Search

Very recent work has demonstrated tremendous improvements in facial expression recognition (FER) on laboratory-controlled datasets. However, recognizing facial expressions under in-the-wild conditions still remains challenging, especially on unseen subjects due to high inter-subject variations. In this paper, we propose a novel Probabilistic Attribute Tree Convolutional Neural Network (PAT-CNN) to explicitly deal with large intra-class variations caused by identity-related attributes, e.g., age, race, and gender. Specifically, a PAT module with an associated PAT loss is proposed to learn features in a hierarchical tree structure organized according to identity-related attributes, where the final features are less affected by the attributes. Then, expression-related features are extracted from leaf nodes. Samples are probabilistically assigned to tree nodes at different levels such that expression-related features can be learned from all samples weighted by probabilities. Furthermore, the proposed PAT-CNN can be learned from limited attribute-annotated samples to make the best use of available data. Experimental results on four spontaneous facial expression datasets, i.e., RAF-DB, SFEW, ExpW, and FER-2013, have demonstrated that the proposed PAT-CNN achieves the best performance when compared to state-of-the-art methods by explicitly modeling attributes. Impressively, a single model PAT-CNN achieves the best performance on the SFEW test dataset when compared to the state-of-the-art methods using an ensemble of hundreds of CNNs. [ABSTRACT FROM AUTHOR]

Multiphase drives exhibit remarkable advantages (e.g., fault tolerance) over three-phase ones. Six-phase drives are particularly attractive, given their moderate complexity and suitability for off-the-shelf three-phase converters. Regarding the stator winding arrangement, the symmetrical one offers superior postfault capabilities in most scenarios. On the other hand, resistance asymmetry in the stator phases or connections may arise due to different causes. The conventional full-range minimum-loss strategy (FRMLS) generates stator-current references under open-phase (infinite resistance) faults so that torque ripple is prevented while minimizing the losses for each torque ($\boldsymbol{d}$ - $\boldsymbol{q}$ current) value and maximizing the torque range; however, this method is unsuitable for unequal resistances of finite value. This article proposes an FRMLS for setting the current references to reach these goals in symmetrical six-phase drives with any resistance asymmetry. The optimum references are found online depending on the resistances, without lookup tables. The phase currents are individually limited by an iterative algorithm, so that minimum stator copper loss (SCL) is achieved over the maximum admissible torque range. In this manner, unlike the conventional FRMLS, minimum SCL and maximum torque range are attained even for finite resistance imbalance. The currents in phases affected by high resistances are suitably reduced. Experimental results are provided. [ABSTRACT FROM AUTHOR]

The traditional neutral point clamped (NPC) inverter has short-circuit problem. The risk of short-circuit can be decreased by using dead-time in the switching signals. However, the dead-time decreases the achievable output voltage and causes distortion in the waveforms. To overcome the short-circuit problem, dual-buck NPC (DB-NPC) and split-inductor NPC (SI-NPC) inverters have been researched. However, the voltage stress of the two external diodes in the DB-NPC inverter is higher. On the other hand, the SI-NPC inverter has a problem of generating huge voltage spikes in the dead-time, which can destroy the semiconductor devices. In addition, the SI-NPC inverter cannot provide reactive power. This article presents a family of NPC inverters consisting of single-phase, three-phase, and cascaded inverters. The proposed inverters have no short-circuit and dead-time issues, therefore no high voltage and current spikes are caused. Also, the dead-time in the switching signals can be minimized. As a result, the magnitude of the output waveforms can be increased, and quality can be improved. Unlike the DB-NPC inverter, the voltage stress of all the semiconductor in the proposed inverter is lower, and unlike the SI-NPC inverter the proposed inverter provides reactive power. In this article, the proposed three-level NPC inverter is analyzed, designed, and tested. The voltage stress of the semiconductor devices in the proposed inverter is half of the source voltage, whereas in the conventional DB-NPC inverter the voltage stress of the two external diodes is the source voltage. In addition to the aforementioned benefits, the proposed cascaded inverter reduces the total number of inductors. To verify the analysis, detailed simulation, and experimental results of the proposed three-level inverter with input voltage 640 V, output power 1.2 kW, and output voltage 220 Vrms are provided. [ABSTRACT FROM AUTHOR]

