This paper proposes a neural current controller for decentralized non-sinusoidal permanent magnet multiphase machines. This architecture has many advantages, such as increased reliability and simplified motor design, but requires new algorithms for decentralized control. The neural network current controller features a so-called hybrid operation, with one part trained offline and a second part working online. Using the machine’s model, the network is first trained offline based on multi-agent reinforcement learning (RL) and is then implemented in the real machine. In a second step, an online calibration using Least Mean Square (LMS) to take account of the uncertainties of the model is applied only to the output layer to reduce computational burden. This solution combines the speed of offline (forward) and the adaptability of online (feedback) training. This proposal has been validated in simulation and tested on a highly non-sinusoidal 7-phase test bench to demonstrate the feasibility of the proposed approach.

Multiphase open-end-winding (OeW) electric machines offer enhanced capabilities that may find a match in high-performance applications requiring further reliability and efficiency. Nevertheless, the control of this highly complex system requires much attention to avoid the appearance of parasitic currents. In topologies with a single DC source and low homopolar inductances, the proper regulation of zero-sequence currents becomes critical, requiring special attention to common-mode voltages (CMV). This work explores the performance of model predictive control (MPC) to regulate five-phase OeW permanent magnet synchronous motor (PMSM) drives, using both standard (single-vector) and multi-vector MPC. To this end, virtual voltage vectors (VVs) with zero average CMV are proposed for the first time for five-phase systems, making a subsequent comparison of single-vector and VV control actions. Simulation results confirm the capability of MPC to supress current harmonics and successfully regulate the five-phase OeW PMSM.

This paper mainly proposes an improved advanced angle field weakening control strategy for multi-phase PMSM. Firstly, voltage vector reference limit of multi-phase PMSM is analyzed and calculated. This method considers the influence of voltage amplitude and phase angle of different harmonic subspaces on the voltage vector reference limit, which greatly improves the accuracy of the calculation. Then, an advanced angle field weakening control strategy based on third harmonic current injection is proposed. The control strategy adjusts the optimal reference current commands of different harmonic subspaces in real time to maximize the voltage limit of the inverter. Finally, the effectiveness of the proposed method are verified by simulation results.

The paper presents a detailed derivation and analysis of the mechanism for the generation of electromagnetic vibration in fractional slot concentrated winding (FSCW) multi-phase permanent magnet synchronous machines (PMSM) with non-sinusoidal back electromotive force, when operating with a supply of non-sinusoidal currents. Firstly, the magnetomotive force (MMF) generated by the stator windings is calculated and analyzed through the product of the winding function and the injected non-sinusoidal current. Secondly, considering the stator slotting effect, the radial electromagnetic force expression of a non-sinusoidal multi-phase PMSM with FSCW is derived based on Maxwell tensor method under the supply of non-sinusoidal currents. Finally, taking a five-phase PMSM with the non-sinusoidal electromotive force (EMF) and FSCW, a multi-physical field finite element model is established to analyze the radial electromagnetic force wave, vibration and noise response, thereby validating the correctness of the theoretical analysis.

This paper presents a new approach to control properly the currents of a dual three-phase PMSM operating in open-circuit fault condition with a wide speed range. Using fault tolerant control strategies proposed in the literature lead to high frequency current components in the rotor frame. Operating at high speed required in some industrial applications induces a strong constraint on the current controllers. In this paper, a second transformation matrix resulting constant currents in the new frame is proposed. However, there is still a coupling between axes. Thus, a simple Adaptive Linear Neuron is proposed to decouple and enhance the performance of the current tracking. Comparative simulation results are shown for a 12slots/8poles dual three-phase PMSM to confirm the validity of the proposed method.

This study aims at eliminating unwanted harmonics in current control of a five-phase non-sinusoidal permanent magnet synchronous machine (PMSM) in an open-end winding configuration. The machine is supplied by two voltage source inverters (VSIs) using a single DC-bus voltage. High-frequency harmonics, caused by the zero-sequence current with the inverter switching frequency, have been significantly reduced by using a proper pulse width modulation (PWM) strategy. Meanwhile, low-frequency current harmonics are generated by unwanted harmonics of the back electromotive force (back-EMF) and by the inverter nonlinearity. In this study, the low-frequency current harmonics are nullified by simple adaptive linear neural networks (ADALINEs) in rotor reference frames combined with the back-EMF compensation. As a result, the quality of current control is improved. The effectiveness of the proposed strategies is verified by numerical results.

