Search

Deep learning (DL) accelerators have been widely deployed in a broad range of application domains. Resilience against hardware failures is a top priority as hardware failures can lead to various undesirable consequences. For inference workloads, hardware failures can generate crashes and significant degradation in inference accuracy. For training workloads, hardware failures can result in failure to converge, low training/test accuracy, numerical errors, and crashes. In this thesis, we establish the new knowledge on how major classes of hardware failures propagate and affect inference and training workloads, and advance state-of-art resilient deep learning systems by devising lightweight, effective detection and recovery techniques to mitigate hardware failures. There are three main contributions in this thesis. First, we devise a novel technique for generating high-quality test programs to detect permanent hardware failures in the field. For compute hardware units, we first adopt the ATPG to generate high-quality test patterns, and then reverse-engineer the dataflow/reuse algorithm of the accelerator to map the test inputs to equivalent DNNs. For control hardware units, our key observation is that typically only one or a few fixed DNNs are deployed at a time, which allows us to target only the hardware failures that can directly affect these DNNs by executing different layers of the DNNs using carefully-crafted weight and input tensors. We demonstrate that our technique successfully detects > 99.9\% of stuck-at faults and > 99.0\% of transition faults, compared to only $< 80\%$ if random test programs are used. Second, we provide (1) an open-source resilience analysis framework targeting transient hardware failures, called FIdelity, as well as (2) the first in-depth study, using this framework, on the impacts of transient failures on DL inference accelerator through fault injection experiments. FIdelity enables accurate and quick resilience analysis by modeling hardware failures in software with high fidelity. This is achieved by taking account of both the spatial and temporal reuse effects of hardware signals to map the effects of a given faulty hardware signal to a set of faulty output neurons in the current DNN layer. Using FIdelity, we perform 46M fault injection experiments running various representative inference workloads. We thoroughly analyze the experiment results, and obtain new insights that can be used in designing efficient, resilient DL inference accelerators. Lastly, we focus on DL training workloads, and provide (1) the first study that reveals the fundamental understanding on how transient and permanent hardware failures affect DL training workloads, and (2) new, light-weight hardware failure mitigation techniques. We extend the FIdelity framework to perform large-scale fault injection experiments targeting DL training workloads, and conduct $>2.9M$ experiments using a diverse set of DL training workloads. We characterize the outcomes of these experiments and derive the necessary conditions that must be satisfied for hardware failures to cause unexpected training outcomes. Based on the necessary conditions, we develop ultra-lightweight software techniques to detect hardware failures and recover the workloads, which only require $24$-$32$ lines of code change, and introduce $0.003-0.025\%$ performance overhead for various representative neural networks.

Objective: To establish a structured and integrated platform of clinical data and biobank data, and a client to retrieve these data. Study Design: Initially, the hospital information system (HIS) a...

Recent years have witnessed the increasing attentions on human action recognition(HAR). Traditional methods are prone to explore the optimum spatiotemporal feature representation of human actions in video clips so as to achieve high recognition performance. However, the optical limitations, such as inappropriate view, dim illumination and object occlusion, usually degrade video quality and affect the recognition performance a lot. Considering that wireless signals are robust against optical limitations, we thus incorporate WiFi signals with video streams for HAR. Specifically, we use WiFi Channel State Information as a compensator for video streams. A great challenge is how to effectively fuse the video and WiFi information to achieve better prediction performance. To this end, we employ convolution neural networks and statistic analysis algorithms to extract video and WiFi features respectively, and propose a novel multi-modal learning approach for video and WiFi feature fusion, where the video and WiFi features are projected to a common space by supervised learning. The experimental results indicate that the recognition precision of human actions in videos improved obviously with the aid of WiFi signals and the proposed multi-modal learning approach rivals the state of art methods.

The joint signatures of binary-state and multi-state (semi-coherent or mixed) systems with i.i.d. (independent and identically distributed) binary-state components are considered in this work. For the comparison of pairs of binary-state systems of different sizes, transformation formulas of their joint signatures are derived by using the concept of equivalent systems and a generalized triangle rule for order statistics. Similarly, for facilitating the comparison of pairs of multi-state systems of different sizes, transformation formulas of their multi-state joint signatures are also derived. Some examples are finally presented to illustrate and to verify the theoretical results established here.

This paper proposes a wind-photovoltaic-thermal energy storage hybrid power system with an electric heater, which adopts the idea of concentrated solar power plant but omits the expensive solar field, and utilizes the reformed power block from conventional small-scale thermal power plant. From the perspective of thermal power plant reconstruction, the hybrid system aims to minimize the levelized cost of energy and maximize the utilization rate of transmission channels. The capacity optimization is solved by multi-objective particle swarm optimization algorithm. Final optimal capacity is obtained by the Pareto front and decision-making method. Moreover, the probability of exceeding is introduced to correct the theoretical power output calculated by mathematical models and improve the practicality of simulation. Finally, by comparing wind-photovoltaic-thermal energy storage system, wind-photovoltaic-battery system and wind-photovoltaic system, it can be concluded that the proposed system effectively increases the utilization rate of transmission channels and presents a better reliability and economy performance.

