Conference Agenda

This paper presents a variant of particle swarm optimization (PSO) algorithm, which was developed for a reliable parameter estimation in thermal analysis of electrical machines. The proposed algorithm uses a varied social network, where both number and size of the network (local neighbourhoods) are randomly adjusted during the optimization process. Such approach has been introduced here to assure improved diversity of the PSO and consequently a more reliable and robust search of the solution space. A case study parameter estimation for a reduced-order thermal-equivalent-circuit (TEC) of an electrical machine has been used to demonstrate effectiveness of the proposed method. The analysed black-box parameter estimation relies on the input and output data (demand data) from a short-transient finite-element-analysis (FEA) of a complete machine assembly. The proposed PSO variant has been benchmarked with a selection of the existing PSO algorithms, which employ alternative social network schemes with the network parameters dynamically varied. The statistical data gathered from multiple runs of the PSO-based estimation suggests that the proposed approach offers considerable improvements in terms of accuracy, efficiency, reliability and robustness as compared with the alterative PSO algorithms.

This work presents a shape optimization algorithm for Soft Magnetic Composites (SMCs) subject to isotropic and volume constraints. The maximal isotropic effective permeability is the design target for a fixed volume fraction. A level-set function represents the shape of the iron inclusion. Shape derivative is deduced to determine the descent direction. A numerical application is implemented for SMCs in magnetostatics in 2D. Local minimum is located at a rapid convergence rate.

A global sensitivity analysis method enhanced with a kriging surrogate model is considered in this work. To decide on the importance of many model parameters in the case of a functional output, the Sobol sensitivity indices—being functions of an independent variable—are to be calculated over a domain of interest. Since for expensive-to-evaluate models obtaining such a sensitivity map is highly computationally demanding, the number of required samples is reduced by a kriging approximation. The confidence levels of the kriging model may provide sufficient information on the importance of the parameters that can be subsequently used for model reduction. The advantages of the approach are illustrated through the example of an electromagnetic interference filter.

Finite-element method (FEM) simulations involve expensive operations that lend themselves to parallelism, particularly the matrix assembly and the solving of linear equations. Simulations require heavy computations of the element matrices, their assembly into the global matrix and the solution of the resulting system of linear equations. The acceleration of current electromagnetic simulation software is fostered by the development of parallel hardware, namely manycore CPUs and GPUs. In this short paper, we illustrate the advantage of using GPUs for the assembly of the FEM matrices and the solution of linear systems in electromagnetic software. In particular, we propose a matrix assembly method for FEM electromagnetic simulations leveraging double CUDA atomic operations on recent GPUs. Our GPU implementation of the matrix assembly with atomic operations on shared and global memory is thirteen times faster than the current multithreaded CPU implementation. Thereafter, the assembled linear system is solved on GPUs with iterative solvers based on Krylov methods, namely the preconditioned Quasi-Minimal Residual (QMR) method. The resulting GPU implementation of the linear system solver can achieve a 19x speedup compared to the multithreaded CPU solver.

The research work presented in this communication deals with the topology optimization of electromagnetic devices represented by an electromagnetic 2D finite element model. We investigate multi-material topology optimization to find an optimal device without a priori knowledge of its shape. First, we discuss the state of the art of topology optimization. Afterward, we review methods and algorithms adapted to this framework. Then, we present a numerical example divided into two problems depending on the number of materials to be used (mono-material or multi-material). In conclusion, the multi-material problem brings gains, allows more freedom to search for the optimal topology but is more challenging to solve, and makes genetic algorithm fail.

Magnetorelaxometry imaging denotes a novel imaging modality enabling a noninvasive localization and quantification of superparamagnetic nanoparticles which is a requirement for a number of promising therapeutic approaches including magnetic hyperthermia therapy and magnetic drug targeting to ensure safe and efficient procedures. Magnetorelaxometry imaging and other linear inverse problems were subject to several different approaches targeting the maximization of the imaging and solution qualities for a predefined set of design parameters throughout multiple simulation studies. These studies utilized various figures of merit to quantify different numerical properties of the forward models' system matrices. Nonetheless, due to the lack of comparable results among the tested approaches, it was impossible to identify the most effective design strategy for magnetorelaxometry imaging setups. Hence, we formulate cost functions based on these figures of merit, enabling an optimization of the magnetic excitation fields that govern the magnetorelaxometry imaging system matrices and a thorough comparison of the different figures of merit. Initialized using two magnetorelaxometry imaging setups, the optimizations with respect to the various cost functions allow a fair and objective comparison of the achievable imaging qualities. The matrix condition number is identified as the most reliable and effective way to enhance the reconstruction accuracies in the course of this extensive simulation study, unprecedented in magnetorelaxometry imaging.

Unlike general spoke-type PMSMs, double-layer spoke-type PMSMs can utilize high reluctance torque by increasing the difference between d-axis and q-axis reluctance. However, as the difference in magnetic resistance increases, vibration and noise are generated, which adversely affects the mechanical part and shortens the life of the motor. Although this problem seemed to be solved by applying core skew in the previous study, it was confirmed that the axial force caused by the axial leakage flux occurred in the maximum torque per ampere (MTPA) control section and the torque ripple was increased. Therefore, in this paper, a model that can apply symmetrical core skew and reduce axial force is proposed. First, the causes of axial force generated in previous studies were analyzed. Based on the analysis of these causes, a new symmetrical core skew structure was proposed, and its justification was verified through FEA.

