Semester theses focus on optimisation methods useful for construction electric machines. Contain elementary sorting optimization methods and brief description of their algorithms. In these theses is foreshadow how shape of asynchronous stator core should be optimized for better effectivity. Next chaps are discuss about application optimization methods to concrete induction motor.

This thesis deals with design of appropriate optimization algorithms for purposes of newly developed tool Mechlab’s parameter estimation, which serves for parameter estimation of simulation models in Matlab/Simulink. Levenberg-Marquardt algorithm had been chosen among other gradient methods. On the other hand, genetic algorithm and simulated annealing had been chosen from category of soft computing techniques to be implemented. Chosen algorithms were tested on artifical problem of mechanical oscilator and also on real datasets from electronic throttle. Proposed simulated annealing worked in both cases whith sufficient results, nevertheless was time-costly. For the oscilator problem, all the algorithms provided accurate solutions, but in the case of real dataset, Levenberg-Marquardt functionality was limited, while genetic algorithm still provided excelent results.

La complexité de la chimie joue un rôle majeur dans la simulation numérique de la plupart des écoulements réactifs industriels. L'utilisation de schémas cinétiques chimiques détaillés avec les outils de simulation actuels reste toutefois trop coûteuse du fait des faibles pas de temps et d'espaces associés à la résolution d'une flamme, bien souvent inférieurs de plusieurs ordres de grandeur à ceux nécessaires pour capturer les effets de la turbulence. Une solution est proposée pour s'affranchir de cette limite. Un outil automatisé de réduction de schémas cinétiques est développé sur la base d'un ensemble de trajectoires de références construites dans l'espace des compositions pour être représentatives du système à simuler. Ces trajectoires sont calculées à partir de l'évolution de particules stochastiques soumises à différentes conditions de mélange, de réaction et d'évaporation dans le cas de combustible liquide. L'ensemble est couplé à un algorithme génétique pour l'optimisation des taux de réaction du schéma réduit, permettant ainsi une forte réduction du coût calcul. L'approche a été validée et utilisée pour la réduction de divers mécanismes réactionnels sur des applications académiques et industrielles, pour des hydrocarbures simples comme le méthane jusqu'à des hydrocarbures plus complexes, comme le kérosène en incluant une étape optimisée de regroupement des isomères. / Complex chemistry is an essential ingredient in advanced numerical simulation of combustion systems. However, introducing detailed chemistry in Computational Fluid Dynamics (CFD) softwares is a non trivial task since the time and space resolutions necessary to capture and solve for a flame are very often smaller than the turbulent characteristic scales by several orders of magnitude. A solution based on the reduction of chemical mechanisms is proposed to tackle this issue. An automated reduction and optimisation strategy is suggested relying on the construction of reference trajectories computed with the evolution of stochastic particles that face mixing, evaporation and chemical reactions. The methodology, which offers strong reduction in CPU cost, is applied to the derivation of several mechanisms for canonical and industrial applications, for simple fuel such as methane up to more complex hydrocarbon fuels, as kerosene, including an optimised lumping procedure for isomers.

<div>With the upsurge of greenhouse gas emissions and rapid depletion of fossil fuels, the pressure on the transportation industry to develop new vehicles with improved fuel economy without sacrificing performance is on the rise. Hybrid Electric Vehicles (HEVs), which employ an internal combustion engine as well as an electric motor as power sources, are becoming increasingly popular alternatives to traditional engine only vehicles. However, the presence of multiple power sources makes HEVs more complex. A significant task in developing an HEV is designing a power management strategy, defined as a control system tasked with the responsibility of efficiently splitting the power/torque demand from the separate energy sources. Five different types of power management strategies, which were developed previously, are reviewed in this work, including dynamic programming, equivalent consumption minimization strategy, proportional state-of-charge algorithm, regression modeling and long short term memory modeling. The effects of these power management strategies on the vehicle performance are studied using a simplified model of the vehicle. This work also proposes an original power management strategy development using a genetic algorithm. This power management strategy is compared to dynamic programming and several similarities and differences are observed in the results of dynamic programming and genetic algorithm. For a particular drive cycle, the implementation of the genetic algorithm strategy on the vehicle model leads to a vehicle speed profile that almost matches the original speed profile of that drive cycle.</div>

Barrier trees are a convenient way of representing the structure of complex combinatorial landscapes over graphs. Here we generalize the concept of barrier trees to landscapes defined over general multi-parent search operators based on a suitable notion of topological connectedness that depends explicitly on the search operator. We show that in the case of recombination spaces, path-connectedness coincides with connectedness as defined by the mutation operator alone. In contrast, topological connectedness is more general and depends on the details of the recombination operators as well. Barrier trees can be meaningfully defined for both concepts of connectedness.

Context. Creating resources for games and 3D environments is an effort consuming process. Some are looking to procedural algorithms to aid in this endeavour but the effort to configure the algorithms can be time consuming in itself. This paper will continue from a set of papers written by Frade et al. where they surrender the process of configuration to the algorithm by using genetic optimization together with a set of fitness functions. This is then tested on procedural generation of height maps.Objectives. The original algorithm utilizes a tree of functions that generates height maps using genetic optimization and a set of fitness functions. The output of the original algorithm is highly dependent on a specic noise function.This paper will investigate if the inverse Fourier transform can be used as an intermediate phenotype in order to decrease the relationship between the set of functions in the algorithm and the types of output.Methods. A reference implementation was first produced and verified. The Fourier transform was then added to the algorithm as an intermediate phenotype together with improvements on the original algorithm. The new algorithm was then put to the test via five experiments, where the output was compared with the reference implementation using manual review.Results. The implementation of Fourier transform that was attempted in this paper exclusively produced noisy output.Conclusions. The modified algorithm did not produce viable output. This most likely due to the behaviour of the Fourier transform in itself and in relation to the implementation of fitness calculation.

