden können. Dazu wurden zwei selbst entwickelte und gebaute Simulationsmodelle evaluiert (Studie 1) und ihr Einsatz in Kombination mit anderen Lehrmedien untersucht (Studie 2). Materialien und Methoden In Studie 1 wurden zwei low-fidelity Modelle zur kaninen Intubation und Katheterisierung entwickelt und evaluiert. Es wurde ein Studiendesign genutzt, das die erworbenen Fertigkeiten zweier Übungsgruppen und einer Kontrollgruppe in einer praktischen Prüfung (OSCE = objective structured clinical examination) am toten Hund vergleicht. Achtundfünfzig Studierende (4. FS) erhielten eine theoretische Einführung zur Intubation und wurden randomisiert auf drei Gruppen aufgeteilt. Gruppe 1 (high-fidelity) übte am kommerziell erhältlichen Intubation Training Manikin, Gruppe 2 (low-fidelity) am entwickelten low-fidelity Modell und die Textgruppe las einen Text, der die Intubation beim Hund beschreibt. Siebenundvierzig Studierende (10. FS) durchliefen dasselbe Studiendesign zum Thema Katheterisierung der Hündin. Sie nutzten das kommerziell erhältliche Female Urinary Catheter Training Manikin, das selbst entwickelte low-fidelity Modell und Lehrtexte. In Studie 2 wurde die Vermittlung zweier spezifischer Fertigkeiten mit Hilfe von Potcasts und Simulationstraining evaluiert. Zwei anleitende Potcasts zu Intubation und Katheterisierung und die oben beschriebenen Modelle wurden innerhalb eines crossover-Studiendesigns genutzt. In dieser Studie sind Potcasts audio-visuell aufbereitete Animationen mit Schritt für Schritt – Anleitungen und Informationen. Die erworbenen praktischen Fertigkeiten zweier Übungsgruppen, die sich in der Art der theoretischen Vorbereitung unterschieden, wurden in einer praktischen Prüfung (OSCE) am toten Hund verglichen. Ein Fragebogen erfasste das Feedback der Teilnehmer. Sechzig Studierende (2. FS) wurden randomisiert auf eine Potcast- und eine Textgruppe aufgeteilt. Die Potcastgruppe sah sich das anleitende Potcast an, die Textgruppe bereitete sich anhand eines Lehrtextes vor. Im Anschluss hatten beide Gruppen separate Übungseinheiten an den low-fidelity Modellen ohne Betreuung durch Lehrende. Ergebnisse In Studie 1 schnitten alle Übungsgruppen signifikant besser ab als die Textgruppen. Gruppe 1 (high-fidelity) und Gruppe 2 (low-fidelity) unterschieden sich weder bei der Intubation noch bei der Katheterisierung signifikant in ihren Leistungen. In Studie 2 schnitt die Potcastgruppe beim Thema Intubation signifikant besser ab als die Textgruppe, beim Thema Katheterisierung ergaben sich keine signifikanten Unterschiede. Insgesamt hatte das Simulationstraining den Studierenden Spaß gemacht, das Lernen ohne Betreuer wurde jedoch als Herausforderung empfunden. Schlussfolgerungen Es ist davon auszugehen, dass low-fidelity Modelle genauso geeignet für das Training klinischer Fertigkeiten sein können wie high-fidelity Modelle. Das Training klinischer Fertigkeiten mit Hilfe von Potcasts und low-fidelity Modellen sollte durch Betreuer ergänzt werden, anstatt als alleiniges Lehrmedium für Studierende des ersten Studienjahres Verwendung zu finden. Eigenständiges Lernen klinischer Fertigkeiten, angeleitet durch Potcasts bietet eine Möglichkeit für vertiefendes und wiederholendes Training höherer Semester. Der Einsatz von Simulationsmodellen in der veterinärmedizinischen Ausbildung wächst seit wenigen Jahren stetig. Diese Arbeit leistet einen zeitgerechten Beitrag bei der Evaluierung von Simulationstraining.
Introduction Students of veterinary medicine are expected to acquire various practical skills in addition to a wide range of theoretical knowledge. There is a strong demand for training opportunities, as every individual learns and acquires practical skills at individual pace. For reasons of animal welfare concerns and availability, live animals and cadavers cannot always be used for clinical skills training. Simulation models, which are models of organisms or body parts can be a considerable alternative for clinical skills training. Models that are commercially produced often have a high price and are not available for all skills. Self-made models are increasingly used in veterinary education. Because there is few published data regarding their use, more scientific research is required. Aims of the Investigation The objective of this study was to determine, if self-made low-fidelity models can be successfully used in veterinary medical education. For this purpose, two self-made low-fidelity models were evaluated (study 1) and their use in combination with other teaching tools was analyzed (study 2). Materials and Methods In study 1, two self-made low-fidelity models for simulation of canine intubation and canine female urinary catheterization were developed and evaluated. We used a study design that compares acquired skills of two intervention groups and one control group in a practical examination (OSCE = objective structured clinical examination). Fifty-eight second-year veterinary medicine students received a theoretical introduction to intubation and were randomly divided into three groups. Group 1 (high-fidelity) was then trained on a commercially available Intubation Training Manikin, group 2 (low-fidelity) was trained on our low-fidelity model, and the text group read a text describing intubation of the dog. Forty-seven fifth-year veterinary medicine students followed the same procedure for training urinary catheterization using the commercially available Female Urinary Catheter Training Manikin, our self-made model, and text. Outcomes were assessed in a practical examination on a cadaver using an OSCE checklist. In study 2 we evaluated the teaching of two specific clinical skills using potcasts and low-fidelity simulation training. Two instructional potcasts describing intubation and catheterization and both low-fidelity models described above were used. In our study, potcasts are audio-visual animations that provide the learner with step by step information and instruction on a clinical skill. We used a crossover study design and compared the acquired practical skills of two intervention groups after a different theoretical preparation. A survey captured the participants’ feedback. Sixty first year veterinary medicine students were randomly allocated to two groups, a potcast group and a text group. The potcast group watched a potcast while the text group read an instructional text for preparation. Then both groups had separate self-directed training sessions on low-fidelity models. Outcomes were assessed in practical examinations on a cadaver using an objective structured clinical examination (OSCE) checklist. Results In study 1 all intervention groups performed significantly better than the text groups. Group I (high-fidelity) and group II (low-fidelity) for both intubation and catheterization showed no significant differences. In study 2 the potcast group performed significantly better than the text group in study intubation but no significant differences were observed in study catheterization. Overall, participants enjoyed clinical skills training but experienced self-directed learning as challenging. Conclusion Low-fidelity models can be as effective as high-fidelity models for clinical skills training. Clinical skills training using potcasts and self-directed low-fidelity simulation training should be complemented by supervisor or peer instruction rather than used as exclusive tool for teaching first year veterinary students. We assume though, that self-directed learning instructed by our potcasts can be a valuable chance for deepening and repetitive training of higher semesters. The use of simulation models in veterinary education has been consistently increasing in the past few years. This study is an important, timely contribution to the evaluation of simulation based education.