Nested array has aroused remarkable attention due to the capability to obtain both enhanced degrees of freedom (DOFs) and enlarged array aperture, whereas only difference co-array is employed. Here, the authors construct the sum co-array of nested array by utilising the non-circular (NC) characteristic and, subsequently, the sum-difference (SD) co-array is obtained by careful crafting of the sum co-array and the typical difference co-array, which doubles the resulting co-array aperture and, simultaneously, further increases the achievable DOF compared with the difference co-array of nested array. Moreover, to circumvent the time-consuming two-dimensional spectral search, the authors proposed the reduced-dimension multiple signals classification (MUSIC) algorithm with the SD (SD-RD-MUSIC) co-array of the nested array by discarding the NC phase. Resultantly, the proposed SD-RD-MUSIC co-array algorithm can resolve more signals and obtain more accurate direction of arrival estimates. In addition, the effectiveness and superiority of the proposed method are validated by numerical simulations.

Redundant systems take an important role in the field of reliability. Most of related researches have considered component redundancy, however, time redundancy which is also very common in practice hasn’t been discussed enough. In this paper, we assume that the time needed for a mission cannot be negligible and the mission must be completed in a given mission window which is long enough for several mission repeated tries. Based on this assumption, the main model is investigated together with its three special cases by the Laplace transform method, and their properties and relationships are discussed briefly. An extended model with different failure modes is taken into consideration by using the Monte Carlo method, and some extended discussions are given to compare the Laplace transform method with the Monte Carlo simulation method. The optimal analysis for cost and revenue are also presented for the main model. Then numerical examples are shown to illustrate results obtained in this paper. Finally, conclusions and discussions are presented.

Tobacco loosening and conditioning process is one of the most important processes in cigarette production, the quality of which affects the final product quality largely. The control of water proportion added in this process is human experience-centric and inconsistent without intensive human tuning. This paper proposes an uncertainty-aware outlet moisture prediction and optimization approach to provide updated water proportion to the PID controller for consistently desirable quality performance and stability. In this research, quantile regression is applied to describe the confidence of prediction, and Monte Carlo simulation is induced in the online optimization to deal with the uncertainty. The approach has been applied to a cigarette production line and the resulted performance showed significant improvement compared with the results of human tuning.

In this paper, we provide a novel idea to achieve underdetermined direction of arrival (DOA) estimation for sparse array by introducing covariance matrix reconstruction into sparsity-based algorithms. Based on the Toeplitz property of covariance matrix of uniform linear array, the covariance vector sparsely representation (CVSR) model is established and DOAs can be retrieved through sparse reconstruction. Moreover, Toeplitz covariance matrix estimation can be transformed as convex optimization problem with less constraints and solved by proposed original covariance matrix reconstruction (OCMR). Finally, CVSR achieve more accurate DOA estimation based on accurate Toeplitz covariance matrix estimation by OCMR. By implementing simulation experiments, we demonstrate that CVSR can outperform existing sparsity-based DOA estimators in terms of estimation accuracy and efficiency which also demonstrates the effectiveness of the novel idea about combinations of covariance matrix reconstruction and sparsity-based algorithms.

To alleviate the dependence on sensor quality and to reduce the accumulated error in traditional inertial navigation systems, this paper proposes RUPT: a millimeter-wave radar aided pedestrian dead reckoning system with dual foot-mounted inertial measurement units (IMU). RUPT in this paper is a comprehensive data processing procedure which pre-processes both inertial data and millimeter-wave data and fuses them in a complementary way. Extensive experiments have demonstrated that the accuracy of RUPT has been improved by up to 65% over the conventional dual-foot mounted pedestrian tracking system.

In this paper, a novel rectangular sparse array with hole-free difference co-arrays and low coupling effect is proposed, while a sparse reconstruction algorithm of partial grid is also proposed for 2-D DOA estimation. Firstly, the nested Toeplitz characteristic of covariance matrix of rectangular uniform array is analysed and the nested Toeplitz covariance matrix is estimated. Based on the precisely estimated covariance matrix, we establish the sparse representation model of covariance vector and achieve 2-D DOA estimation by proposed partial grid covariance vector sparse reconstruction (PGCVSR). Simulation results demonstrate that our proposed algorithm can achieve superior 2-D DOA estimation performance and high estimation accuracy.

