Search

Aero-turbines usually work in various non-stationary and harsh operating environments. Its blades easily appear early crack faults in long-time operation, and it is difficult to dig out discriminant features of early crack. To solve these problems, in the paper, a novel intelligent approach for early crack diagnosis of turbine blades using three-dimensional blade tip clearance is presented. In order to improve feature learning ability to obtain better generalization ability, the paper firstly develops a novel deep learning method based on deep belief networks (DBNs). Considering the fact that the feature degradation easily occurs in deeper layers because of the change of the distribution in each layer's outputs with the increase of the layers, it is hard to decide which layer to learn features is useful for fault diagnosis. Accordingly, in the pre-training process, the global back-reconstruction (GBR) mechanism is introduced into DBNs to optimize the feature learning ability. The GBR mechanism can be realized between the input layer and hidden layers by “shortcut connection”, and the layer to learn more discriminant features can be determined automatically without prior knowledge. Moreover, due to three-dimensional blade tip clearance (3-DBTC) acquired from three different directions of turbine blades contains much more useful crack failure multiscale information, it is suitable to be used as an input from which to extract multiscale discriminant features. Eventually, in the supervised training, the softmax regression model is employed to classify the health conditions of turbine blades using these sensitive features learned from 3-DBTC. The experimental results show that the proposed method can effectively identify the crack of turbine blades with fairly high diagnostic accuracies and significantly outperform other methods considered in the paper.

We propose a method to search the inverse matrix for a type of special multiplied triangular matrix quickly. The proposed method can be used for the decomposition of Diagonal matrix, Lower triangular matrix and Upper triangular matrix (LDU) of a square matrix, and can effectively reduce the complexity of calculation and improve the operation efficiency. The simulation results verify the convenience and effectiveness of the proposed algorithm.

In recent years, blended learning has become one of the major trends in higher education. However, there are few researches on the evaluation of blended learning. Effective learning evaluation can guide the effective development of learning. This paper proposed a new method of student evaluation to evaluate the effect of blended learning for basketball courses in universities from three aspects: knowledge, skill and attitude. We designed a questionnaire for students who have accepted blended learning to take basketball courses in Tsinghua University by using this method and finally collected 222 valid questionnaires. The results indicate that the evaluation questionnaire of blended learning proposed in this study has good reliability and validity, and the blended learning effect of basketball courses in knowledge, skill and attitude is recognized by students.

In this paper, the trajectory tracking control problem of second-order underactuated system with unknown dynamic and prescribed performance constraint was discussed. Event-triggered sliding mode control combined with adaptive method is investigated to solve the problem. By introducing the constraint transformation, which is used to convert the constrained system into unconstrained one, and prescribed performance function (PPF), the steady and transient performance of closed-loop system are guaranteed. After obtained the unconstrained system, an event-triggered adaptive sliding mode control was proposed. The proposed method is robust to uncertain disturbance and saves computing and communication resources. Then, through the Lyapunov theory, the update laws of control law and adaptive laws were obtained, and also the stability of closed-loop control system was proved. Finally, the simulation results demonstrate the effectiveness of proposed control approach.

The task of named entity recognition is to identify certain types of entities with special meanings from the text. It is a basic task in natural language processing and the foundation of higher-level tasks such as relation extraction, knowledge graph, and question answering system. The correctness of the entity recognition has a huge influence on the effectiveness of the upper layer application.This paper mainly studies the problem of Chinese named entity recognition in the medical field. By extracting the information about the disease in the insurance text and labeling the entity of disease, treatment, and symptom, the data set for entity recognition is established. On the basis of the BILSTM-CRF model, we use different methods to improve the recognition effectiveness of the model. By incorporating word boundary information and adding attention mechanism in the BiLSTM layer, the effectiveness of entity recognition is further improved.

Over the past few years, the volumes of digital information have increased dramatically. Web search engines become indispensable to our daily lives due to their capability of filtering relevant information. The web search engines collect users’ online behavior data which implies their personalized interest, even contains sensitive information. A malicious attacker may benefit from advertising or manipulating political campaigns, which poses a serious threat to public security. In this paper, we propose two new methods to generate dummy queries base on global and personalized queries’ probability. The generated anonymous queries set has greater information entropy which helps to protect user’s intent in web search. Experiment results verify the validity of our methods.