The demand for collaborative, easy to use robots has increased during the last decades in hope of incorporating the use of robotics in smaller production scales, with easier and faster programming. Artificial intelligence (AI) and Machine learning (ML) are showing promising potential in robotics and this project has attempted to automatically solve a specific assembly task with Behavior trees (BTs). BTs can be used to elegantly divide a problem into different subtasks, while being modular and easy to modify. The main focus is put towards developing a Genetic algorithm (GA), that uses the fundamentals of biological evolution to produce BTs that solves the problem at hand. As a comparison to the GA result, a so-called Automated planner was developed to solve the problem and produce a benchmark BT. With a realistic physics simulation, this project automatically generated BTs that builds a tower of Duplo-like bricks and achieved successful results. The results produced by the GA showed a variety of possible solutions, a portion resembling the automated planner's results but also alternative, perhaps more elegant, solutions. As a conclusion, the approach used in this project shows promising signs and has many possible improvements for future research.

The aim and objective of this research is to formulate and solve a multi-objective optimization problem for the optimal placement of multiple rows and multiple columns of fixed flat-plate solar collectors in a field. This is to maximize energy collected from the solar collectors and minimize the investment in terms of the field and collector cost. The resulting multi-objective optimization problem will be solved using genetic algorithm techniques. It is necessary to consider multiple columns of collectors as this can result in obtaining higher amounts of energy from these collectors when costs and maintenance or replacement of damaged parts are concerned. The formulation of such a problem is dependent on several factors, which include shading of collectors, inclination of collectors, distance between the collectors, latitude of location and the global solar radiation (direct beam and diffuse components). This leads to a multi-objective optimization problem. These kind of problems arise often in nature and can be difficult to solve. However the use of evolutionary algorithm techniques has proven effective in solving these kind of problems. Optimizing the distance between the collector rows, the distance between the collector columns and the collector inclination angle, can increase the amount of energy collected from a field of solar collectors thereby maximizing profit and improving return on investment. In this research, the multi-objective optimization problem is solved using two optimization approaches based on genetic algorithms. The first approach is the weighted sum approach where the multi-objective problem is simplified into a single objective optimization problem while the second approach is finding the Pareto front. / Dissertation (MEng)--University of Pretoria, 2012. / Electrical, Electronic and Computer Engineering / MEng / Unrestricted

Although the concept of tensegrity structures was invented in the beginning of the twentieth century, the applications of these structures are limited, partially due to their low stiffness. The stiffness of tensegrities comes from topology, configuration, pre-stress and initial axial element stiffnesses.  The first part of the present work is concerned with finding the magnitude of pre-stress. Its role in stiffness of tensegrity structures is to postpone the slackening of cables. A high pre-stress could result in instability of the structure due to buckling and yielding of compressive and tension elements, respectively. Tensegrity structures are subjected to various external loads such as self-weight, wind or snow loads which in turn could act in different directions and be of different magnitudes. Flexibility analysis is used to find the critical load combinations. The magnitude of pre-stress, in order to sustain large external loads, is obtained through flexibility figures, and flexibility ellipsoids are employed to ensure enough stiffness of the structure when disturbances are applied to a loaded structure.  It has been seen that the most flexible direction is very much sensitive to the pre-stress magnitude and neither analytical methods nor flexibility ellipsoids are able to find the most flexible directions. The flexibility figures from a non-linear analysis are here utilized to find the weak directions.  In the second part of the present work, a strategy is developed to compare tensegrity booms of triangular prism and Snelson types with a truss boom. It is found that tensegrity structures are less stiff than a truss boom when a transversal load is applied. An optimization approach is employed to find the placement of the actuators and their minimum length variations. The results show that the bending stiffness can be significantly improved, but still an active tensegrity boom is less stiff than a truss boom. Genetic algorithm shows high accuracy of searching non-structural space. / QC 20110524

The university timetabling problem is an NP-complete problem which schools all over the world face every semester. The aim of the problem is to schedule sets of events such as lectures and seminars into certain time slots without violating numerous specified constraints. This study aimed to automate this process with the help of simulated annealing and compare the results with a genetic algorithm. The input data sets were inspired by the Royal Institute of Technology in Stockholm. The results showed a great run time difference between the two algorithms where the simulated annealing performed much better. They also showed that even though the simulated annealing algorithm was better during all stages, the genetic algorithm had a much better performance in early stages than it had in latter. This led to the conclusion that a more optimized, hybrid algorithm could be created from the two algorithms provided that the genetic algorithm could benefit from the improvements suggested in previous research. / Universitetsschemaläggningsproblemet är ett NP-fullständigt problem som skolor över hela världen måste hantera innan varje termin. Syftet med problemet är att schemalägga händelser, såsom föreläsningar och seminarier, utan att bryta flertalet fördefinierade villkor. Denna studie hade som mål att automatisera denna process med hjälp av algoritmkonstuktionsmetoden simulerad glödgning och sedan jämföra resultatet med en genetisk algoritm. De datamängder som användes är inspirerade av den verkliga situationen på KTH. Resultaten visar stora tidsmässiga skillnader där algoritmen baserad på simulerad glödgning går snabbare. De visar dock också att den genetiska algoritmen har en bättre prestanda i tidigare stadier än i senare. Detta ledde till slutsatsen att en mer optimerad hybridalgoritm kan skapas av de två algoritmerna, förutsatt att den genetiska algoritmen kan dra nytta av förbättringar som föreslagits i tidigare forskning.