To achieve underdetermined 2-D DOA estimation of signals in different band, a novel integrated sparse array based on four rectangular sparse subarrays with reduced coupling effect and a 2-D DOA estimator via matrix completion are proposed. Firstly, the concept of original rectangular uniform array (ORUA) is proposed and the nested Toeplitz characteristic of covariance matrix is analysed. Then nested Toeplitz covariance matrix is estimated quickly by augmented Lagrange multiplier. Based on the precisely estimated covariance matrix, we achieve 2-D DOA estimation with unitary-ESPRIT. Simulation results demonstrate that our proposed algorithm can achieve superior 2-D DOA estimation performance and high estimation accuracy.

In the optical transport network, the high-dimension equiprobable constellation is used to increase the information transmission rate and the forward error correction code is used to improve the system performance. However, adopting the equiprobable constellation will lead to some information loss according to the Shannon theory and the performance of the FEC code needs to be strengthened. Based on this, unequiprobable constellation is designed to reduce the information loss by optimizing the probability of the first bit taking the value zero or one. In order to further improve the system performance, polar code, an advanced channel code with good performance and low decoding computation complexity, is combined with the probabilistic shaping to realize the unequiprobable constellation in this paper. The simulation results show that the unequiprobable constellation can provide nearly 1dB gain compared to the equiprobable constellation.

Systems with self-healing mechanism have been successfully applied in many practical fields. However, the self-healing action has less been considered when discussing the reliability performance of systems subject to degradation or shocks. In this paper, we study a damage self-healing system subject to random shocks where the shock arrival times are described by a Poisson process with different intensities due to the accumulated system damage. When the system is majorly damaged, i.e., the state of the system is worse than a specified state, it loses the ability of self-healing. By using the discrete state method, the evolution of the system state can be modeled by a Markov renewal process. Then the system performance indexes, for example, the system reliability, the remaining lifetime distribution, the expected number of shocks and the availabilities are derived in the analytical forms. Finally, a numerical example is illustrated to demonstrate the framework we derived in this paper and show the influence of self-healing mechanism.

In the present paper, a discrete-time semi-Markovian system is studied for both theoretical and practical demands. The state space of a semi-Markov chain is divided into three subsets, namely working state subset, changeable state subset and failure state subset, resulting from applications for this kind of models in degradation analysis on system performances. Performances of changeable states depend on recent evolutions of the system, including the immediate history and future. The Z-transform is used to investigate stochastic properties of the semi-Markovian system and distributions of some sojourn time lengths, based on which, for example, the working period length and failure period length, are discussed. Reliability measures such as reliability, point availability and interval availability are derived for the system. Besides, two new reliability measures, point covering availability and interval covering availability, are discussed for discrete-time semi-Markovian systems for the first time. Preventive maintenance and corrective maintenance from outside of the system are also investigated with respect to related costs. Finally, a numerical example is given to illustrate the results obtained in the paper. Those models and results presented in the paper can be used in reliability, maintainability, supply chain management and other areas.

System signature is a crucial measure in the description and comparison of system structures in the field of reliability, and it does not depend on the component distributions under the assumption that all components are independent and identically distributed. A Markov process method is proposed to compute signatures for general coherent systems in this paper, which is more efficient for the computation of the system signature. As applications of the method, signature formulas of consecutive‐k‐out‐of‐n: F systems and related consecutive‐type systems are investigated, including both the linear and circular ones. Particularly, the related signature formulas for consecutive‐type systems with sparse d are derived for the first time. Illustrative examples are presented for methane reformer furnace systems.

A novel damage identification method for concrete continuous girder bridges based on spatially-distributed long-gauge strain sensing is presented in this paper. First, the variation regularity of the long-gauge strain influence line of continuous girder bridges which changes with the location of vehicles on the bridge is studied. According to this variation regularity, a calculation method for the distribution regularity of the area of long-gauge strain history is investigated. Second, a numerical simulation of damage identification based on the distribution regularity of the area of long-gauge strain history is conducted, and the results indicate that this method is effective for identifying damage and is not affected by the speed, axle number and weight of vehicles. Finally, a real bridge test on a highway is conducted, and the experimental results also show that this method is very effective for identifying damage in continuous girder bridges, and the local element stiffness distribution regularity can be revealed at the same time. This identified information is useful for maintaining of continuous girder bridges on highways.

A concept of Sequential Probability Series System (SPSS) is developed in this paper, which widely exists in many practical sectors such as power plants, inventory management and security management. In SPSS the failure states of each unit are divided into two classes according to their consequences: dangerous failure and safe failure, where the former results in system failure while the latter has no impact on the system. Suppose that when a failure unit appears in SPSS, the system fails with probability p while the other units in SPSS can continue working with probability 1 − p . This paper treats the problem of achieving optimal allocation of units in SPSSs that maximizes expected total working time of all units. Three optimal allocation models are formulated. We derive the analytical expressions for the optimal allocation solutions under certain assumptions. A genetic algorithm and a Monte Carlo method are provided to solve the allocation problems whose analytical solutions are difficult to obtain. An application can be found in Remote Power Feeding System (RPFS). Numerical examples for a RPFS are presented to demonstrate the application of the developed approach in each model.

