Book Description

"The role of computers in the design of electrical machines has been a subject of interest to many designers and researchers in the field. As processing units become stronger, it becomes possible for computers to assist in and take over more parts of the design procedure. Since the goal in designing electrical machines is to have a device that operates at maximum possible performance, and since the number of performance parameters is usually greater than one, it is a popular approach to formulate the design of an electrical machine, or parts of it, as a multi-criterion optimization problem. These problems are usually solved with the help of simulation models, numerical methods, and optimization algorithms. In this thesis, two important topics in multi-objective design optimization of electrical machines have been the focus of study.The first issue pertains to the lack of infinite precision in manufacturing and difficulty in controlling the environment of operation. These undesirable factors cause the performance of a design to deviate from its nominal or previously calculated values. This is an important issue in many, if not all, engineering design problems and electric machines are not exempt. Since limiting the impact of these factors on a manufactured device is usually very costly, one that is inherently less sensitive to these effects is sought in the design process. This lack of sensitivity is referred to as robustness, and a higher degree of robustness is always more desirable in an engineering design.In this research work, a metric for quantifying robustness in multi-objective problems has been developed. Additionally, a new role for robustness in the process of optimization, with the goal of finding acceptable trade-offs between the robustness and the performance objectives of a design, is introduced. Several artificial and electromagnetic test problems have been used in order to analyze the performance of the proposed methodologies.The second topic visited in the context of this thesis is the use of statistical analysis techniques in the process of multi-criterion design optimization. In recent decades, with masses of data becoming available in different fields, machine learning and data mining techniques have gained a lot of application. These techniques are used to extract information for artificial intelligence units and provide knowledge to users. Since most of these techniques require large amounts of data to work with, using them in the design of electrical machines was infeasible in the past. But recently, with the processing power of computers growing significantly, it has become possible to simulate large numbers of designs. Furthermore, such databases are usually created when performing optimization on machine design problems.Subsequently, the application of a few machine learning and data mining techniques for facilitating the process of design optimization of electrical machines is studied in this dissertation. It is demonstrated how information regarding the location and the innate dimensionality of the optima in the design space can be extracted and transferred to similar problems. Additionally, the validity of this information transfer strategy has been assessed. Moreover, it is shown how these techniques can help the user in the process of design space exploration and decision making. In this part, two electric machine design cases are used as test beds and the results are presented and explained." --