Vertical-axis wind turbines (VAWTs) represent a valuable alternative to horizontal- axis ones for non-conventional installations like highly turbulent environments and off- shore floating. Due to their inherently complex aerodynamics, nonetheless, characterized by a continuous oscillation of the angle of attack on the blade, often above the static stall limit, their development has fallen behind. The nature of this technological gap is twofold. On the one hand, new strategies for increasing their performance and stabilizing their operation must be developed. In this perspective, one can work both on improving the airfoils' performance with passive flow control devices and on controlling the angle of attack (e.g., with active blade pitching). On the other hand, analysis tools with a fidelity higher than the ubiquitous Blade Element Momentum (BEM) method are needed. Blade- resolved Computational Fluid Dynamics (CFD) has shown its potential for this application, but its elevated computational cost makes it suitable only for analyses of few cases. In between the two, interest is being devoted at developing hybrid approaches, able to conjugate the accuracy of CFD and the calculation cost reduction coming from a lumped-parameter modeling of airfoil aerodynamics. Among the different methodologies available, the Actuator Line Method (ALM) is particularly promising. Several shortcomings of this approach have not been solved by the scientific community yet, in particular: the spreading of aerodynamic forces in the domain, the sampling of the angle of attack from the resolved flow field, and a robust dynamic stall modelling. Moving from this background, this thesis presents a comprehensive approach to the power augmentation of vertical-axis rotos. Two strategies have been investigated, i.e., Gurney Flaps and active blade pitching. To this end, high-fidelity, blade-resolved CFD simulations were sided by a new generation ALM tool, here developed within the commercial solver ANSYS® FLUENT®. In the effort of tailoring the ALM to this type of machines, different features have been implemented and discussed in the present study, including a novel strategy for sampling of the angle of attack from the resolved flow field, a sensitivity analysis on the force spreading within the domain and several sub-models to account for secondary aerodynamic effects. Particular attention has been given to dynamic stall and to tip effects modelling. Validation on selected test cases, for which high-fidelity blade forces and wake field data were available from wind tunnel tests and blade-resolved simulations, has proved the reliability of the developed ALM tool. Effectiveness of the proposed power augmentation strategies has been demonstrated also via their application to a hydrokinetic rotor (HVAT - hydrokinetic vertical-axis turbine), designed in collaboration with an industrial partner. Both ALM and blade-resolved CFD simulations showed a simultaneous increase in the turbine aerodynamic efficiency and a reduction in fatigue loading.

abstract: This investigation develops small-size reduced order models (ROMs) that provide an accurate prediction of the response of only part of a structure, referred to as component-centric ROMs. Four strategies to construct such ROMs are presented, the first two of which are based on the Craig-Bampton Method and start with a set of modes for the component of interest (the β component). The response in the rest of the structure (the α component) induced by these modes is then determined and optimally represented by applying a Proper Orthogonal Decomposition strategy using Singular Value Decomposition. These first two methods are effectively basis reductions techniques of the CB basis. An approach based on the “Global - Local” Method generates the “global” modes by “averaging” the mass property over α and β comp., respectively (to extract a “coarse” model of α and β) and the “local” modes orthogonal to the “global” modes to add back necessary “information” for β. The last approach adopts as basis for the entire structure its linear modes which are dominant in the β component response. Then, the contributions of other modes in this part of the structure are approximated in terms of those of the dominant modes with close natural frequencies and similar mode shapes in the β component. In this manner, the non-dominant modal contributions are “lumped” onto the dominant ones, to reduce the number of modes for a prescribed accuracy. The four approaches are critically assessed on the structural finite element model of a 9-bay panel with the modal lumping-based method leading to the smallest sized ROMs. Therefore, it is extended to the nonlinear geometric situation and first recast as a rotation of the modal basis to achieve unobservable modes. In the linear case, these modes completely disappear from the formulation owing to orthogonality. In the nonlinear case, however, the generalized coordinates of these modes are still present in the nonlinear terms of the observable modes. A closure-type algorithm is then proposed to eliminate the unobserved generalized coordinates. This approach, its accuracy and computational savings, was demonstrated on a simple beam model and the 9-bay panel model.
Dissertation/Thesis
Doctoral Dissertation Mechanical Engineering 2017