Conference Agenda

Design optimization plays more and more important role on the electrical machine design. Currently, both the size optimization, shape optimization, and topology optimization can be employed for the electrical machine design optimization. However these methods have their own characteristics. In this paper, a combined topology-shape optimization method is proposed for optimizing the rotor magnetic barrier of a synchronous reluctance machine. For evaluating the effectiveness of the proposed topology-shape optimization method, the optimization results of synchronous reluctance machine by using the size optimization, shape optimization, and topology optimization are determined as the benchmark.

Amorphous alloy materials used in reactor cores can reduce the core losses, but its large magnetostrictive coefficient and strong stress sensitivity will increase the core vibration. To reduce the vibration of the amorphous alloy core reactor, a magnetic-mechanical coupling topology optimization algorithm considering inverse magnetostriction effect is proposed. Firstly, the force-magnetic coupling relationship of amorphous alloy materials considering the inverse-magnetostriction effect is established based on the quadratic domain rotation model. Combined with the Solid Isotropic Material Penalty (SIMP) model, the relative permeability and Young's modulus of the material in the optimized area are expressed as continuous functions. Secondly, using the finite element method(FEM), the topology optimization of a reactor core is built and simulated with the inductance value as the constraint and the vibration minimization as the optimization objective. Finally, the vibration reduction effect was verified by experiment and simulation. The results show that the vibration acceleration in the air gap area of the core is reduced by 56% after optimization.

In this paper, a novel global search algorithm circular area movement optimization (CAMO), and its hybridization with pattern search method (PSM) are proposed to solve the multimodal optimization problem. The CAMO is an optimization method that creates a circular search area across the problem region and searches the entire problem region using different movement strategies for each type of sample within the area. Also, the hybridization with the PSM supports the fast convergence on adjacent optima from the points that are discovered in global search using the CAMO. The CAMO was applied to the two test functions and the effectiveness was verified through comparison with the niching genetic algorithm. Last, the algorithm is applied to the optimal design of the interior permanent magnet synchronous motor for hybrid electric vehicles.

To simplify the complexity of the control strategy during the switching process from the constant current mode to the constant voltage mode of the wireless charging system, an integrated coil wireless charging system based on a switchable hybrid structure is proposed. Through switching the compensation element at the receiving pad, constant current(CC) and constant voltage(CV) switching can be achieved under the condition of zero phase angle input (ZPA). Meanwhile, through the optimized design of integrated coils and reasonable parameter configuration, the system can still maintain good CC-CV output characteristics when the misalignment happens. The simulation results show that the maximum fluctuation of output voltage and output current is less than 8%, and the charging efficiency can be maintained at a high level within 100mm misalignment in X-axis.

This paper proposes the multimodal optimization based on machine learning techniques such as k-means clustering and kernel support vector machine. The proposed algorithm reduces function calls and improves diversity characteristics. The outstanding performance of the proposed algorithm is verified by comparison with niching genetic algorithm and conventional surrogate model at the two test functions. Finally, we applied the proposed algorithm to the optimal design of interior permanent magnet synchronous motor for electric vehicles.

In this paper, a novel global search algorithm sub region assisted hybrid algorithm (SAHA) and its hybridization with adaptive simplex are proposed to solve the multimodal optimization problem. Conventional global search algorithms often have a problem that searched areas are already explored. But the SAHA method reduces exploration of already searched area and simultaneously enhances diversification strategy. In other words, it reduces meaningless calculations and enhances its ability to find undiscovered optimal point. The adaptive simplex adopted as an intensification strategy supports the fast convergence on adjacent optima from the points that are discovered in global search by the SAHA with less calculation number. The effectiveness of proposed algorithm is verified by comparison with conventional algorithm at the two test functions, and the proposed algorithm is applied to the optimal design of the permanent magnet-assisted synchronous reluctance motor for electric vehicle drive.

In the field of electrical engineering, the multiphysics optimization of electrical equipment has been a research hotspot in recent years. Aiming at the multiphysics optimization problem of multi-objective and multivariate, this paper investigates a new optimization algorithm based on hybrid kernel extreme learning machine (HKELM). Firstly, to improve the modeling precision and efficiency of the HKELM, the sparrow search algorithm is applied for parameter identification of HKELM, the parallel infill strategy based on the distance density and the minimum response surface criteria are adopted. Finally, the sparrow search algorithm assisted HKELM method is applied to the test function and the complex electromagnetic problem.

Aiming at the problem that it is difficult to achieve stable numerical simulation of electromagnetic field at high speed, a convolutional long short-term memory (ConvLSTM) model is proposed to realize the prediction method of electromagnetic field cloud image of high-speed electromagnetic rail launcher. Taking the electromagnetic rail launcher with the exit velocity of 2400 m/s as the calculation case, the mean square error (MSE) of each pixel is in the range of 0.2×10-3-0.7×10-3, which proves the feasibility and correctness of this prediction method and prediction model. Furthermore, the electromagnetic field cloud image in the "pseudo-oscillation" stage is predicted and analyzed, which provides a new idea for the analysis of the electromagnetic thermal dynamic characteristics of high-speed electromagnetic rail launcher.

