Ziyan Ren

Multi-objective worst-case scenario robust optimal design of switched reluctance motor incorporated with FEM and Kriging

In this paper, one multi-objective robust optimization algorithm is applied to the optimal design of switched reluctance motor. The performance robustness against uncertainty in design variables is evaluated utilizing the first order sensitivity assisted-worst case scenario approximation. In order to reduce the computing cost required by the finite element analysis, the Kriging surrogate model is used to predict performance of switched reluctance motor during optimization process. With the help of multi-objective particle warm optimization algorithm, a set of robust optimal designs are obtained through making a balance between maximizing average torque and minimizing torque tipple.

Multi-objective optimization approach to reliability-based robust global optimization of electromagnetic device

Purpose – The purpose of this paper is to propose a multi-objective optimization algorithm, which can improve both the performance robustness and the constraint feasibility when the uncertainty in design variables is considered. Design/methodology/approach – Multi-objective robust optimization by gradient index combined with the reliability-based design optimization (RBDO). Findings – It is shown that searching for the optimal design of the TEAM problem 22, which can minimize the magnetic stray field by keeping the target system energy (180 MJ) and improve the feasibility of superconductivity constraint (quenching condition), is possible by using the proposed method. Originality/value – RBDO method applied to the electromagnetic problem cooperated with the design sensitivity analysis by the finite element method.

A New Reliability Analysis Algorithm With Insufficient Uncertainty Data for Optimal Robust Design of Electromagnetic Devices

Due to the existence of randomness and fuzziness in engineering problems, the reliable performance prediction is very complex. In addition, in the early design stage of a new product, the insufficient information on uncertainty further aggravates the difficulty of performance prediction and robustness evaluation. This paper suggests a new reliability analysis method using possibility theory to evaluate constraint feasibility even with insufficient uncertainty data in design variables. In order to save the computing time, two strategies of approximating performance constraints are proposed: 1) design sensitivity-based method and 2) dynamic Kriging-based method. Through applications to analytic example and engineering problem, the proposed methods are investigated and discussed.

Numerically Efficient Algorithm for Reliability-Based Robust Optimal Design of TEAM Problem 22

The robust optimization and reliability-based optimization have been proven effective to deal with uncertainties in design variables. However, there are scarcely any publications about comparison of robust and reliable designs, not to mention the combination of robustness and reliability in the electrical engineering. In this paper, the optimal design of superconducting magnetic energy storage system is taken as an example to comparatively investigate robust and reliable designs. Furthermore, the performance robustness and constraint feasibility are integrated into a single optimization model - reliability-based robust design optimization (RBRDO). The proposed RBRDO formulation yields results that provide new alternatives to the designer.

A Possibility-Based Robust Optimal Design Algorithm in Preliminary Design Stage of Electromagnetic Devices

In the early design stage of an electromagnetic device, sufficient information on uncertainties of design variables is not available. Therefore, a reliable optimal design cannot be achieved by the conventional reliability analysis, in which the probabilistic method is applied. This paper, considering the insufficient uncertainty data, proposes a possibility-based optimal design algorithm to get a robust and reliable optimal design of electromagnetic devices. The suggested algorithm adopts a possibility analysis utilizing the fuzzy set theory. In addition, to mitigate the expensive performance analysis during possibility analysis, the design sensitivity analysis is employed to construct a surrogate model. Finally, the developed algorithm is validated through applications to several examples.

A Novel Subdivision-Based Optimal Design Algorithm for Multidimensional Electromagnetic Problems

For the large-scale electromagnetic problems, the existing optimization methods normally cannot search for a global optimal solution in the design space accurately and have a lower numerical efficiency. In order to mitigate these problems, this paper presents a novel optimal design algorithm based on the subdivision strategy. In the proposed algorithm, first, the whole design space is decomposed into a set number of subregions. Correspondingly, the problem is optimized simultaneously by applying any global optimizer to each subregion. The utilization of subregion strategy in the global optimizers can guarantee a faster convergence and a wider diversity of solutions. The implementation of parallel optimization will effectively reduce the expensive computational cost, especially for the design of the large-scale/multidimensional electromagnetic problems. Finally, the numerical efficiency and the searching ability of the proposed algorithm are validated through applications to two multidimensional electromagnetic problems. One is the design of a superconducting magnetic energy storage (SMES) system, and the other is the design of a Thomson-coil actuator used in the arc eliminator system.

Optimal Design of Powder-Aligning and Magnetizing Fixtures for an Anisotropic-Bonded NdFeB Permanent Magnet

This paper presents a systematical optimal design method for powder aligning fixture during forming process and magnetizing fixture during magnetizing process for an anisotropic-bonded NdFeB magnet. Before the forming process, the mixture ratio of magnet powder and resin, molding tool temperature, and filling density are selected to improve the magnetic performance. During the forming process, a newly multiobjective optimization model is proposed to obtain the required orientation and maximum aligning field. Thus, in the magnetizing process, an anisotropic-bonded NdFeB magnet with required orientation and high residual magnetic flux density is obtained by applying a single-objective optimization process. To accurately predict the residual magnetic flux density, the transient finite element method combined with the Jiles-Atherton hysteresis model, taking the aligning field into account.

A Novel Subregion-Based Multidimensional Optimization of Electromagnetic Devices Assisted by Kriging Surrogate Model

A novel subregion-based optimization strategy utilizing an adaptive dynamic Taylor Kriging (ADTK) surrogate model is developed for a multidimensional optimal design of electromagnetic devices. In the algorithm, the whole design space is divided into a series of subregion, which has its own local ADTK model with optimal set of basis functions. For all subregions, a global optimal solution is found by using particle swarm optimization with the help of the local ADTK models of the objective and constraint functions. The proposed algorithm improves remarkably the accuracy of the ADTK model by reducing the computational complexity. It also significantly reduces the computational cost especially for an optimal design of large-scale multidimensional problems. Finally, the effectiveness of the proposed method is demonstrated through applications to two benchmark problems: TEAM problems 22 and 25.

Modeling of anisotropic magnetostriction under alternating magnetization based on neural network-FFT model

Purpose This paper aims to investigate the magnetostrictive phenomenon in a single electrical steel sheet, which may cause vibration and noise in the cores of transformers and induction motors. A measurement system of magnetostriction is created and the principal strain of magnetostriction is modeled. Furthermore, the magnetostriction property along arbitrary alternating magnetization directions is modeled. Design/methodology/approach A measurement system with a triaxial strain gauge is developed to obtain the magnetostrictive waveform, and the principal strain is computed in terms of the in-plane strain formula. A three-layer feed-forward neural network model is proposed to model the measured magnetostriction property of the electrical steel sheet. Findings The principal strain of magnetostriction of the non-oriented electrical steel has strong anisotropy. The proposed estimation model can be effectively used to model the anisotropic magnetostriction with an acceptable prediction time. Originality/value This paper develops the neural network combined with fast Fourier transform (FFT) to model the principal strain property of magnetostriction under alternating magnetizations, and its validation has been verified.

Subdivision-based optimal design of multi-dimensional electromagnetic devices

A novel method that subdivides the design space region into many subregions is introduced in this study for the efficient optimal design of electromagnetic devices. The particle swarm optimization algorithm is applied to each subregion, which can be run simultaneously. The proposed algorithm is invalidated by using two engineering design problems - superconducting magnetic energy storage system and Thomson coil actuator.