BCI based on machine learning could makes use of the EEG signals to communicate to output under the condition of without the participation of peripheral nerves and muscles. Extracting the essential features of the EEG signals in the presence of artifacts, training the classification algorithms and optimizing the performance of classifier is critical procedure for BCI system. In the realization of BCI, the most important step is the feature extraction and classification of EEG signals. Due to the obvious individual difference and low signal-to-noise ratio of EEG signals, the current feature extraction and classification algorithms have low accuracy. The emergence of deep learning has attracted much attention in many fields. At present, some researchers try to apply deep learning algorithm to the recognition of EEG signals, and obtain good results. Based on convolutional Neural Networks (CNN), this paper studies the application of deep learning in motor imagery task classification by end-to-end deep learning.

This paper mainly analyzes the stability of high-dimensional fractional-order gene regulatory network systems with time delay. Based on the judgment of eigenvalues, this paper proposes sufficient information about the stability of high-dimensional fractional-order gene regulatory network systems with time delay. Conditions, and by analyzing the characteristic equations, the bifurcation conditions of the high-dimensional fractional-order gene regulatory network system with time delay are derived. Finally, the theoretical part of this paper is verified by numerical simulation.

In 5G networks, wide-band mmWave can be used to provide extreme data rates to user equipments (UEs). However, since mmWave is coverage limited, it necessitates dense placement of base stations, which in turn can significantly increase the fiber deployment cost. One solution being considered is to replace fibers with Integrated Access and Backhaul (IAB) network, where a part of the wireless spectrum is used for connecting base stations. In an asymmetric IAB network, standard proportional fair scheduling algorithm (PF) fails to distribute resources among UEs fairly. We propose a new weighted proportional fair (WPF) scheduling algorithm to improve the fairness of UE’s achieved throughput in an IAB network. Also to mitigate congestion in IAB nodes and improve system throughput, we propose an IABaware end-to-end flow control (I-EEFC) algorithm. Through detailed analyses for both symmetric and asymmetric network topologies, we show that our proposed combined scheduling and flow control algorithm (WPF+I-EEFC) improves both fairness and system throughput.

With the rapid development of deep convolutional neural networks, more and more computer vision tasks have been well resolved. These convolutional neural network solutions rely heavily on the performance of the hardware. However, due to privacy issues or the network instability, we need to run convolutional neural networks on embedded platforms. Critical challenges will be raised by limited hardware resources on the embedded platform. In this paper, we design and implement an embedded inference framework to accelerate the inference of the convolutional neural network on the embedded platform. For this, we first analyzed the time-consuming layers in the inference process of the network, and then we design optimization methods for these layers. Also, we design a memory pool specifically for neural networks. Our experimental results show that our embedded inference framework can run a classification model MobileNet in 80ms and a detection model MobileNet-SSD in 155ms on Firefly-RK3399 development board.

In order to promote the study of system dynamics (SD) and multi-agent theory and its wide application in the field of renewable energy, a hybrid modeling method of SD and Agent is proposed in this paper, which is suitable for the multi-energy collaborative management and control of regional cooling, heating and power under the background of the ubiquitous Internet of Things. Based on the detailed analysis of the lack of single modeling paradigm of system dynamics and the necessity and feasibility of mixed modeling of SD and Agent, the basic types of hybrid modeling of SD and Agent for regional multi-energy cooperative management and control are defined, and the application forms, simulation platforms, problems and ideas of hybrid modeling of SD and Agent in regional multi-energy cooperative management are discussed and analyzed, which can deepen the innovative application of SD and Agent hybrid modeling in multiple fields and assist relevant managers in scientific decision-making.

There has been an increasing amount of open educational resources (OERs) on the Internet. However, not so many instructors are familiar with those resources, or can easily use them, which leads to insufficient application of these resources in daily teaching. By analyzing the requirements of physical education courses in universities, this paper categorizes the resources for these courses, and proposes the design objectives, technical architecture and main functions of the resource database for physical education courses. We integrate resources of physical education courses from different online sources into a centralized resource database, and link this database to a learning management system so that college instructors and students can get easily access the resources and take advantage of them in personalized instruction and selflearning. Based on the resources in the database, online and offline integrated blended courses are designed and applied preliminarily in the basketball classes in a university in China. The results indicate that the resource database can effectively help instructors develop blended courses, increase the quality of instruction, and substantially improve learning.

BCI based on machine learning could makes use of the egg signals to communicate to output under the condition of without the participation of peripheral nerves and muscles. Extracting the essential features of the EEG signals in the presence of artifacts, training the classification algorithms and optimalizing the performance of classifier is critical procedure for BCI system. To some extent the BCI system can be treated as a pattern recognition system whose performance depends on both the features extraction and the features classification algorithm employed. Independent component analysis (ICA) can remove the artifact in the electroencephalogram (EEG) signal spontaneous evoked by left and right hand motor imagery and the classifier based on the Support Vector Machine (SVM) algorithm and on the common spatial pattern (CSP) algorithm apply the feature extracted from purified EEG signal to recognize and discriminate distinctive motor imagery pattern.