The simulation of electromagnetic field distribution in the launching process of electromagnetic railgun is a necessary condition for the fine design of armature and track structure. In view of the difficulty in numerical simulation of electromagnetic field stability at high speed, the skin effect and proximity effect were considered, and the physical-augmented convoluted long short-term memory neural network(PCNN) prediction method of electromagnetic field of high speed electromagnetic gun was realized by reconstructing loss function and minimizing loss function. Taking the electromagnetic rail gun as an example, the electromagnetic field data at the stability stage of the numerical solution is obtained as an experimental sample. Experiments show that the prediction method can accurately predict the electromagnetic field of the electromagnetic gun under the premise of satisfying the actual physical constraints, which provides a new way for the prediction of electromagnetic field.

An inverse rheological hysteresis model is proposed in this paper for the first time, and to avoid the problem of requiring too much experimental data to identify this kind of hysteresis model, an efficient and practical parameter determination technique based on the optimization algorithm (genetic algorithm here) and the numerical method of producing the first-order reversal curves (FORCs) is thereby developed. The inverse rheological hysteresis model and its parameter identification method are subsequently applied to the simulate different kinds of hysteresis loops of soft magnetic materials, that are symmetrical loops without dc bias and major loops imbedded with the unsymmetrical minor ones. By comparing these simulated loops with the corresponding measured loops, the accuracy and practicability of the proposed hysteresis model and its identification method are verified, showing appealing features in the fields that need to simulate the hysteresis characteristics accurately and efficently under complex working conditons.

This paper proposes a continuum sensitivity analysis for shape optimization of steady-state electrothermal systems. The targeted electrothermal system in this paper comprises the steady-state heat conduction and eddy current systems. The shape sensitivity formula for the steady-state electrothermal system was derived using the continuum approach. In the derivation of the sensitivity formula, the material derivative concept and adjoint variable technique were used. Two numerical examples having the known solutions were tested to show the feasibility of the derived sensitivity formula.

In this paper, a new solution is proposed for TEAM Problem 35, an international standard example. The multi-objective guided optimization method based on the magnetic field distribution characteristics combines the physical characteristics with the optimization process to improve the convergence rate of the optimization process and the continuity and stability of the Pareto solution. With the uniformity of magnetic field distribution and power loss of solenoid as optimization objectives and coil radius as design variable, the optimization results are compared with the results obtained by the traditional optimization algorithm, which proves that it has good optimization performance.

External rotor permanent magnet machine is a key technology of distributed electric drive system in electric vehicles and it is widely used in wheel hub. In this paper, an external rotor permanent magnet machine for in-wheel applications with fractional slot concentrated winding (FSCW) structure is introduced as a benchmark motor. To elevate performance of the benchmark motor rapidly, a BP neural network is proposed as a surrogate model to describe the relationship between design properties and objectives. The BP neural network model is trained by results of orthogonal experiment based on Taguchi method. The surrogate model can reduce the computation time significantly comparison with FEA model. Adopting multi-objective genetic algorithm (GA) as an optimization method, and surrogate model as a fitness function, the optimization point is confirmed based on Pareto front. Finally, the performance of optimized motor is improved significantly comparison with benchmark motor, the effectiveness of optimization is validated.

This digest proposes optimization design of a novel partitioned stator hybrid excitation permanent magnet vernier machine (PS-HEPMVM) with DC biased sinusoidal current. Compared with traditional permanent magnet vernier machine, it can effectively enhance the torque density, efficiency and has good flux regulation capability. Firstly, the working principle of the proposed PS-HEPMVM is presented, and the influence of different parameters of iron piece, inner stator tooth and inner stator-PM on the performance is analyzed. Then a multi-objective optimization method based on Finite Element Analysis (FEA) and Genetic algorithm is proposed to optimize the performance of the proposed machine. Finally, various FEA simulation, optimization and comparison studies of the proposed machine are performed. The results verified the effectiveness of the proposed machine.

In this work, a Play model based on recurrent neural network is proposed to predict hysteresis characteristics of electrical steel sheet under complex excitation conditions. The proposed model combines neural network with the Play hysteresis operator, replacing the distribution function in the hysteresis model with the trained neural network parameter structure. An automatic selection method for neural network hyperparameters based on genetic algorithm is suggested, which improves the prediction accuracy. Comparing the experimental and model prediction results, the proposed model can accurately predict the hysteresis characteristics of materials under sine, harmonic and DC bias conditions.

In this paper, the distribution of electromagnetic from the two-dimensional (2-D) finite element analysis of the simplified transformer is estimated by the full convolutional neural network (CNN) model U-net improved with ResNet. By changing the geometric dimension, material property and current excitation, the image dataset with different parameters is obtained and additionally expanded by the operation of scaling, flipping and rotating. Based on it, the weight parameters of improved U-net model are trained and the optimization for the model is carried out by hyperparameters searching, which improves the estimation precision of the electromagnetic and temperature field distribution of transformer. The estimation results prove the effectiveness of the method.