Search

Biomarkers are one of the primary medical signs to facilitate the early detection of Alzheimer’s disease. The small beta-amyloid (Aβ) peptide is an important indicator for the disease. However, current methods to detect Aβ pathology are either invasive (lumbar puncture) or quite costly and not widely available (amyloid PET). Thus a less invasive and cheaper approach is demanded. MRI which has been used widely in preclinical AD has recently shown the capability to predict brain Aβ positivity. This motivates us to develop a method, SDF sparse convolution, taking MRI to predict Aβ positivity. We obtain subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and use our method to discriminate Aβ positivity. Theoretically, we provide analysis towards the understanding of what the network has learned. Empirically, it shows strong performance on par or even better than state of the art.

Shape analysis has been playing an important role in early diagnosis and prognosis of neurodegenerative diseases such as Alzheimer's diseases (AD). However, obtaining effective shape representations remains challenging. This paper proposes to use the Alexandrov polyhedra as surface-based shape signatures for cortical morphometry analysis. Given a closed genus-0 surface, its Alexandrov polyhedron is a convex representation that encodes its intrinsic geometry information. We propose to compute the polyhedra via a novel spherical optimal transport (OT) computation. In our experiments, we observe that the Alexandrov polyhedra of cortical surfaces between pathology-confirmed AD and cognitively unimpaired individuals are significantly different. Moreover, we propose a visualization method by comparing local geometry differences across cortical surfaces. We show that the proposed method is effective in pinpointing regional cortical structural changes impacted by AD.

In recent years, quantum communication project is greatly supported by policies in China. Based on the theoretical models and research experience at home and abroad, starting from the existing stock index futures in China, through VEC model and GARCH model, this article reveals the impact of stock index futures on the corresponding stock index components as well as the quantum communication industry. Then further predicts the influence of the new stock index futures in Shenzhen stock market on quantum communication industry. The corresponding investment opportunities will also be analyzed.

With the improvement of computer technology and the widespread popularization of the Internet in China, e-commerce has greatly changed people's way of life, become a major promoting industry for China's economic development, and also increased the income of rural and urban residents in China to a certain extent. Since 2013, China's e-commerce development has reached a certain scale, so this article based on the system of literature, using data from 2013–2019, analyzes the electronic commerce to the influence of the Chinese people's income, found that in general the development of e-commerce of our country resident income and great role in promoting, and the impact on rural income is greater than the effect on the income in cities and towns. Finally, combined with the results of empirical analysis, put forward the corresponding countermeasures and suggestions.

Three-level T-type neutral-point-clamped converters (3L-TNPC) are widely used in motor drive and PV applications because of the higher efficiency, improved output THD, and lower common-mode noise. But such topology suffers from high stray inductance because of the complicated power loop structure. This paper proposes a SiC-MOSFET based 3L-TNPC power module with a hybrid packaging method. The module has a printed circuit board (PCB) directly soldered on top of direct bounding copper (DBC), with bare dies soldered on DBC and connecting to PCB by bonding wires. Such a structure allows the power loop on both DBC and PCB. This reduces stray inductance through enhanced mutual-inductance cancellation. The introduced PCB can provide extra flexibility in gate-loop optimization, enabling synchronous gate drive between parallel dies. The heat from dies can also be directly dissipated through DBC and the proposed direct-soldered coldplate. A 1.2 kV, 400 A SiC 3L-TNPC module is fabricated with measured loop inductance of 2.5 nH. Faster driving speed by using 0.33Ω gate resistance is enabled to reduce switching loss. Eventually, double-pulse tests and continuous tests are carried out to evaluate the module.

This paper proposes an insulation design and optimization methodology for laminated busbars in more-electric aircraft (MEA) motor drive system. The design assumes that the converter works at around 11-km altitude with decompressed air pressure, which expedites partial discharge (PD) compared to it at sea-level. The trade-off between the insulation margin and power-loop stray inductance is revealed and optimized. By using the method, a busbar for a 450-kVA 3-level T-type neutral-point-clamped (3L-TNPC) converter is designed. Its partial discharge inception voltage (PDIV) is 43% higher than the rated voltage at the air pressure of 0.2 atm and 170% higher at normal air pressure. Busbar stray inductance of 12.0 nH is achieved, lower than published literature of 3-level converters.

