Search

Audiovisual scenes are pervasive in our daily life. It is commonplace for humans to discriminatively localize different sounding objects but quite challenging for machines to achieve class-aware sounding objects localization without category annotations, i.e., localizing the sounding object and recognizing its category. To address this problem, we propose a two-stage step-by-step learning framework to localize and recognize sounding objects in complex audiovisual scenarios using only the correspondence between audio and vision. First, we propose to determine the sounding area via coarse-grained audiovisual correspondence in the single source cases. Then visual features in the sounding area are leveraged as candidate object representations to establish a category-representation object dictionary for expressive visual character extraction. We generate class-aware object localization maps in cocktail-party scenarios and use audiovisual correspondence to suppress silent areas by referring to this dictionary. Finally, we employ category-level audiovisual consistency as the supervision to achieve fine-grained audio and sounding object distribution alignment. Experiments on both realistic and synthesized videos show that our model is superior in localizing and recognizing objects as well as filtering out silent ones. We also transfer the learned audiovisual network into the unsupervised object detection task, obtaining reasonable performance. [ABSTRACT FROM AUTHOR]

The Ultra High Voltage (UHV) converter transformer may generate serious inrush current during closing. It will cause commutation voltage distortion and easily result in commutation failure of converters on the same commutation bus. Due to the change of flux amplitude and phase caused by closing resistance in the Ultra High Voltage Direct Current (UHVDC) system, it is difficult to determine the closing time of the converter transformer when considering the influence of residual flux. Therefore, this paper first analyses the existing closing resistance strategy, and constructs the magnetic bias function according to the influence characteristics of closing resistance on the flux and phase. Then an inrush current suppression strategy based on magnetic bias simulation is proposed for star connection and delta connection of UHV converter transformers, which can accurately calculate the ideal closing time of mutual cancellation between the residual flux and bias flux during closing. Finally, the no-load closing model of the UHV converter transformer is established by PSCAD/EMTDC for simulation analysis of the feasibility, real-time verification is carried out using the RTDS closed-loop experimental platform. The results show that the proposed inrush current suppression strategy can effectively suppress the inrush current under different residual flux conditions. [ABSTRACT FROM AUTHOR]

Anomaly detection is a common technology for condition monitoring that is important for safe and reliable production in industry. In this work, we proposed an anomaly detection method for rotating machinery based on vibration vectors,whichwas inspiredby Polar plot in vibration analysis. The vibration vector, consisting of amplitude and phase, was selected as monitored indicator at different characteristic frequencies (CFs). The original vibration vectors obtained by fast Fourier transform-based order analysis (FFT-OA) were mapped from Polar to Cartesian plot for simplified calculation and better visualization. The acceptance region defined to distinguish anomalies from normal conditions was learned by support vector data description (SVDD) model with normal vibration samples. We proposed the anomaly score to indicate the degree of deviation from normal condition, and the fit degree to reduce underfitting of data description model. Two datasets, obtained from a real steam turbine and a real steam feed pump respectively, were used to validate the method under constant and variable speed operation conditions. Compared to the amplitude-based methods, the proposed method exhibited an extremely strong ability to detect anomalies in advance. Moreover, SVDD performed better than the minimum covariance determinant (MCD) and isolated forest (iForest) methods in terms of describing the acceptance region in the present work. With low timeconsuming calculation, the acceptance region can be self-evolving easily and has a prospect to the edge computing for machine condition monitoring. [ABSTRACT FROM AUTHOR]

Over the course of both evolution and development, the human musculoskeletal system has been well shaped for the cushion function of the foot during foot-strike and the impulsive function of the ankle joint during push-off. Nevertheless, an efficient energy interaction between foot structure and ankle joint is still lacking in the human body itself, which may limit the further potential of economical walking. Here we showed the metabolic expenditure of walking can be lessened by an unpowered exoskeleton robot that modulates energy among the foot-ankle complex towards a more effective direction. The unpowered exoskeleton recycles negative mechanical energy of the foot that is normally dissipated in heel-strike, retains the stored energy before mid-stance, and then transfers the energy to the ankle joint to assist the push-off. The modulation process of the exoskeleton consumes no input energy, yet reduces the metabolic cost of walking by $8.19~\pm ~0.96$ % (mean ± s.e.m) for healthy subjects. The electromyography measurements demonstrate the activities of target ankle plantarflexors decreased significantly without added effort for the antagonistic muscle, suggesting the exoskeleton enhanced the subjects’ energy efficiency of the foot-ankle complex in a natural manner. Furthermore, the exoskeleton also provides cushion assistance for walking, which leads to significantly decreased activity of the quadriceps muscle during heel-strike. Rather than strengthening the functions of existing biological structures, developing the complementary energy loop that does not exist in the human body itself also shows its potential for gait assistance. [ABSTRACT FROM AUTHOR]

To satisfy the huge storage space and organization capacity requirements in addressing big multimodal data, hashing techniques have been widely employed to learn binary representations in cross-modal retrieval tasks. However, optimizing the hashing objective under the necessary binary constraint is truly a difficult problem. A common strategy is to relax the constraint and perform individual binarizations over the learned real-valued representations. In this paper, in contrast to conventional two-stage methods, we propose to directly learn the binary codes, where the model can be easily optimized by a standard gradient descent optimizer. However, before that, we present a theoretical guarantee of the effectiveness of the multimodal network in preserving the inter- and intra-modal consistencies. Based on this guarantee, a novel multimodal deep binary reconstruction model is proposed, which can be trained to simultaneously model the correlation across modalities and learn the binary hashing codes. To generate binary codes and to avoid the tiny gradient problem, a novel activation function first scales the input activations to suitable scopes and, then, feeds them to the tanh function to build the hashing layer. Such a composite function is named adaptive tanh. Both linear and nonlinear scaling methods are proposed and shown to generate efficient codes after training the network. Extensive ablation studies and comparison experiments are conducted for the image2text and text2image retrieval tasks; the method is found to outperform several state-of-the-art deep-learning methods with respect to different evaluation metrics. [ABSTRACT FROM AUTHOR]