The deployable composite cylindrical thin-walled hinges have important potentiality in the use of deployment mechanisms of satellite and solar array, but the progressive damage analysis of composite hinges hasn’t been solved, which means the mechanical properties of the composite hinge are overestimated. The analytical and finite element models of composite tape springs are established, and the bending characteristics are analyzed. Theoretical and analytical solutions show good consistency. In addition, the progressive damage model of composite cylindrical thin-walled hinges is established and its folding and torsion characteristics are studied. The peak moment of folding is 4.41% reduced and the steady moment of folding is 16.22% reduced while taking progressive damage into consideration. In addition, the hinge has peak torsional moment of 18.48050 Nm and torsional angle of 15.37 degree correspond with Peak torsional moment.

In today's modern society, residential dwellings make up a large proportion of the world's energy consumption. Therefore, predicting dwellings' energy consumption and needs are highly useful for energy conservation and evaluating future demand. Data-driven energy consumption prediction models are widely researched to forecast energy consumption of buildings. However, in the majority of studies, the training data is collected through smart energy meters and sensors of some target sample buildings. Thus, the source of data intrinsically restricts the usable range of prediction models. In this paper, we build a set of innovative partitioned parallel prediction models include Support Vector Regression (SVR) and Random Forest Regression (RFR) to predict annual heating costs, hot water costs and lighting costs of dwellings. The model training and valuation procedures are basing on the data of the Standard Assessment Procedure and Energy Performance Certificate (SAP-EPC), covering all 13 main regions in the United Kingdom. The feature selection comparative experiments prove that the merge feature selector can select valid feature sets according to different prediction targets The comparative experiment results manifest the fact that the parallel modeling approach can improve the model's accuracy. Meanwhile, the experimental results indicate that the proposed prediction models can accurately forecast the energy consumption of the dwellings.

An efficient calculation unit is developed on the basis of programmable neuromorphic circuit for implementing arbitrary mathematical functions. To retrieve any specific function approximately, the regression algorithm through neural network (NN) is realized by a compact analog circuitry. The memcapacitor technology is associated to Neuron-MOS structure, which couples multiple memcapacitors on the floating gate of a MOS transistor to achieve multiply-accumulation (MAC) operation. In this manner, each synapse of NN is emulated by only one memcapacitor device and the weight is post-fabrication programmable due to the memcapacitive characteristics of memcapacitor. Two types of neurons including sigmoid and linear are implemented by differential pairs with Neuron-MOS transistors. For proof-of-concept, the memcapacitor-based approximate calculation unit (MC-ACU) for arbitrary two-operand computations is achieved with 461 devices. From the real circuit simulation results, all the example functions are retrieved with the maximum inaccuracy of 7.8%. The energy- and device-count of MC-ACU are only 36.7% and 15.7% of that of state-of-the-art works in approximate computing, respectively.

There is growing concern about the energy usage of machining processes. Analysis of the specific energy consumption (SEC) makes contribution to this trend of green manufacturing. In this paper, several key aspects of the SEC are analyzed in detail. First, computing models of MRR for turning and milling processes are proposed. Then, every component of SEC and the interaction of STE and SUE is considered. Furthermore, the limitations and shortcomings in application are illustrated. Finally, the change tendency of SEC when MRR increases without limit is investigated through the dissection of the two coefficients of $C_{\theta}$ and $C_{1}$, which are later verified theoretically in the turning processe.

The increasing switching frequency of power converters always leads to severer radiated electromagnetic interference (EMI), especially with the boom of wide bandgap (WBG) devices. Therefore, the development of radiation model and the analysis of the influence of WBG devices on the radiated EMI is important. This paper proposed a technique of developing the radiation model and compared the radiated EMI generated by converters with silicon MOSFETs and GaN HEMTs. Experiments are carried out to verify the analysis.

This paper explores the course design and student satisfaction with a Chinese university basketball course offered in blended mode through the support of mobile technology. There are four parts to this study: (1) the design of a basketball course using the blended learning approach supported by flipped classrooms and a mobile app; (2) a 3-week trial of this approach as a pilot study in the spring semester of 2017; (3) informed by this pilot study, a semester-long implementation of the approach in the autumn semester of 2017; (4) an evaluation of students' satisfaction with this blended learning approach through the use of surveys. The results indicate that this approach to course design and delivery effectively facilitated student learning. The survey data further confirmed the students' satisfaction with this new learning and teaching approach as well as the course, as they believed that this approach had improved their learning quality, in-class interaction and learning outcomes in physical education.