The modular multilevel converter (MMC) is a promising candidate for high-/medium-voltage high-power applications such as medium voltage ac drives due to its modularity, reliability, and scalability. The main challenge of a conventional MMC is low-frequency operation. Capacitor voltage ripple is inversely proportional to the operating frequency. Hence, operating at low frequency necessitates employing large submodule capacitances to limit voltage ripple or the utilization of energy equalization modules (EEMs). The first option negatively affects converter lifetime, and the second option requires a large number of semiconductor devices and isolating transformers. In this article, a modified MMC topology with self-energy equalization is proposed to ensure converter operation with balanced capacitor voltages with low ripple during zero-/low-frequency conditions. Operating with balanced capacitor voltages with low ripple is achieved by maintaining the same energy level in the involved upper and lower arms. A low device count alternative to the EEMs approach is proposed in this work. Detailed illustration of the operational concept and design of the associated passive components is presented. The performance of the proposed medium voltage ac drive is elucidated through simulation and experimental results. [ABSTRACT FROM AUTHOR]

This article presents the operation of onboard integrated battery chargers tied to a nine-phase EV-drive train, shedding light on the charging performance during the postfault operation, when one or more faulty inverter legs are disconnected. The objective in that case is to ensure balanced grid currents, while maintaining minimum machine losses at zero average torque production. The proposed concept requires a minimalistic change from a control structure perspective to switch between the pre- and postfault cases. Limitations and fault cases are discussed and corroborated with experimental tests on a 1.5-kW prototype.The derived concept is parameter independent and applicable to nine-phase drives, irrespective of the type of rotor used. [ABSTRACT FROM AUTHOR]

Wrapping a computation domain with a perfectly matched layer (PML) is one of the most effective methods of imitating/approximating the radiation boundary condition in the Maxwell and wave equation solvers. Many PML implementations often use a smoothly increasing attenuation coefficient to increase the absorption for a given layer thickness, and, at the same time, to reduce the numerical reflection from the interface between the computation domain and the PML. In discontinuous Galerkin time-domain (DGTD) methods, using a PML coefficient that varies within a mesh element requires a different mass matrix to be stored for every element and therefore significantly increases the memory footprint. In this work, this bottleneck is addressed by applying a weight-adjusted approximation to these mass matrices. The resulting DGTD scheme has the same advantages as the scheme that stores individual mass matrices, namely, higher accuracy (due to reduced numerical reflection) and increased meshing flexibility (since the PML does not have to be defined layer by layer), but it requires significantly less memory. [ABSTRACT FROM AUTHOR]

Nine-Switch inverter (NSI) has been recently introduced for independent control of dual ac motors. Although this topology reduces the number of active switches by 25% compared to conventional inverters, it has a main conceptual shortcoming due to voltage gain limitations. In NSI, the voltage gains of both outputs are relatively low, while their maximum achievable magnitudes depend on the phase angle between the two sets. In order to mitigate this problem, this article introduces a new dual output nine-switch boost inverter (NSBI) topology with a minimum number of passive elements. The detailed analysis of NSBI is presented, while the required modifications to conventional pulsewidth modulation (PWM) strategies of the NSI are explained to ensure an acceptable voltage boosting ratio while a high-quality voltage output is obtained. Among different PWM schemes, a modified discontinuous PWM scheme is found optimum with this proposed topology to minimize the voltage stress across switches and improve the input current and output voltage quality. The proposed topology is also validated under both common-frequency and different-frequency modes of operations. Moreover, a clear picture of the NSBI performance compared with other possible counterparts is introduced. The theoretical findings have been corroborated using simulation and experiments. [ABSTRACT FROM AUTHOR]