Partial discharge (PD) issue of high voltage power module packaging insulation is increasingly critical due to the development of high voltage wide bandgap power modules towards high power density, which leads to smaller packages and higher operating voltages. However, existing standards for high voltage power modules focus on PD measurement under the sinusoidal waveform rather than actual DC or square waveform that poses actual electric stress on packaging insulation, and thus cannot fully represent the PD phenomenon in power electronics systems. To bridge this gap, this paper introduces a PD testing platform trimmed for high voltage power electronics systems, including a high-voltage (10 kV) square wave generator using supercascode structure with off-shelf SiC devices, and a contactless PD monitor using Super High Frequency (SHF) signatures from the PD event. The challenges in the design of this platform include the development of a PD-free high voltage square wave generator, and SHF PD signature separation from EMI noises induced by high dV/dt switching actions. Experiment validation is included in the last section of this paper.

Parasitic capacitance in power loop can impact switching performance of converters. Depending on the locations of those capacitors, they can mitigate switching stress if paralleled to DC-link capacitors, or introduce additional switching loss if paralleled to device output capacitance, Coss, or intensify common-mode EMI noise if paralleled to switching-node stray capacitance. Thus high-accuracy characterization of parasitic capacitance is necessary to estimate the performance of converter. This paper proposes novel method to measure stray capacitance in laminated busbar. Such method is based on a vector network analyzer (VNA). The direct assessment of each node-to-node capacitance is the key advantage compared to the traditional method. It also simplifies measurement process, avoids complicated math operation, and improves measurement accuracy in applications of assessing capacitance of the multiple-node systems, for example, three-level converters. This paper illustrates proposed method by measuring a laminated busbar. Design of customized fixtures is discussed for improving measurement consistency and accuracy. Measurement results are compared to the traditional method and ANSYS Q3D simulation to prove its accuracy.

In recent years, deep learning like stacked autoencoder (SAE) has been widely applied to soft sensor modeling for industrial process quality prediction. However, the original SAE based models cannot extract the quality-relevant features due to its unsupervised feature learning, which may even cause negative impact on the final prediction. To cope with this problem, a novel stacked attention-based autoencoder (SAAE) is proposed. By introducing quality information and attention mechanism to the pre-training process, the quality-relevant features can be obtained and simultaneously their importance can be distinguished. Furthermore, to fully utilize these features, feature fusion strategy is integrated with the SAAE method. Finally, a hydrocracking industrial process is used to validate the effectiveness of the proposed method.

Exploring the partial discharge (PD) characteristics of power modules under square wave high electric fields with short rise and fall time is of great significance for improving the reliability of its packaging insulation. In this paper, a PD detection method based on super high frequency (SHF, 3-30 GHz) technique and signal down-mixing principle is applied to detect the PD under square wave. The power module packaging insulation specimen consisting of metal baseplate, direct bonding copper (DBC), substrate and housing is fabricated. A repetitive square wave generator is built to supply high voltage square waves with short transition time. PD experiments are performed under square wave and sinusoidal wave. Experimental results show that the SHF-down-mixing PD detection system can effectively suppress the electromagnetic interference (EMI) from the square wave and extract PD signals successfully. As the radius of the electrode corner increases, the partial discharge inception voltage (PDIV) values under the sinusoidal wave also gradually increases. The PDIV under square wave does not show a monotonically increasing and is usually higher than that under sinusoidal wave probably because of the effect of space charge accumulation.

Multivariate statistical analysis methods like PCA have been widely utilized for fault diagnosis and quality control. Nevertheless, the traditional PCA based methods have some limitations in dealing with the dynamic data information which extensively exists in modern process industry. This paper developed a time window based 2D-PCA model to strengthen the capability of extracting dynamic features for process monitoring. First, this model uses the time window to construct 2-dimensional data matrices piece by piece to keep as much dynamic information as possible between samples. Second, 2D-PCA is applied to these two-dimensional data samples for dynamic feature learning. Finally, general statistics are calculated for online monitoring to detect abnormal states. Finally, Tennessee Eastman (TE) process is used to verify the effectiveness of the developed 2D-PCA monitoring strategy.

In a data-driven fault detection strategy, not only the data model, but the detection index also influences the detection performance a lot. Deep learning has a flexible and powerful ability for modeling strongly nonlinear processes with good generalization performance. However, lack of statistical interpretations, the detection indices have to resort to nonparametric density estimation methods in most of deep models. Thus, the control limits are severely influenced by a specific set of training data, causing a poor robustness. Simultaneously, the squashed activations in neural networks make the detection margin limited. To solve these problems, a novel sample deviation degree penalty strategy with stacked autoencoder is proposed to regularize the neurons and the reconstruction errors, which makes the neurons in each layer and the reconstruction error of different samples tend to be converged. Thus, the robustness and the detection margin are improved. Experiment data in a multiphase flow facility are used to verify the efficacy of the proposed algorithm.