Research on reliability indices has received considerable attention recently in the field of reliability. In this work, two new reliability indices — multi-point bounded-covering availability and multi-interval bounded-covering availability — are introduced for discrete-time systems. Their explicit formulas are derived for discrete-time first- and second-order aggregated semi-Markov systems, whose state spaces are divided into three subsets corresponding to perfect functioning states, imperfect functioning states and failure states, respectively. Finally, some numerical examples are presented to illustrate the developed results. The results established here will also be useful for many practical reliability systems.

Grid connection of distributed generator (DG) is highly increasing in power distribution system. Network energy loss (NEL) reducing for multiple DGs is very significant. To minimize the NEL, we propose a methodology based on the particle swarm optimization (PSO) initialized with K-means cluster to dynamically allocate the penetration for multiple DGs. The random initialization of particles in the PSO can lead to a premature convergence to a local solution. In order to avoid this problem, we employ the K-means clustering method to initialize particle with the designed formulas. We divide the nodes into the clusters with various NEL objects by K-means clustering method. The NEL objects include annual NEL, probabilistic annual NEL and hourly NEL. Based on the statistical probability of hourly NEL cluster, we obtain the time-varying cluster. Since nodes have similar characteristic in the same cluster, cluster with less centre distance of NEL and the node with less NEL are allocated more penetration by the designed formulas. We utilize this penetration to initialize the particle in PSO algorithm. Because the variation of power generated from DG and load demand can impact on NEL, we utilize the PSO based on the K-means clustering initialization to allocate the penetration at each hour with the historical data. The PSO algorithm and proposed method are applied to a 38 nodes power distribution system. We carry on the simulation in various types of time period. We also present load growth to compare the proposed method with PSO algorithm. The results indicate that the proposed allocation strategy performs better than PSO in reducing NEL.

Video feature extraction is the basis of research on video data. At present, many models based on convolution neural networks for features extracting need to stack deeper network layers to improve the robustness, which means that the amount of parameters and computation is greatly increased while the practicability of such models is constrained. In this paper, a global projection self-attention mechanism is proposed to empower shallow models generate robust video representations. Compared to Non-local [17] and Squeeze Excitation [18] networks, we adopt more succinct but sufficient context modeling strategy to establish complete global interaction space, introduce a lighter transform block for attention coefficients predicting, which improves the running efficiency of attention models greatly. We further embed our attention module to several modern networks including ResNet[3], $\mathbf{R}(\mathbf{2+1})\mathbf{D}[\mathbf{9}]$ and I3D[7], in that ResNet-18 and $\mathbf{R}(\mathbf{2+1})\mathbf{D}-\mathbf{18}$ meet considerably promotion upon its original version by more than 18.9 percent on UCF101 dataset, this result implies huge potential of proposed attention blocks.

With the explosive growth of video data in real applications, near-duplicate video retrieval (NDVR) has become indispensable and challenging, especially for short videos. However, all existing NDVR datasets are introduced for long videos. Furthermore, most of them are small-scale and lack of diversity due to the high cost of collecting and labeling near-duplicate videos. In this paper, we introduce a large-scale short video dataset, called SVD, for the NDVR task. SVD contains over 500,000 short videos and over 30,000 labeled videos of near-duplicates. We use multiple video mining techniques to construct positive/negative pairs. Furthermore, we design temporal and spatial transformations to mimic user-attack behavior in real applications for constructing more difficult variants of SVD. Experiments show that existing state-of-the-art NDVR methods, including real-value based and hashing based methods, fail to achieve satisfactory performance on this challenging dataset. The release of SVD dataset will foster research and system engineering in the NDVR area. The SVD dataset is available at https://svdbase.github.io.

The traditional Region Growing (RG) algorithm is a semi-automatic image segmentation algorithm that requires manual selection of seed points, manual setting of thresholds, which may cause the cavity and over-segmentation while deals with the uneven and undivided image for noise and grayscale. Therefore, the classical Otsu method is adopted to find the optimal threshold in HSV space to obtain the initial binarized segmentation image. Then, the Adaptive Region Growing (ARG) algorithm is applied to find the image edge. To be specific, the initial seed point is confirmed automatically through histogram firstly. Secondly, the threshold of growth condition is determined according to the average similarity of image. Finally, the image is grown into the contour edges. This image is mixed together with the initial Otsu segmentation image can obtain the final processed image. Experimental results show that the proposed algorithm has strong anti-interference, which can effectively reduce the mis-segmentation rate.

The development of machine tool is toward high-speed, high-precision and smarter with high-end compensation techniques. The sensor and the data acquiring system are required in compensation system. However, the aliasing phenomenon can be observed in the existing data acquiring system in machine tool application. In this paper, one solution is proposed to overcome signal aliasing. The phenomenon which appears in spindle thermal growth compensation system is carried, and the improved system. is validated.