This paper presents a particle-swarm-optimized 2D Otsu segmentation algorithm based on fractional order differential and cloud model (hereinafter called the 'FOCPSO') to solve the low efficiency of the traditional 2D Otsu segmentation algorithm. The proposed algorithm uses 2D Otsu as fitness function, and adaptively changes the fractional order according to the state of particle swarm. It updates the speed and position values by obtaining the ?-order derivative at the current moment. Based on the fitness value, the particles are divided into different populations. Experiments show that the proposed segmentation process improves the convergence speed of the algorithm under the premise of guaranteeing the segmentation effect.

The difference of the radiated EMI modeling between the non-isolated and isolated dc-dc power converters is discussed. Detailed modeling process and parameter extraction method for both the cables and the non-isolated power converter are provided. A boost converter is used in experiment for demonstration.

Deep convolutional neural network (DCNN) based computer vision methods have great progress in object detection tasks. And object detection is essential to many applications such as autonomous robots and vehicles. However, it is difficult for object detection system to be deployed on embedded platforms due to its intensive computing architecture. We present an object detector MobileNet-SSD which can be deployed on embedded platforms. The object detector presented in this work based on Single Shot Detector (SSD) framework and replace the feature extractor with a more light weight network MobileNetV1. General matrix to matrix multiplication is used to simplify the calculation of convolution operation and is further optimized using ARM NEON technology. Finally, the object detector MobileNet-SSD is deployed on embedded platform and achieves 1.13FPS with 0.72 mean average precision on PASCAL VOC dataset.

Power converters should comply with the radiated emission regulation standards. For isolated power converters with long input / output cables, radiated EMI is mostly caused by the CM currents. Accurate and repeatable measurement for the radiated CM currents, impedance and voltages remains a fundamental challenge to the EMI/EMC research in power electronics. In this paper, two measurement issues in the CM current and CM impedance measurement are identified. They are the coupling between the equipment under test (EUT) and the measurement equipment, and the influence of the varying utility grid impedance. Based on the radiated CM current model, this paper proposes improved measurement techniques. By increasing the CM impedances of the measurement equipment cables and the cable connecting the utility grid, the undesired coupling can be eliminated and the utility grid impedance can be isolated. Furthermore, the issues in the CM voltage measurement are identified: low resolution for the low voltage component in the radiated frequency range due to the high magnitude fundamental frequency voltage component; and the small input impedance of the high pass filter which can bypass the radiated emission of the EUT. This paper proposes an improved measurement technique to improve the resolution of voltage measurement in the radiated frequency range and to isolate the input impedance of the high pass filter. Effectiveness and accuracy of the proposed measurement techniques are verified through experiments performed on a flyback converter with input and output long cables.

In this paper, we describe a model-free reinforcement learning method for gait controlling of humanoid robots, which combines Q-learning with Radial Basis Function Network. With the help of RBF Network, this method can solve the approximation problem caused by continuous state space and action space. The approach is applied to the controllers on hip joints of humanoid robots that receives sensory data and constantly adjusts the outputs of steering engines on hip joints, finding an optimal policy that can guide humanoid robots to walk stably on different uneven terrains. We have tested the approach on Webots, a simulation platform, and experiment results have proven the validity of the proposed method.

A novel MEMS ring vibrating gyroscope with S springs is proposed, and the working principle, vibrating characteristics and sensitive working mode are analyzed, according to the working mechanism and vibrating characteristics of axially symmetric ring vibrating gyroscope. On this basis, the finite element model of the ring vibrating gyroscope is established by using ANSYS. Furthermore, the modal analysis, harmonic response analysis, impact transient analysis and static analysis is performed. The results of simulation and analysis show that the natural frequency difference between the driving and sensing mode is 1Hz and the minimum frequency difference between the working mode and other vibrating modes is 3304Hz, which can effectively resist the interference of environmental vibration. The maximum stress of the resonant structure is 104.46MPa under the transient influence of 10 000g, which can work stably.

Humanoid robot has always been a popular field in robotics. Being able to walk stably in complex environments makes humanoid robots more advantageous. Stair is a common walking environment in human life, and the ability of climbing stairs is one of the basic function of humanoid robot. In this paper, the gait planning of humanoid robot climbing stairs is discussed. We use multi-objective genetic algorithm to optimize the key parameters of robot climbing stairs and simulation experiments are performed in Gazebo. The ideal gait parameters are obtained by genetic algorithm, so that the robot can walk stably and quickly.