The nine-phase six-terminal induction machine has recently been proposed as a promising contender to the conventional six-phase asymmetrical winding machine in terms of torque density, phase current quality, stator winding simplicity, and fault-tolerant capability. However, the relatively lower dc-link voltage utilization of a single neutral arrangement in multiphase machines with multiple three-phase windings represents, in general, a technical challenge when compared to windings with isolated star points. Therefore, this paper proposes a new pseudo six-phase winding layout suitable for medium-voltage high-power induction machines, which employs quadruple three-phase stator winding sets, while providing the same terminal behavior of a nine-phase six-terminal winding. Additionally, like the traditional six-phase winding, two possible neutral arrangements can be configured. The proposed winding configuration provides the same dc-link voltage utilization as in conventional dual three-phase winding machines with isolated neutrals. The effect of the circulating zero-sequence current component experienced with a single neutral arrangement can also be avoided. A 1.5 Hp prototype induction machine is used to experimentally validate the proposed six-phase winding layout under both healthy and fault conditions. [ABSTRACT FROM AUTHOR]

In this article, three different types of pulse generators based on multimodule boost converter ringing circuit are presented, namely, interleaved, stacked, and interleaved–stacked with full design and operational concept. The interleaved version is advantageous in generating a burst of pulses with moderate voltage magnitude, where the used modules are fed from a common dc source. On the other hand, the stacked multimodule version can be used to generate high-voltage pulsed output from synchronized modules with a moderate voltage rating. This can be done by feeding the modules from isolated dc sources while connecting their outputs in series. Finally, the interleaved–stacked version can be used for generating a burst of pulses with high-voltage magnitude. The presented pulse generators are valid for resistive loads with exponential decay pulses. The suggested pulse generators are simple and require semiconductor devices with moderate voltage ratings. For given pulse specifications, a detailed design for the passive components of each version is presented in this article. The simulation and experimental results are used to validate the suggested topologies. [ABSTRACT FROM AUTHOR]

In order to overcome the limitations of the existing classical and solid-state Marx pulse generators, this paper proposes a new modular multilevel voltage-boosting Marx pulse generator (BMPG). The proposed BMPG has hardware features that allow modularity, redundancy, and scalability as well as operational features that alleviate the need of series-connected switches and allows generation of a wide range of pulse waveforms. In the BMPG, a controllable, low-voltage input boost converter supplies, via directing/blocking (D/B) diodes, two arms of a series modular multilevel converter half-bridge sub-modules (HB-SMs). At start up, all the arm's SM capacitors are resonantly charged in parallel from 0 V, simultaneously via directing diodes, to a voltage in excess of the source voltage. After the first pulse delivery, the energy of the SM capacitors decreases due to the generated pulse. Then, for continuous operation without fully discharging the SM capacitors or having a large voltage droop as in the available Marx generators, the SM capacitors are continuously recharged in parallel, to the desired boosted voltage level. Because all SMs are parallelly connected, the boost converter duty ratio is controlled by a single voltage measurement at the output terminals of the boost converter. Due to the proposed SMs structure and the utilization of D/B diodes, each SM capacitor is effectively controlled individually without requiring a voltage sensor across each SM capacitor. Generation of the commonly used pulse waveforms in electroporation applications is possible, while assuring balanced capacitors, hence SM voltages. The proposed BMPG has several topological variations such as utilizing a buck–boost converter at the input stage and replacing the HB-SM with full-bridge SMs. The proposed BMPG topology is assessed by simulation and scaled-down proof-of-concept experimentation to explore its viability for electroporation applications. [ABSTRACT FROM AUTHOR]

The nine-arm modular multilevel converter (9A-MMC) has been recently proposed as a reduced MMC topology variation for six-phase drive applications, with 25% reduction in the number of employed arms and associated components, compared to a standard dual three-phase MMC, however with a limited output voltage amplitude. This paper proposes a hybrid 9A-MMC comprised of half-bridge submodules (SMs) in both the upper and lower arms, and full-bridge SMs in the middle arms. By employing the negative-voltage state of the full-bridge SMs, the hybrid 9A-MMC avoids the limitations imposed on the dc-link voltage utilization, while achieving further reduction in the component count, compared to a standard 9A-MMC with identical half-bridge SMs. The operating principles of the proposed hybrid 9A-MMC are illustrated with mathematical analysis, whereas its characteristics are verified through both simulation and experimentation. An assessment of the proposed topology quantifying its employed components is also provided, in comparison to other MMC-based six-phase machine drives. [ABSTRACT FROM AUTHOR]