Due to the fluctuation and randomness of renewable resources, such as solar irradiation resource and wind resource, independent renewable power plants are not easy to generate stable and reliable power. However, multi-energy complementary power generation with energy storage can improve the power quality of renewable energy generation and meet the requirements of grid-connected, so it will be the mainstream of renewable energy generation in the future. Capacity optimization of multi-energy complementary system is the basis and key to improve the power quality and reduce cost of renewable power generation. This paper describes the capacity optimization model of concentrated solar power-photovoltaic-wind (CSP-PV-Wind) combined power generation system. The optimization objectives are as follows: (1) the power is as close to the load as possible; (2) the low overall investment of the combined power supply; (3) the high annual total power generation revenue. The improved particle swarm optimization algorithm is used to optimize the capacity configuration of CSP-PV-Wind combined power generation system, and obtain the optimal dispatch strategy. The results show that, power quality of CSP-PV-Wind combined power generation system is obviously better than that of PV-wind combined power generation system, while Surplus of Power Supply Probability (SPSP) and Loss of Power Supply Probability (LPSP) are all below 15%. However the power generation cost is still a little higher. Therefore, the strategy of reducing the area of collector and increasing the storage tank capacity will be used to decrease the generation cost in the future.

Recently, most of the community discovery algorithms are based on the structural information of undirected networks, and the social characteristics of users are less considered. Based on the academic social network, we propose a label propagation algorithm that integrates the network structure and multi-dimensional user information (LPA-NU). Through the fusion of multi-dimensional social networks, the algorithm firstly uses the LDA model to mine the similarity of user research directions to derive the hidden social edges between users. Secondly, it constructs a comprehensive directed weighted network, and then classifies the community according to the initial sub-group information. In order to evaluate the quality of community discovery, this paper proposes the definition of overlapping modules of directed networks. We conduct relevant experiments on real social network datasets (SCHOLAT). Experiments show that the LPA-NU algorithm can better divide the structure of the community, and the quality of community division is higher.

In this paper, the survival signature is introduced in a matrix form for a multi-state coherent system with multi-state components and then some related properties are discussed. Using the finite Markov chain imbedding approach and the relationship between survival signatures and state distributions, a computational method for survival signatures is proposed for multi-state consecutive type systems. Some illustrative examples are also presented to demonstrate how survival signatures are calculated by this method and as to how they are applied.

To provide investors with a selection method of energy storage technology, this paper proposes a quantitative techno-economic comparison method of battery, thermal energy storage, pumped hydro storage and hydrogen storage in wind-photovoltaic hybrid power system from the perspective of multi-objective capacity optimization. The multi-objective capacity optimization models are developed based on minimizing the levelized cost of energy (economy) and loss of power supply probability (reliability) simultaneously. Comprehensive metrics based on hypervolume are proposed to compare the performance of four multi-objective evolutionary algorithms. Moreover, the operation characteristics of devices is considered in the model to improve the simulation accuracy. The performance comparisons of algorithms show that the average rank of nondominated sorting genetic algorithm, multi-objective evolutionary algorithm based on decomposition, multi-objective particle swarm optimization and strength Pareto evolutionary algorithm are 2.8, 3.6, 1.8 and 1.8 respectively, which demonstrates that multi-objective particle swarm optimization and strength Pareto evolutionary algorithm have relatively better overall performance when applied in capacity optimization problems. The quantitative techno-economic comparisons of energy storage show that the levelized cost of energy of thermal energy storage, battery, hydrogen storage and pumped hydro storage under the same reliability are 0.1224 $/kWh, 0.1812 $/kWh, 0.1863 $/kWh and 0.2225 $/kWh respectively, which demonstrates that thermal energy storage is the most cost-effective alternative. Furthermore, the sensibility analysis demonstrates that thermal energy storage is always the most cost-effective alternative for different load profile, different resources level and different energy storage cost. Finally, the conclusions can help investors to select a cost-effective and reliable energy storage technology.

This paper considers a mission-based non-repairable system which has multiple and cyclic mission sequences, and whose components have multimode failures. For example, the valve system with two failure modes cannot only fail to close when required to close but can also fail to open when required to open. A discrete-time semi-Markov chain within the time evolution framework is constructed, which is more properly to model the realistic problems because the sojourn times between the successive missions may be not satisfied Markov property. Using the aggregated stochastic theory and Z-transform, the explicit form of the solution of system reliability is obtained, which can simplify calculations and get some distributions of some sojourn times that the readers relish, including perfect and imperfect operations. In addition, two special valve systems with two-mission and three-mission switching are provided respectively. Finally, numerical examples are presented to illustrate the proposed model and the method.

for the current development of our country, the development of industry is going hand in hand with the economic development of our country, with the development of industry, the overall maturity of machinery manufacturing industry is gradually increasing. In the mechanical manufacturing industry, connecting rod parts are one of the important components of piston engine and compressor, so for these two kinds of machinery, the importance of connecting rod parts is self-evident. This paper will take the machining technology of connecting rod and the corresponding furniture design in the processing process as the starting point to analyze the problems that need attention in the design and processing process.

