Vahid Ghorbanian

Design and optimization problems typically require running thousands of motor simulations which can take several hours, if not days. To overcome this bottleneck, surrogate models are often used in engineering design to expedite the process. When inverter-fed motor drives are involved, the cost of generating a finite element (FE) database from a pulsewidth modulated (PWM) current simulations to fit such models can be prohibitive. This article compares an ensemble of surrogate models of synchronous motors generated with sinusoidal and PWM stator excitations, in terms of computational burden and performance. The main contribution resides in showing whether computational resources can be saved with sinusoidal excitation models without compromising the optimization results of a real inverter-fed permanent magnet synchronous motor design.

A new method for predicting efficiency maps of electric motor drives is proposed using deep learning. Since many operating points need to be simulated using finite element analysis to estimate the efficiency map of a single motor drive topology with certain geometry dimensions and materials, incorporating the whole efficiency map into the design optimization process is an overwhelmingly time-consuming task and may be impossible, depending on the availability of computational resources. Therefore, two deep learning (DL) network architectures are employed in this work to quickly and accurately predict efficiency maps. In the first architecture, the two important stages of efficiency map calculations, i.e. the flux linkage maps and the torque-speed envelopes, are replaced by a combination of recurrent and feedforward neural networks to account for geometric and operating point variations. For the second architecture, an end-to-end DL model was trained to predict the same efficiency maps. The output of the proposed methods has a good match with that of the finite element solution, indicating a high prediction accuracy as well as low run-time useful for design and optimization problems.

Design and optimization problems typically require running thousands of motor simulations, which could take several hours if not days. Finding alternative means of reducing the solution time has recently gained research interest. Surrogate models can emulate the outputs of computer simulations with less computational effort. This paper proposes the use of surrogate models to predict the acoustic noise, applied to an interior permanent-magnet synchronous motor (IPMSM). The simulation procedure involves using finite element analysis to evaluate the acoustic performance across a design space of stator and rotor geometric variations. Then, four different classes of surrogate models are used to learn a portion of the design space before attempting to generalize and make predictions in a much larger space with relatively less computational burden. It is demonstrated that the trained models can be considered as appropriate replacements of the time-consuming FEA for future design and optimization problems of the same motor case study.

This paper investigates the feasibility of novel data-driven deep learning (DL) models to predict the solution of Maxwell's equations for low-frequency electromagnetic (EM) devices. With ground truth (empirical evidence) data being generated from a finite-element analysis solver, a deep convolutional neural network is trained in a supervised manner to learn a mapping for magnetic field distribution for topologies of different complexities of geometry, material, and excitation, including a simple coil, a transformer, and a permanent magnet motor. Preliminary experiments show DL model predictions in close agreement with the ground truth. A probabilistic model is introduced to improve the accuracy and to quantify the uncertainty in the prediction, based on Monte Carlo dropout. This paper establishes a basis for a fast and generalizable data-driven model used in the analysis, design, and optimization of EM devices.

The effects of temperature-dependent material properties, including those of silicon steel and permanent magnet (PM) materials, on the design process of inverter-fed PM motors are investigated in this paper. The magnetization curves of these materials are measured experimentally at different temperatures and are imported into a finite-element package. Then, a statistical study of an integrated motor-drive system consisting of an interior PM motor connected to a space vector modulated inverter is performed by assigning four different temperatures and building a design space of various objectives including the integrated system efficiency, speed limits, motor startup quality, torque ripple, total harmonic distortion, and power factor as the geometrical dimensions are varied. The results reveal that incorporating the realistic material properties into the design process is essential. Moreover, different motor-drive performances have totally different variation rates as the material properties are changed. Finally, a set of optimal designs at different temperatures is proposed to show the effect of changing material properties.

Electrical machines are one of the most growingly manufactured devices which are widely used not only in industries but also in domestic applications. Due to their higher efficiency level, they are common sources of electromechanical energy conversion. Various applications including control, automotive and power generation applications are served by means of these devices, and no one can deny their ability in being precisely controlled in terms of position, speed and torque which are some of the essential quantities in electromechanical devices.

This paper provides the fundamentals of integrated motor-drive system design knowledge that could be used as a basis to change the existing machine design approach from being a separate machine design tool to a more advanced engineering package in which the inverter performance can also be considered. Various users' preferences including motor performances in the transient, rated, and flux-weakening operations along with the inverter quality are studied by means of a detailed cosimulation process which utilizes finite element method, MATLAB, and SIMULINK packages to build the framework based on which magnetic, electric, and electronic devices and quantities are modeled, simulated, and postprocessed. A case study of an interior-permanent-magnet motor connected to a field-oriented controlled drive is investigated and the design process concepts are developed by means of a comprehensive statistical analysis. It is shown that incorporating the inverter quality into the design process changes the idea of optimum motor design, and hence, not only the design parameters but also the expectations from motor performances have to be revised. In fact, an integrated motor-drive system design process regarding the best motor operations in the transient, rated, and flux-weakening modes is targeted with the purpose of addressing design challenges of interior-permanent-magnet motors. To this end, the start-up torque, the rise time, the motor efficiency, the torque ripple, the constant power speed range, the characteristic current, the inverter efficiency, and the system cost, which cover a group of important objectives of different applications, are investigated. Finally, a design package will be able to address different designers' expectations more efficiently using the approach proposed herein.