A power transformer fault diagnosis method based on Improved Particle Swarm Optimization and BP neural network is proposed. The particle swarm algorithm that used to optimize the parameters of the BP neural network is prone to “premature”. By optimizing the inertia weight, in the process of increasing the number of iterations, the inertia weight can be gradually reduced, and the algorithm can avoid the partial optimization. By introducing momentum terms, the weights adjusting method of BP neural network is optimized. The training speed of BP neural network is speeded up and the training accuracy is improved. Simulation results show that the improved PSO algorithm can optimize the BP neural network effectively in transformer fault diagnosis, and improve the efficiency of diagnosis.

For the reason that the algorithm PFIM needs to scan database repeatedly, to produce a great deal of redundant candidate itemset, and to compute more time-complexity of frequent probability, an improved algorithm UPro-Eclat which is based on PFIM and Eclat is proposed. It uses a vertical mining method which is extension-based, adds probabilistic information in Tid, builds recursively the subset of search tree, and mines probabilistic frequent pattern by depth-first traversa. The algorithm UPro-Eclat can swiftly find probabilistic frequent itemset rather than compute their probability in each possible world.

The high switching frequencies of power converters always lead to severe radiated EMI. The modeling and analysis of the converter with undesired antenna such as input/output cables are very important for the research on radiation. This paper first develops the model for undesired antenna in converters and discusses its relationship with the radiated power. Then the antenna's interaction with converter source impedances is taken into account and is explored in detail. Finally, the analysis is applied to a specific converter and verified by simulations and experiments.

For nonlinear estimation, the Gaussian sum filter (GSF) provides a flexible and effective framework. It approximates the posterior probability density function (pdf) by a Gaussian mixture in which each Gaussian component is obtained using a linear minimum mean squared error (LMMSE) estimator. However, for a highly nonlinear problem with large measurement noise, the estimation performance of the LMMSE estimator is largely limited, since it is the best only within the class of linear estimators. This may further degrade the performance of the GSF, especially if a small number of these components are used. To improve the estimation performance, this paper proposes a Gaussian sum uncorrelated conversion (UC) based filter (GS-UCF), where the recently proposed uncorrelated conversion based filter (UCF) is applied to obtain the Gaussian components for Gaussian sum filtering. The UCF which is the LMMSE estimator using the measurement augmented by its uncorrelated conversions can outperform the original LMMSE estimator. Thus, the first two moments of the Gaussian component obtained by UCF can be more accurate than those obtained by the LMMSE estimator, which further improves the performance of the GSF. As an integration of the UCF and the GSF framework, the obtained filter is named as the Gaussian sum uncorrelated conversion based filter (GS-UCF). Simulation results show the effectiveness of the proposed estimator.

In this paper, we present the design of Chinese Speech Robot Control System on Android Embedded Platform, in order to improve the application of service robot in the Chinese environment. With using this software system, we can have the intelligent dialogue with users in daily life by a wireless network based on Android Programming and ROS (Robot Operating system). Furthermore, after collecting user speech message by the Embedded System, the Android client can achieve Semantic Understanding, Voice Wakeup, Speech Synthesis and other services functions. Forward the speech message to robot ROS network through a wireless network, users can control the robot motion by inputting different speech commands. Experiments show that the user not only can communicate with the robot through a good human-computer interaction, in addition to the robot direct dialogue but also can have good performance in the room through the mobile side to the robot emergency motion control.

The performance of a frequency division duplexing (FDD) massive multiple input multiple output (MIMO) system is traditionally limited by the large amount of overhead for downlink channel training and uplink channel state information (CSI) feedback. In this paper, we propose an environment-aware scheme to exploit massive MIMO gains in the FDD mode. Under a quasi-static scattering geometry and slow user mobility, the propagation environment can be known at a low cost. Given a priori environment information, the angular domain channel statistics can be obtained accordingly. In the proposed scheme, the angular domain is partitioned into several angular bins and the same number of predefined precoding vectors are generated accordingly. Based on the environment-specific angular domain information, the system topology is modeled as a bipartite graph. An efficient user scheduling algorithm is proposed, which is equivalent to finding a match of the bipartite graph, and a remarkable multiplexing gain is achieved by removing the overlapped angular bins. After user scheduling, each selected user is allocated to one predefined precoding vector. The numerical results have confirmed the validity of the proposed scheme.