Parameter Optimal State Transition Algorithm (POSTA) is a global metaheuristic method with adaptive parameter adjustment capabilities. However, its fixed local search candidate solution space easily leads to the computational redundancy and the problems that are difficult to jump out of local optimization. Therefore, a novel state transition algorithm with variable local candidate solution space is proposed in this paper, called VLCSSSTA. First, a dynamic logic of distance control is designed to speed up the search, which adaptively changes the size of the candidate solution space. Then, the larger candidate solutions are randomly generated to enhance the ability to jump out of the local optimum. Statistical research results show that the proposed algorithm has better performance than POSTA and some classic optimization algorithms in five benchmark function experiment. Finally, the proposed method is applied to the operation optimization in residue hydrogenation fractionation (RHF). The experimental results also verify the effectiveness of the proposed method.

Modern industrial processes are often characterized with nonlinearities and dynamics. The long short-term memory (LSTM) based soft sensor models have been used for process quality variable prediction. However, most traditional dynamic models like LSTM cannot handle dynamic data sequences with irregular sampling measurements, which is very common in many industrial processes. In this paper, a novel sampling-interval-attention LSTM (SIA-LSTM) is proposed to solve the irregular sampling problem in industrial process. In SIA-LSTM, an additional attention network is designed to learn sampling-interval relevant weight to adjust the dynamic relationship between two adjacent sampled data. The effectiveness and performance of the proposed method is validated on an industrial hydrocracking process to predict the 90% boiling point of aviation kerosene and heavy naphtha.

High-frequency electric machines have been developed to provide propulsion in the electric aircraft system. However, the partial discharge (PD) issue has been paid significant attention to electric machine's insulation design due to the high temperature and low pressure operating environment, which poses threatens to the reliable operation of electric machine. This study proposes an insulation design procedure to avoid PD for air-core form-wound stator windings of high frequency permanent magnet synchronous machines. The effect of air pressure and temperature on partial discharge inception voltage (PDIV) is investigated by experiments. A three-dimensional stator model is built by the finite element method and used to simultaneously obtain thermal the electric field distribution in the insulation. Besides, a design margin is also provided by considering delamination due to manufacturing process or insulation degradation. The proposed procedure may provide useful references for electric machine insulation design for more electric aircraft propulsion.

The heterogeneity of the sintering transversal thermal state is one of the decisive factors of sinter quality and productivity. However, it is difficult to quantify directly the heterogeneity. In the paper, a computer vision-based online assessment method is proposed for the heterogeneity of the sintering transversal thermal state. The method makes use of the flame front distribution in the real-time collecting image of the sintering machine end section. First, the flame front was extracted using image segmentation methods. Second, the heterogeneity of the sintering transversal thermal state was converted into the spatial heterogeneity of the flame front distribution, by introducing the spatial point patterns. Finally, the sintering transversal heterogeneity index was established to evaluate quantifiably the transversal heterogeneity, by integrating the conventional spatial heterogeneity assessment methods and the features of the flame front. The proposed method performs more accurately, robustly, and comprehensively, in comparison with other existing heterogeneity assessment methods, and the sintering expertise. It was integrated into a sintering condition assessment and quality prediction system in a steelmaking plant. The industrial application of it proves its efficiency and significance for guiding the sintering operation, and improving the sinter quality and productivity.

Learning from the multimodal brain imaging data attracts a large amount of attention in medical image analysis due to the proliferation of multimodal data collection. It is widely accepted that multimodal data can provide complementary information than mining from a single modality. However, unifying the image-based knowledge from the multimodal data is very challenging due to different image signals, resolution, data structure, etc.. In this study, we design a supervised deep model to jointly analyze brain morphometry and functional connectivity on the cortical surface and we name it deep multimodal brain network learning (DMBNL). Two graph-based kernels, i.e., geometry-aware surface kernel (GSK) and topology-aware network kernel (TNK), are proposed for processing the cortical surface morphometry and brain functional network. The vertex features on the cortical surface from GSK is pooled and feed into TNK as its initial regional features. In the end, the graph-level feature is computed for each individual and thus can be applied for classification tasks. We test our model on a large autism imaging dataset. The experimental results prove the effectiveness of our model.

