Axial flux permanent magnet (AFPM) machines have recently attracted significant attention due to several reasons, such as their specific form factor, potentially higher torque density and lower losses, feasibility of increasing the number of poles, and facilitating innovative machine structures for emerging applications. One such machine design, which has promising, high efficiency particularly at higher speeds, is of the coreless AFPM type and has been studied in the dissertation together with more conventional AFPM topologies that employ a ferromagnetic core. A challenge in designing coreless AFPM machines is estimating the eddy current losses. This work proposes a new hybrid analytical and numerical finite element (FE) based method for calculating ac eddy current losses in windings and demonstrates its applicability for axial flux electric machines. The method takes into account 3D field effects in order to achieve accurate results and yet greatly reduce computational efforts. It is also shown that hybrid methods based on 2D FE models, which require semi-empirical correction factors, may over-estimate the eddy current losses. The new 3D FE-based method is advantageous as it employs minimum simplifications and considers the end turns in the eddy current path, the magnetic flux density variation along the effective length of coils, and the field fringing and leakage, which ultimately increases the accuracy of simulations. After exemplifying the practice and benefits of employing a combined design of experiments and response surface methodology for the comparative design of coreless and conventional AFPM machines with cores, an innovative approach is proposed for integrated design, prototyping, and testing efforts. It is shown that extensive sensitivity analysis can be utilized to systematically study the manufacturing tolerances and identify whether the causes for out of specification performance are detectable. The electromagnetic flux path in AFPM machines is substantially 3D and cannot be satisfactorily analyzed through simplified 2D simulations, requiring laborious 3D models for performance prediction. The use of computationally expensive 3D models becomes even more challenging for optimal design studies, in which case, thousands of candidate design evaluations are required, making the conventional approaches impractical. In this dissertation a new two-level surrogate assisted differential evolution multi-objective optimization algorithm (SAMODE) is developed in order to optimally and accurately design the electric machine with a minimum number of expensive 3D design evaluations. The developed surrogate assisted optimization algorithm is used to comparatively and systematically design several AFPM machines. The studies include exploring the effects of pole count on the machine performance and cost limits, and the systematic comparison of optimally designed single-sided and double-sided AFPM machines. For the case studies, the new optimization algorithm reduced the required number of FEA design evaluations from thousands to less than two hundred. The new methods, developed and presented in the dissertation, maybe directly applicable or extended to a wide class of electrical machines and in particular to those of the PM-excited synchronous type. The benefits of the new eddy current loss calculation and of the optimization method are mostly relevant and significant for electrical machines with a rather complicated magnetic flux path, such is the case of axial flux and of transvers flux topologies, which are a main subject of current research in the field worldwide.