To meet the required huge data analysis, organization, and storage demand, the hashing technique has got a lot of attention as it aims to learn an efficient binary representation from the original high-dimensional data. In this paper, we focus on the unsupervised spectral hashing due to its effective manifold embedding. Existing spectral hashing methods mainly suffer from two problems, i.e., the inefficient spectral candidate and intractable binary constraint for spectral analysis. To overcome these two problems, we propose to employ spectral rotation to seek a better spectral solution and adopt the alternating projection algorithm to settle the complex code constraints, which are therefore named as Spectral Hashing with Spectral Rotation and Alternating Discrete Spectral Hashing, respectively. To enjoy the merits of both methods, the spectral rotation technique is finally combined with the original spectral objective, which aims to simultaneously learn better spectral solution and more efficient discrete codes and is called as Discrete Spectral Hashing. Furthermore, the efficient optimization algorithms are also provided, which just take comparable time complexity to existing hashing methods. To evaluate the proposed three methods, extensive comparison experiments and studies are conducted on four large-scale data sets for the image retrieval task, and the noticeable performance beats several state-of-the-art spectral hashing methods on different evaluation metrics. [ABSTRACT FROM AUTHOR]

A novel concept of acoustic fiber Bragg grating (AFBG) was conceived and experimentally validated as an effective technique for high-temperature sensing. Similar to the well-known optical fiber Bragg grating, the AFBG employs interactions between waves and periodic structures on an elongated waveguide. Acoustic waves reflected from periodic structures interfere with each other and result in the emergence of resonance frequency when acoustic wavelength matches with pitch length of the periodic structures. When the single-mode operation condition is satisfied, the resonance frequency location can be accurately resolved and thus be used as an effective indicator for temperature sensing. To demonstrate this concept, an AFBG sample was fabricated on a telecommunication optical fiber (125- $\mu \text{m}$ diameter) with femtosecond laser micromachining setup, and it demonstrated sensing capability up to 700 °C, which was limited by the fiber material (fused silica) and could be over 1400 °C if other material, such as sapphire, was used as the waveguide. With the features of various applicable material, low system cost, and potential multiplexing capability, this AFBG technology is a promising candidate for high-temperature sensing, even in a distributed manner. [ABSTRACT FROM AUTHOR]

In this letter, a frequency-diverse transmission metamaterial aperture (FDTMA) with a bunching random beam (BRB) is proposed for coincidence imaging. The FDTMA consists of a random modulation metamaterial lens and a broadband microstrip patch antenna array, which is used as a feed source. First, an equivalent medium containing both dielectric layer and metamaterial structure is modeled and analyzed using the electromagnetic theory. The zoom coefficients of the metamaterial elements are analyzed to simplify the design of the broadband aperture. Then, the basic metamaterial elements are designed based on the proposed model. Moreover, the FDTMA is designed using the basic metamaterial elements considering the requirements of the coincidence imaging system. Finally, the waved metal plate leveraged in the FDTMA with a BRB is optimized using the High-Frequency Structure Simulator software. The radiation patterns of the proposed FDTMA with a BRB are randomly distributed in a limited angle range in the X-band. The proposed equivalent model and the FDTMA with a BRB are validated through simulations and experiments. [ABSTRACT FROM AUTHOR]

In this paper, the transport ac losses in a triangular epoxy-impregnated high-temperature superconducting (HTS) coil made from YBCO coated conductor, intended for use in a prototype axial flux HTS electric machine, are measured using two different electrical techniques at 77 K. The first set of ac loss measurements of the coil are carried out at the University of Cambridge using a technique based on a lock-in amplifier. The coil is then measured at the Center for Advanced Power Systems, Florida State University, using a technique based on a high-accuracy data acquisition measurement system. The two different methods show consistent results, validating the accuracy of these two techniques for transport ac loss measurements of superconducting coils. Multiple voltage taps are utilized within the coil to study the details and distribution of the ac loss in different sections of the coil. Losses are also measured with a flux diverter made of ferromagnetic material to analyze its effect on the ac losses. [ABSTRACT FROM PUBLISHER]

The saturated-core-type superconducting fault current limiter (SCSFCL) is an important device to limit increasing fault currents in electrical power grids. However, there have been few numerical simulations carried out to simulate the properties of the superconducting DC coil in the SCSFCL, but such modeling is indispensable to the design and analysis of such devices. In this paper, a simulation method to simulate the SCSFCL with COMSOL Multiphysics is presented, and the AC losses of the DC coil in the normal, nonlimiting state, which is its usual operating state, are simulated and analyzed for the first time. A strategy to simulate the AC loss of the superconducting coil with a three-dimensional (3-D) model is given first. The nonlinear electric and magnetic problems are decoupled first. Next, the model is divided into smaller models for separate calculations. The 3-D model is further simplified into a 2-D model based on the worst case scenario for a conservative design. The AC loss distribution within the coil is given and the relationship between the AC loss and an AC component of the current is analyzed, and it is found that the outermost coil regions have much larger AC loss than the coils near the centre. [ABSTRACT FROM PUBLISHER]