For an open-end windings integrated Permanent Magnet Synchronous Machine, the demagnetization of the permanent magnets is analyzed when a transistor is short-circuited and no specific control strategy is adopted. Depending on the temperature, the high currents due to the inverter fault may locally demagnetized the permanent magnets leading to an accelerated aging of the machine and torque loss. A co-simulation, using a Finite Element software for the machine coupled with an average modeling of the transistor, gives interesting local prediction of the machine behavior in healthy and degraded mode.

For a particular five-phase synchronous machine, this paper investigates the sensitivity of a vectorial control strategy on the required peak phase voltage whose value is fundamental for the choice of the DC bus voltage. The specificity of the machine is that the first and third harmonic components of the back electromotive force (back-emf) have the same amplitude. As a consequence, the torque can be produced by one of them or both with suitable currents. This degree of freedom is interesting for optimizing the efficiency and generating high transient torque. However, using two harmonics having the same amplitude leads to a necessity to analyze the constraints on the required phase machine voltage. Considering a Maximum Torque Per Ampere (MTPA) strategy, the paper examines the impact of some parameters such as the phase shift between currents and back-emfs or the ratio between the third and the first harmonic of current on the torque and maximum voltage value. Experimental tests with a limited DC bus voltage have been carried out and compared to the results obtained by a Finite Element Analysis.

This papers presents a sensorless control for five-phase PM synchronous machines. An adaptive method, based on a linear neural network called Adaline (Adaptive Linear Neural Networks), has been achieved to estimate the rotor position with a high precision even at low speed without high frequency signal injection. Non-sinusoidal three-phase PM machines require more complex algorithm for sensorless control because of harmonics in the back-EMF. This is not the case for multiphase PM machines thanks to the property of equivalent machines in the eigenspace. Some given simulation and experimental results in laboratory confirm the possibility of real-time implementation of Adaline networks and the good performance of sensorless control based on this.

This paper presents a new method by using the Artificial Neural Networks (ANNs) for estimating the parameters of the machine which achieving the optimal efficiency of the Synchronous Reluctance Motor (SynRM). This model take into consideration the magnetic saturation, cross-coupling and iron losses. With Finite Element Analysis (FEA), the characteristics of the SynRM including inductances and iron loss resistance are determined. Because of the non-linear characteristics, an ANN is trained to obtain the d-q inductances and the iron loss resistance from Id, Iq currents and rotor speed. After learning process, an analytical expression of the optimal currents is given thanks to Lagrange optimization. Therefore, the optimal currents will be obtained online in real time. This method can be achieved with maximum efficiency and high-precision torque control. Simulation and experimental results are presented to confirm the validity of the proposed method.

An original analytical expression is presented in this paper to obtain optimal currents minimizing the copper losses of a multi-phase Permanent Magnet Synchronous Motor (PMSM) under fault conditions. Based on the existing solutions [i]opt1(without zero sequence of current constraint) and [i]opt2(with zero sequence constraint), this new expression of currents [i]opt3is obtained by means of a geometrical representation and can be applied to open-circuit, defect of current regulation, current saturation and machine phase short-circuit fault. Simulation results are presented to validate the proposed approach.

Multi-phase machines are known for their fault-tolerant capability. However, star-connected machines have no fault tolerance to inverter switch short-circuit fault. This paper investigates the fault-tolerant operation of an open-end five-phase drive, i.e. a multi-phase machine fed with a dual-inverter supply. Inverter switch short-circuit faults are considered and handled with various degrees of reconfiguration. Theoretical developments and experimental results validate the proposed strategies.

This paper presents a new method based on Artificial Neural Networks to obtain the optimal currents, for reducing the torque ripple in a Non-sinusoidal Synchronous Reluctance Motor. Optimal current control has to develop a constant electromagnetic torque and minimize the ohmic losses. In d-q reference frame without homopolar current, the direct and quadrature optimal currents will be determined thanks to Lagrange optimization. A neural control scheme is then proposed as an adaptive solution to derive the optimal stator currents. Thanks to learning capacity of neural networks, the optimal currents will be obtained online. With this neural control, either machine's parameters estimation errors or current controller errors can be compensated. Simulation results using Matlab/Simulink are presented to confirm the validity of the proposed method.