Qingfengling Small Hydropower Training Base is a base under planning and construction of Sichuan Electric Power Company. The training is based on the relevant requirements of Technical Specification for Concrete Dam Safety Monitoring (DL/T5178-2003), Technical Specification for Earth-Rockfill Dam Safety Monitoring (SL 551-2011) and Technical Specification for Dam Safety Monitoring Automation (DL/T5211-2005) as well as the guidance of some laws and regulations such as Detailed Rules for Dam Safety Inspection of Hydropower Station and Regulations on the Safety Management of Reservoir and Dam. Combining the practice teaching characteristics of hydraulic instrument observation and drawing on the actual monitoring experience of Baozhusi Hydropower Station and Gongzui Hydropower Station, the content of the monitoring training is mainly divided into three parts: field inspection, instrument monitoring, and the collation, compilation and analysis of monitoring data. Combined with the three simulated dams that have been built in Qingfengling Teaching Power Plant, the training of hydraulic instrument observers mainly includes external observation and internal observation, which covers deformation monitoring, seepage observation, stress and strain and temperature monitoring.

This paper proposes an intelligent monitoring system to adapt motorized high speed spindle on machine tools to meet the requirement of Industries 4.0, based on the techniques of power electronic engineering, mechatronics and reliability engineering. In this paper, a force-sensor-integrated spindle is proposed to monitor the bearing stress during operation. The implementation of charge amplifier and mechanical design are described as well. As a result, raw data is transmitted through a filter to create fruitful information and allow the remote monitoring system to operate effectively. The information supports predictive maintenance plans and improves the quality of after-sales service.

This paper studies and explores a suitable blockchain solution for the IoT platform, enabling the platform user and physical devices or the virtual devices to participate in and make data or service transaction. We research on suitable blockchain-related technologies in use; consider the advantages when using blockchain to improve IoT platform’s ability, efficiency, and security, based on the different existing blockchain consensus; benchmark the performance in the IoT scenario; and make a comprehensive discuss.

This paper presents a new short-term reliability assessment method based on time-varying probability ordered tree selection algorithm for the islanded micro-grid under master-slave control. First of all, depending on the different influencing factors, two types of load loss are defined. Secondly, a selection algorithm based on time-varying probability ordered tree is proposed to get the probabilities of several system states with higher probability at each moment. On this basis, the time-varying probabilities of two types of load loss are calculated, and then the probability of the load-point load shedding, and than the expected value of system blackouts of the load-controlled micro-grid system is calculated for short term reliability evaluation. Finally, the case study verifies the correctness and high efficiency of the proposed method and is used as a basis for the operator to control the operation and scheduling of the micro-grid.

Topology control is a basic issue in wireless sensor networks. It has great significance to establish the hierarchy clustering topology for improving the network performance. Characteristics of the immune system, such as memory learning, feedback regulation and distributed autonomous without center, provide a novel approach to the issues that exist in the clustering topology control in wireless sensor networks. Mechanism of immune system based topology control clustering algorithm is presented for wireless sensor networks in this paper. It mainly contains the definition of immune problem, algorithm's design and implementation, simulation test. Through the simulation on the performance of clustering optimization, the algorithm shows the good clustering performance based on the mechanism of immune system.

It is of great significance to study the robustness of different networks under different attacks. To attack the network more effectively, a new attack centrality strategy for a path based on the multi-hop by mapping the problems into percolation, which is called multi-hop centrality, is proposed in this paper. This method is tested respectively in two kinds of model networks (ER network, SF network) and some real networks by removing the path with the largest centrality from a network iteratively, and then calculating the network efficiency, the maximum size of the connected sub-graph and the corresponding number of attacks. The experiment results demonstrate that our method performs good. It also shows that more homogeneous the denser of networks is, the higher robust of the network is, and the better it against with iterative path attack.

As a kind of accurate, rapid and real – time data acquisitions and processing technology, RFID can give unique identification to entity object, has been widely used in various industries, such as manufacturing, sales, transportation and so on. But with its widely application, many relevant problems, especially the safety issues of RFID system and the low cost issue of label have been raised more and more attention by people; therefore, a new and more appropriate security authentication protocol becomes necessary. In this paper, firstly introducing several now available security authentication protocols, and analyzing their strengths and drawbacks, then proposed a new authentication scheme based on Hash function, doing security properties and feasibility analysis of it in theory, and proving that the security properties of this scheme is more efficient, and it is more applicable to meet the needs of people through test at last.