Human visual cortex is organized into several functional re-gions/areas. Identifying these visual areas of the human brain (i.e., V1, V2, V4, etc) is an important topic in neurophysiology and vision science. Retinotopic mapping via functional magnetic resonance imaging (fMRI) provides a noninvasive way of defining the boundaries of the visual areas. It is well known from neurophysiology studies that retino-topic mapping is diffeomorphic within each local area (i.e. locally smooth, differentiable, and invertible). However, due to the low signal-noise ratio of fMRI, the retinotopic maps from fMRI are often not diffeomorphic, making it difficult to delineate the boundaries of visual areas. The purpose of this work is to generate diffeomorphic retinotopic maps and improve the accuracy of the retinotopic atlas from fMRI measurements through the development of a specifically designed registration procedure. Although there are sophisticated existing cortical surface registration methods, most of them cannot fully utilize the features of retinotopic mapping. By considering unique retinotopic mapping features, we form a qua-siconformal geometry-based registration model and solve it with efficient numerical methods. We compare our registration with several popular methods on synthetic data. The results demonstrate that the proposed registration is superior to conventional methods for the registration of retinotopic maps. The application of our method to a real retinotopic mapping dataset also results in much smaller registration errors.

A threshold-free feature in brain network analysis can help circumvent the curse of arbitrary network thresholding for binary network conversions. Here, Persistent Homology is the inspiration for defining a new aggregation cost based on the number of cycles, or for tracking the first Betti number in a nested filtration network within the graph. Our theoretical analysis shows that the proposed aggregated cost of cycles (ACC) is monotonically increasing and thus we define a univariate persistent feature based on the shape of ACC. The proposed statistic has advantages compared to the First Betti Number Plot (BNP1), which only tracks the total number of cycles at each filtration. We show that our method is sensitive to both the topology of modular networks and the difference in the number of cycles in a network. Our method outperforms its counterparts in a synthetic dataset, while in a real-world one it achieves results comparable with the BNP1. Our proposed framework enriches univariate measures for discovering brain network dissimilarities for better categorization of distinct stages in Alzheimer’s Disease (AD).

Alzheimer's disease (AD) is a neurodegenerative disorder with progressive impairment of memory and cognitive functions. Sparse coding (SC) has been demonstrated to be an efficient and effective method for AD diagnosis and prognosis. However, previous SC methods usually focus on the baseline data while ignoring the consistent longitudinal features with strong sparsity pattern along the disease progression. Additionally, SC methods extract sparse features from image patches separately rather than learn with the dictionary atoms across the entire subject. To address these two concerns and comprehensively capture temporal-subject sparse features towards earlier and better discriminability of AD, we propose a novel supervised SC network termed Temporally Adaptive-Dynamic Sparse Network (TADsNet) to uncover the sequential correlation and native subject-level codes from the longitudinal brain images. Our work adaptively updates the sparse codes to impose the temporal regularized correlation and dynamically mine the dictionary atoms to make use of entire subject-level features. Experimental results on ADNI-I cohort validate the superiority of our approach.

Continual learning (CL), the problem of lifelong learning where tasks arrive in sequence, has attracted increasing attention in the computer vision community lately. The goal of CL is to learn new tasks while maintaining the performance on the previously learned tasks. There are two major obstacles for CL of deep neural networks: catastrophic forgetting and limited model capacity. Inspired by the recent breakthroughs in automatically learning good neural network architectures, we develop a nonexpansive AutoML framework for CL termed Regularize, Expand and Compress (REC) to solve the above issues. REC is a unified framework with three highlights: 1) a novel regularized weight consolidation (RWC) algorithm to avoid forgetting, where accessing the data seen in the previously learned tasks is not required; 2) an automatic neural architecture search (AutoML) engine to expand the network to increase model capability; 3) smart compression of the expanded model after a new task is learned to improve the model efficiency. The experimental results on four different image recognition datasets demonstrate the superior performance of the proposed REC over other CL algorithms.

