Model-based predictive control that highly depends on system parameters has been widely investigated. In contrast, model-free predictive current control (MFPCC) can be performed based on input or output measured data, rather than on any system model information. In such a strategy, the current gradients due to each of the possible voltage vectors are stored and used to predict future currents. Current gradient knowledge therefore provides a significant foundation for MFPCC. In conventional MFPCC, however, the stagnation existing in the current gradient update always impacts the reliability of current gradients and further worsens the control performance. In this paper, an improved MFPCC with an advanced current gradient updating mechanism is proposed. In the proposed strategy, to guarantee the reliability of the current gradients, two contiguous measured current gradients due to the applied voltage vectors are used to estimate the current gradients for all the possible voltage vectors. This simple method takes advantage of the mathematical relationships between the voltage vectors. In this way, all the current gradients can be obtained within one control period, effectively reducing the stagnation effect in conventional MFPCC. The proposed MFPCC scheme is evaluated on a permanent magnet synchronous machine (PMSM) drive setup to demonstrate its effectiveness

With the standstill frequency-response (SSFR) test, accurate electrical-machine models can be identified. However, it can be a time-consuming method, particularly in case the machine has to be identified at low frequencies. To shorten the required time for identification, in this paper, the response on a broadband signal is measured, resulting in a multisine SSFR test. To generate the broadband signal, a high-power linear amplifier can be applied as a waveform generator. As this signal generator is not commonly available in the field, the application of a voltage-source inverter (VSI) is discussed. The multisine SSFR test with a VSI allows swift evaluation of the influence of frequency, saturation, and cross saturation on the q-and d-axis parameters with a signal generator that is often already available to control the machine. Extensive measurements are performed on several permanent-magnet synchronous machines and the method can be extended to synchronous machines with rotor-field winding. By applying a switching converter instead of a linear amplifier, it can be expected that the identification results are affected by the switching actions. Therefore, multisine SSFR tests with either a VSI or a high-power linear amplifier as well as conventional tests as described in the IEEE standard are performed, and the results are compared.

As inverters are often used in critical applications, reliability is an important issue. In particular, the power electronic switches and gate drivers, the most essential components of the inverter, are vulnerable parts in real live operation. Therefore, this paper focuses on open-switch fault detection for multilevel inverters. When a single-switch open-circuit fault occurs in one of the power electronic switches, the algorithm can detect the fault and the switch that is causing it. The detection is worked out for both a linear resistive inductive load and an induction motor. The proposed algorithm is an extension of an already available finite-set model-based predictive control algorithm. Therefore, no extra hardware or measurements are required. This paper also discusses a suggested method for reconfiguration after fault detection. Computer simulation and experimental verifications validate the proposed methods.

The rotor position is required in many control schemes in electrical drives. Replacing position sensors by machine self-sensing estimators increases reliability and reduces cost. Solutions based on tracking magnetic anisotropies through the monitoring of the incremental inductance variations are efficient at low-speed and standstill operations. This inductance can be estimated by measuring the response to the injection of high-frequency signals. In general, however, the selection of the optimal frequency is not addressed thoroughly. In this paper, we propose discrete-time operations based on a rotating voltage injection at frequencies up to one-third of the sampling frequency used by the digital controller. The impact on the rotation drive, the computational requirement, the robustness, and the effect of the resistance on the position estimation are analyzed regarding the signal frequency.

At low speed, the rotor-position estimation in sensorless control is often carried out based on the evaluation of the phase-current ripples resulting from the supply of high-frequency voltage test signals. However, the rotor-position estimation is affected by cross-saturation in the machine, resulting in less accurate position estimations at higher loads. As the importance of sensorless control of interior permanent-magnet synchronous machines (IPMSMs) increases, it is useful to design IPMSMs in such a way that they are optimized for accurate sensorless control. The goal of this paper is to determine design aspects in the rotor geometry of an IPMSM to minimize the position estimation error due to cross-saturation. Simulations of a sensorless drive are usually based on a state-space model with constant q- and d-axis inductances and no mutual inductances. In this paper, this technique is improved by calculating the inductance matrix from several finite-element models, which allows the study of the effect of variable q- and d-axis inductances and cross-saturation on the performance of the sensorless control. The rotor design is discussed, for both IPMSMs with and without flux barriers, in order to reduce the estimation error caused by cross-saturation.

Conventional model-based predictive current control (MBPCC) suffers from high current ripple and the system parameter dependence problem. To address these issues, a novel reference current slope based modulated model-free predictive control (MFPC) is proposed in this article. In the proposed method, a reference current slope is obtained based on the current deadbeat solution. A reduced amount of active voltage vectors can be then determined in a straightforward way according to the reference current slope and the current slopes for the possible voltage vectors, and thus avoiding the enumerating operation in the conventional MBPCC. To reduce the current ripples, a two-vector modulation strategy is introduced to the proposed method. In addition, the current slopes for the possible voltage vectors are obtained in a model-free manner based on online measured data only to reduce the influence of parameter uncertainties on the controller. Here, the principles of the two-vector modulation are well considered, and the current variations with the input voltage vectors from two sampling time intervals are used to estimate the current slopes for all the voltage vectors. As a result, even though two voltage vectors with variable time durations are applied in each control period, the current slope information can be refreshed every half control period, and thus guaranteeing the reliability and accuracy of current predictions. The proposed method is verified on a permanent magnet synchronous machine (PMSM) setup to demonstrate its improved parameter robustness and steady-state performance.