Consider a repairable series system consisting of n units and each of them follows a Markov process with finite state and continuous time. Under independence assumption among units, the repairable series system has been widely studied by using the Markov process method and Lz-transform method. However, both methods have faced the problem of state exploration although some approximation methods have been used. Thus, it is still an interesting and significant problem to be explored. In this chapter, we investigate repairable series systems by using matrix method which has been widely used in aggregated stochastic processes especially in ion channel modeling and aggregated repairable systems. The formulas for reliability, instantaneous and interval availabilities are given in matrix form for four kinds of repairable series systems, general repairable series system, general repairable series system with neglected failures, phased-mission repairable series system and phased-mission repairable series system with neglected failures, respectively. Numerical examples are shown to illustrate the results for the four kinds of systems and present how matrix method is used to solve the problem of state exploration in the Lz-transform method. Finally, the conclusions and some future possible applications are given.

In this paper, a novel well structure for PMOS single-event transient (SET) mitigation is studied by way of technology computer-aided design (TCAD) numerical simulations. Based on a 90-nm CMOS technology, the simulation results show that the proposed selectively implanted deep-N-well (SIDNW) can significantly reduce the SET pulsewidth without area, power, and performance overheads, when compared with the conventional dual-well process. A comparison is also made with the triple-well process.

This paper considers a novel k-out-of-n: F repairable balanced system with m sectors. The characteristic of this balanced system is that the numbers of working components in the m sectors must keep the same at any time. Thus, some components should be shut down manually when the balance is broken. Specifically, two Markov repairable system models are proposed in this paper in terms of different restart mechanisms of shut down components. The excellent Markov process imbedding method is used to obtain the system reliabilities and availabilities. Phase-type (PH) distributions for the lifetime and repair time of the components in the first model are considered. For the second case of the second model, three different maintenance policies are proposed and their performances are compared through several numerical examples. To the best of our knowledge, the k-out-of-n: F repairable balanced system models proposed in the research of this paper are first reported in literature. Finally, numerical examples are given to illustrate the models and the approaches.

The primary test stand (PTS) is a multiterawatt facility with a current of 8-10 MA and rise time of 90 ns, which is being built for Z-pinch studies at the Institute of Fluid Physics, China Academy of Engineering Physics. The PTS consists of 24 modules that are connected in parallel. Each module is based on the Marx generator and water pulse forming line as well as laser-triggered spark gap switch working at 5-MV level. The nominal total stored energy of the PTS is 7.2 MJ. All the Marx generators and laser-triggered spark gap switches had been tested at full voltage level before the PTS was assembled. The design and installation of the PTS are described. In this paper, the most recent results of the PTS commissioning will be introduced.

Covert communication can effectively improve the security of communication content, which has been widely used in wireless communication systems. Most approaches for covert communication are based on the principle of text-embedded in upper layer of the OSI model. In this paper, a novel covert communication scheme is proposed. This method takes advantage of scramble codes and message processing rules of the WCDMA system. It uses a downlink scramble code in the normal WCDMA uplink signal, so that the covert messages will not be received by the base station (BS) and the user equipment (UE). Simulation result shows that the proposed covert communication method has a low bit error rate (BER) performance, and will only make the signals of different business to a slightly lower data transmission rate.

We present a novel image stitching approach, which can produce visually plausible panoramic images with input taken from different viewpoints. Unlike previous methods, our approach allows wide baselines between images and non-planar scene structures. Instead of 3D reconstruction, we design a mesh-based framework to optimize alignment and regularity in 2D. By solving a global objective function consisting of alignment and a set of prior constraints, we construct panoramic images, which are locally as perspective as possible and yet nearly orthogonal in the global view. We improve composition and achieve good performance on misaligned areas. Experimental results on challenging data demonstrate the effectiveness of the proposed method.

Aiming at project characteristics of taekwondo and combining with multi-Agent technique, this paper has put forward a taekwondo athlete physical fitness evaluation system based on multi-Agent. Moreover, it has made elaborate about the system module.

In this paper, we present the leader-driven algorithm (LDA) for learning community structure in networks. The algorithm allows one to find overlapping clusters in a network, an important aspect of real networks, especially social networks. The algorithm requires no input parameters and learns the number of clusters naturally from the network. It accomplishes this using leadership centrality in a clever manner. It identifies local minima of leadership centrality as followers which belong only to one cluster, and the remaining nodes are leaders which connect clusters. In this way, the number of clusters can be learned using only the network structure. The LDA is also an extremely fast algorithm, having runtime linear in the network size. Thus, this algorithm can be used to efficiently cluster extremely large networks.