Projection regression is an important tool for process soft sensing in order to eliminate redundant information and obtain proper data features. As most industrial process is intrinsically nonlinear and process variables are collected in random noise environment, it is significant to adopt probabilistic nonlinear latent variable model to carry out dimension reduction for feature extraction before regression modeling. Generative topographic mapping (GTM) is such a probabilistic nonlinear model. However, GTM is an unsupervised method, in which the extracted features may include irrelevant ones with the output information. Thus, it may result in inaccuracy of soft sensor performance. To deal with this problem, a generative topographic mapping regression is developed based on supervised GTM in this paper, which incorporates the output information to guide feature extraction and projection regression. By utilizing the output to jointly generate the latent variables, output-related features can be extracted for output prediction. The effectiveness and flexibility of the proposed method are validated on a numerical example and an industrial process.

Wide bandgap power electronic devices with higher breakdown voltages, faster switching, and lower switching losses than silicon-based power devices will be widely used in high-temperature and high-voltage applications. Corresponding packaging materials are urgently required to have higher thermal conductivity and satisfactory insulating properties in wide bandgap power electronic modules. This study introduces a silicone elastomer (rubber) with micron and nano-sized boron nitride (BN) and silicon carbide (SiC) particles to improve thermal conductivity and partial discharge resistance. The thermal conductivity, polarization and depolarization current, and partial discharge were measured to investigate the effect of mixed micro-nanoparticles on thermal and electrical properties. Experimental results show that the composite material has a higher thermal conductivity than the original silicone elastomer. The conductive current of the composite expresses nonlinearity with the increase of load level. Regarding the trap density distribution, the composite's trap depth is generally reduced. The shallow trap density increases, whereas the deep trap density decreases with ascending load level. The partial discharge inception voltage of the composite is higher than the original silicone elastomer, and it increases as the load level rises, which may be explained by the local electric field decrease resulting from trapped charge diffusion around the high voltage electrode. The multiphase BN/ SiC/ silicone elastomer with proper load level has a higher thermal conductivity and better partial discharge resistance, indicating its advantages as packaging material to ease thermal and electrical stress in wide bandgap power electronic modules.

Nowadays, data-driven soft sensors have been developed to estimate the quality variables which are difficult-to-measure in industrial processes. Feature representation plays a significant role in constructing accurate soft sensors. Recently, deep learning has been introduced for feature representation in process data modeling. However, traditional deep networks cannot capture quality-related features for output prediction. To handle this problem, a nonlinear variable-wise weighted stacked autoencoder (NVW-SAE) is proposed to learn deep quality-related features in this paper. By measuring the nonlinear Kendall correlations of input or feature variables with the quality variable in each autoencoder, a corresponding weighted reconstruction objective function is designed to learn quality-related features layer by layer in NVW-SAE. Finally, the proposed NVW-SAE based soft sensor method is applied to a debutanizer column to estimate the concentration of butane, which shows its effectiveness and superiority.

In order to investigate the electrical tree properties during the voltage rise and the influence of the tree defect on the tree propagation, the cross-linked polyethylene (XLPE) samples with different electrical tree defects which were generated by an AC stress were prepared and the voltage-regulating experiment is carried out. The typical needle plate was used where the tip radius of needle was 5µm and the distance between the needle tip and the sample bottom was about 1.5mm. The regulating voltage rose at a constant speed of 500V/ s for 10 seconds and maintained for 5 minutes. The progress was repeated until the electrical tree grew. After that the voltage was maintained at 30kV for 2 hours and then declined as the same regulation way. It was shown that the electrical tree grew remarkably at a typical branch shape during the voltage rise, while the tree had little change at DC voltage or declining voltage. As the tree defect length increases, the tree increment during the voltage rise decreases but the total tree increment over the experiment would increase. Space charge behavior around the tree tip would be applied to describe the tree characteristics at the rising DC voltage and DC voltage.

The learning neural network (LNN) has shown good performance in prediction and regression fields, because the number of hidden neurons can be adjusted and the structure can be trained to fit any unknown function. However, those redundant hidden neurons may make the structure of LNN complicated, which is likely to increase the computational cost and time consumption. Furthermore, the standard Morris method provides an index named elementary effect (EE) to evaluate each hidden neuron, but standard Morris method cannot work well using the input variables with large fluctuation. To overcome these limitations, an improved screening method based on standard Morris is proposed in this paper. Based on the standard Morris method, the EE index is redefined by expanding its discrete domain and improving sampling strategy with the probability distribution of parameters. Then, the mean and standard deviation of multiple EEs are combined into a comprehensive index for contribution calculation and pruning neurons. Finally, the effectiveness of the proposed method is verified in several cases, in which the pruned LNN model shows good performance with low computational cost.