The objective of this paper is to demonstrate the relevance of using a global deterministic algorithm to solve optimal design problems even with computationally expensive black-boxes. The authors highlight the efficiency of the proposed method by applying recent improvements of their methodology to problems of magnetic coupling design.

This digest presents the optimal design of a permanent magnet synchronous generator (PMSG). FE models are used to construct metamodels which are then used to build the objective functions that define the performance of the PMSG. Kriging modeling is used together with the design of experiments based on Latin Hypercube sampling. The usage of surrogate models allows to speed up the optimization process while keeping accuracy because they are developed from FEA. On the other hand, it has been reported that NSGA-III is better than NSGA-II. The NonSorting Genetic Algorithm III can solve multi and many objective optimization problems. The above is useful in electrical machines where it can be found many objectives to minimize o maximize.

Power transformer modelling and simulation can require detailed material and design information. However, this detailedtransformer information is usually not entirely available. Therefore, we use a single-phase supplement excitation setup to measurethe transformer core hysteresis characteristic. In this study we compare two transformer network models, based on standard dataand supplement measurements, with measurement from a 3-limb, 2-winding, 50 kVA power transformer. The first model uses thecapacitance-permeance analogy to model the transformer in the magnetic domain. The second model uses the principle of duality andthe Jiles-Atherton model for hysteresis implementation. We have found that both transformer network models show a deviation in power losses below 10 % compared to the single-phase measurement setup. In addition, the validity of the models under AC and DC excitation is analysed using measurement from no-load, back-to-back and back-to-back tests, where the transformer is excited with AC and superimposed DC. These results may be useful for studies where already installed power transformers need to be modelledincluding their hysteresis characteristic.

The interest of industry domain and research community for the performance evaluation of 3D-printed electrical machines grows rapidly recently, focusing on challenges for development of low cost small size high speed motors. For this reason, a recently realized magnetic iron polylactic acid filament with low mass and cost is considered in this paper, which is very promising for the construction of low-power high-speed permanent magnet motor's stator cores. Also, the machine's casing and other individual parts are made of polyethylene terephthalate. The motor's behavior is evaluated through both finite element analysis and experimental results.

The paper investigates demagnetization effects in V-shaped interior permanent magnet motors under loading conditions and temperature variations which could cause demagnetization of permanent magnets. Four demagnetization models are considered involving different computational effort, namely the limited model, the linear, the exponential and the stepping ones applied for the representation of both NdFeB and SmCo permanent magnet types. A particular regression modeling technique is proposed in finite element analysis, enabling computational efficiency in conjunction with demagnetization effects precision under all conditions considered.

For the design optimization of a novel bendable permanent magnet tubular linear synchronous motor (PM-TLSM) for applications in human support systems, this paper presents an analytical method for the computation of the motor force. In this method, the magnetic field generated by permanent magnets in the mover is calculated, while the field generated by the stator coils is preliminarily neglected. From the mover field, the force can be determined numerically from the Lorentz-Force acting on the stator coils. The preliminary negligence of the stator field and the utilization of tangential symmetry and axial periodicity in cylindrical coordinates reduces the problem to a mathematical model, which is easy to handle and therefore enables a fast but yet accurate force computation. The method is validated by Finite Element (FE) simulation. Finally, it is used for the optimization of the PM-TLSM winding geometry regarding the force-to-weight-ratio for an ironless design variant and a variant containing ferromagnetic material.

The application of surrogate models in engineering design is becoming popular due to their mimicking capabilities. In the design of electric motor drives for example, where a control-drive circuit is connected directly to the electric motor, the finite element simulations of the integrated motor drive system can be time consuming. A complete motor simulation can run into several hours, if not days. Therefore, this paper addresses the computational burden associated with the design process by replacing the finite element models with surrogate models in an effort to expedite the design process. Briefly, the procedure involves filling the design space with multiple motor geometries created from an 8 pole 12 slot surface-mounted permanent magnet synchronous motor. Then, a multiphysics performance evaluation is performed over the design space for different switching frequencies over a range of DC bus voltages. The design variables of the motor, the inverter and the calculated performance objectives are used to train an ensemble of surrogate models to obtain the objective functions that could potentially be used as replacements of the finite element models in order to reduce the solution time.

This paper proposes a computationally efficient thermal study method that substitutes Computational Fluid Dynamics (CFD) with Finite Element Analysis (FEA) involving minimal loss of accuracy and omission of the increased computational load. The suggested technique contributes towards the realization of multiphysics optimization of high-speed electrical machines that effectively includes critical thermal evacuation phenomena. The strategy is regression-based and succeeds in constructing convection coefficient regression models that typically involve many variables and small training data sets. The proposed scheme is verified in indicative high-loss density and high-speed outer rotor Surface Mounted Permanent Magnet (SMPM) motor geometries that an optimization algorithm typically encounters during execution. The reference target specifications correspond to 20kW output power, 15krpm rotating speed and 2kHz electrical frequency.