The continued scaling of feature size has brought increasingly significant challenges to conventional optical lithography.[1-3] The rising cost and limited resolution of current lithography technologies have opened up opportunities for alternative patterning approaches. Among the emerging patterning approaches, block copolymer self-assembly for device fabrication has been envisioned for over a decade. Block copolymer DSA is a result of spontaneous microphase separation of block copolymer films, forming periodic microdomains including cylinders, spheres, and lamellae, in the same way that snowflakes and clamshells are formed in nature - by self-assembly due to forces of nature (Fig. 1a). DSA can generate closely packed and well controlled sub-20 nm features with low cost and high throughput, therefore stands out among other emerging lithographic solutions, including extreme ultraviolet lithography (EUV), electron beam lithography (e-beam), and multiple patterning lithography (MPL).[2;6]Previous research has shown a high degree of dimensional control of the self-assembled features over large areas with long range ordering and periodic structures.[5; 6] The exquisite dimensional control at nanometer-scale feature sizes is one of the most attractive properties of block copolymer self-assembly. At the same time, device and circuit fabrication for the semiconductor industry requires accurate placement of desired features at irregular positions on the chip. The need to coax the self-assembled features into circuit layout friendly location is a roadblock for introducing self-assembly into semiconductor manufacturing. Directed self-assembly (DSA) and the use of topography to direct the self-assembly (graphoepitaxy) have shown great potential in overcoming the current lithography limits.[4]Recognizing that typical circuit layouts do not require long range order, we adopt a lithography sub-division approach akin to double-patterning and spacer patterning, using small guiding topographical templates. Guiding topographical templates with sizes of the order of the natural pitch of the block copolymer can effectively guide the self-assembly of block polymer (Fig. 1b-c). Therefore, circuit contact hole patterns can be placed at arbitrary location by first patterning a coarse guiding template using conventional lithography.[7; 8] This procedure enables generating a higher resolution feature at a location determined by the coarse lithographic pattern. The size and registration of the features are determined by parameters of the template as well as the block copolymer itself.Using this technique, we have proposed a general template design strategy that relates the block copolymer material properties to the target technology node requirements, and demonstrated contact hole patterning at the technology node from 22 nm to 7 nm, for both memory circuits and random logic circuits.[11] The critical dimension of DSA patterns is highly uniform, with their position controlled precisely. As technology scales down, the contact/via density scales up, which simultaneously opens the possibility of using multiple-hole DSA patterns for contact hole patterning and brings in the challenge of printing guiding templates at a small pitch. Using DSA for patterning IC contacts requires further knowledge of the placement of contacts in an IC layout, as the placement of contacts in the IC layout determines the shape and size of the required templates.We hypothesize that there exists a limited set of guiding templates analogous to the letters of an alphabet that can cover all the possible contact hole patterns of a full chip contact layer.[12] This alphabet approach would significantly simplify DSA contact hole patterning when the total number of letters of the alphabet is small and would allow us to focus on fully characterizing only the design spaces for the letters of the alphabet. By positioning these letters in various locations we would be able to pattern the full chip contact layer in the same way that the 26 letters of the English alphabet is sufficient to compose an English newspaper. Some of the most basic letters, such as circular templates for 1-hole DSA patterns and elliptical templates for 2- and 3-hole DSA patterns, have been studied extensively.[12] To establish a complete alphabet, though, requires the examination of the entire standard cell library, as well as the optimization of the layout to further reduce the number of letters in the alphabet.[5]The broad community of DSA researchers has made tremendous progress in the past few years. However to make DSA fully qualified for large-scale semiconductor manufacturing, technical issues such as defectivity reduction and overlay optimization must be solved. While many researchers are developing new block copolymer materials for better chemical properties, there remains more works to be accomplished from the circuit and system design level, including IC layouts optimization for the improvement of DSA process yield and DSA full-chip hotspot detection. Challenges such as optimizing and tuning the template design based on overlay, defectivity, and lithography requirements will need to be further investigated in order for practical implementation in industry.

Directed Self-Assembly (DSA) of block copolymers (BCP) has attracted increasing attention as the potential next generation lithography technology. One of the most promising applications of DSA is the patterning of contact holes in IC circuits using physical guiding templates. In previous studies, researchers have demonstrated that DSA patterns are determined not only by the size and shape of guiding templates, but the template density as well. However, the influence of the pattern density has not been explored systematically, nor is there a fast inspection methodology to visualize and quantify the influence. In this paper, we introduce the concept of DSA Interaction Range (DSAIR). The influence of template density on the DSA patterns is examined using Gaussian convolution of x with y [say what it is a convolution of]. This approach provides us with a fast and quantitative way to model the influence of template density and predict the location of overfilled conditions.

Cylindrical directed self-assembly (DSA) nanostructures is a promising candidate for patterning the contacts and vias in integrated circuits. To match the contact patterns in an IC layout, physical guiding templates have been adopted to generate aperiodic DSA patterns, and templates of different sizes could lead to various DSA patterns. It is found in the experiment that the density of guiding templates has a strong influence on the DSA patterns. At a low template density, templates tend to become overfilled and result in DSA defects. In this paper, we experimentally demonstrate an effective solution to counteract the influence of template pattern density on the quality of DSA using sub-DSA-resolution Assist Features (SDRAFs). We show that SDRAFs can reduce the DSA defects significantly.