Brakes are responsible for the stability of the vehicle. Brake failure is one of the key elements where more attention is required. Normally, a brake system failure is not an instantaneous process. It is caused by faults due to reasons like wear, mechanical fade, and oil leak, which started long before the failure progresses. Hence, it is essential to build a model that can recognize the condition from the signal. Condition monitoring is one such supervision approach, which continuously monitors the system and gives characteristics data. These data can be analysed and the condition of the component can be extracted using a machine learning approach. This study focuses on one such machine learning approach using the vibration characteristics of the brake system. The machine learning approach was carried out using feature extraction and feature classification. The statistical information extracted from the vibration signals under various fault conditions were used as features. The features were classified using machine learning algorithms, namely, Simple logistics, Logit boost and Multinominal Regression. Results were compared and discussed. The Logit boost algorithm, which produced 98.91 % classification accuracy, has been suggested as an effective approach for the brake fault diagnosis study.

Submitted on June 30, 2017; Revised on November 22, 2017; Accepted on November 27, 2017
References: 18

Turbocharger is used to increase the efficiency of an engine. In the turbocharger, a centrifugal compressor is used. Centrifugal compressor failure has been a main issue in recent years. The failure of the centrifugal compressor is because of surge and stall. Surge occurs because pressure at the receiver is greater than the pressure at the compressor. So, gas flow will reverse and surge occurs. This work-study is based on reducing surge and stall. We can prevent the tendency of surge and stall by changing the factors affecting the change in pressure. We can change the pressure by changing the density of gas|Noof impellers, impeller diameter, inlet and volute geometry and flow regulation. Hence, in this paper, we perform the CFD analysis of the centrifugal compressor by changing the geometry.

Submitted on July 19, 2017; First Revised on September 29, 2017; Second Revised on November 9, 2017; Accepted on November 29, 2017
References: 12

A CART decision tree algorithm based on attribute weight is proposed in this paper because of the present problems of complex classification, poor accuracy, low efficiency, and severe memory consumption of CART decision. What is more, the algorithm is combined with the parallel computing model of MapReduce. Theory of attribute weights is used in the algorithm. A decision tree is built through the sum of weights, which is decided by the degree that the attributes affect a decision. Thus the accuracy of classification through decision tree is improved. Parallel sorting algorithms of CART decision tree for massive data is implemented through the MapReduce programming technology of cloud computing. All the results of theoretical analysis and experimental comparison show that it is very important to mark attributes by weights through MapReduce. Furthermore, the accuracy of the classification of large sample data sets is improved significantly, classification efficiency of decision tree is improved and the trained time is also significantly reduced.

Submitted on July 24, 2017; Revised on September 25, 2017; Accepted on November 29, 2017
References: 28

In the process of user-based collaborative filtering algorithm, finding similar users effectively plays a crucial role in obtaining a high recommendation accuracy. The original rating matrix is very sparse, resulting in similarity information loss during similarity calculating and degrading the efficiency of similar users extracting. To tackle this problem, we propose an improved collaborative filtering algorithm based on Latent Semantic and Bayesian (ICFLSB). ICFLSB first utilizes Latent Semantic to extract meaningful features in the original rating matrix. Then, we establish a Bayesian model based on these extracted features to predict items which users have not rated but may be interested. Further, we fill the sparse original rating matrix with these predicted items and find similar users. After that, we adopt the collaborative filtering algorithm to conduct recommendations. Experiments show that ICFLSB proposed in this paper has a better recommendation performance than the traditional collaborative filtering algorithm. In particular, the evaluation results demonstrate that our ICFLSB can achieve a 2.152% higher and 1.152% higher on recommendation accuracy and recall rate respectively when compared to the traditional collaborative filtering algorithm.

Submitted on November 1, 2017; Revised on December 10, 2017; Accepted on December 20, 2017
References: 21

In the recent advocated reform of education and teaching by China’s education ministry, the MOOCS teaching concept and mode is widely used in higher education teaching systems. This paper focuses on the Sino-foreign cooperative design major and examines its unique curriculum, teaching system, talent cultivation and other characteristics, through the selection of different levels and different types of representative colleges and universities and Sino-foreign cooperative design majors. Utilizing varieties of methods including questionnaire investigation, face-to-face interview and discussion, statistical analysis, long-term tracking research and so on, this paper aims to research, analyze, demonstrate and reveal the effects of introducing MOOCS mechanisms in education. The teaching of the Sino-foreign cooperative design major, especially points to embodied advantages and disadvantages to show its reliability, performability and existing risks. The paper puts forward suggestions for improvement; countermeasures and sustainability assessments so as to further improve and optimize MOOCS, deepen application and better promote the reform of education and teaching in the corresponding fields in China.

Submitted on October 20, 2017; Revised on December 17, 2017; Accepted on December 24, 2017
References: 16

Analysis of forest canopy hemisphere images is one of the most important methods for measuring forest canopy structure parameters. This paper focuses on improvements to contour extraction method for the circular imaging region in forest canopy. We proposed an effective contour extraction algorithm based on circular hough transform for the circular imaging region of digital canopy hemispherical photography (DHP). In order to eliminate the threshold segmentation error, the morphological opening and closing was used to remove these desultory points and smooth the outline of the imaging region. After the boundary points of the imaging area were extracted through erosion, the circular hough transform was applied to fit the boundary point for finding the position of the center and gaining the radius. The experimental results proved its feasibility in terms of effective implementation and robustness.

Submitted on September 29, 2017; Revised on November 2, 2017; Accepted on December 21, 2017
References: 25

After studying the topological structure of neighboring nodes in the WSN, we present a local localization (LLA) algorithm by combining the ideas of principal manifold learning and nonlinear dimension algorithm. This algorithm is particularly suitable for determining the relative locations of sensor nodes in the large-scale and low-density WSNs, where the low connectivity between nodes and the large ranging error between long-distance nodes usually make accurate localization quite difficult. In this algorithm, based on the pair-wise distance between each node and its neighbor nodes within a certain communication range, the local geometry of the global structure is firstly obtained by constructing a local subspace for each node, and those subspaces are then aligned to give the internal global coordinates of all nodes. Combined with the global structure and the anchor node information, we can finally calculate the absolute coordinates of all unknown nodes in the least squares algorithm.

Submitted on November 4, 2017; Revised on December 1, 2017; Accepted on December 19, 2017
References: 15

Due to force-directed algorithms’ capabilities of producing aesthetically pleasing graph layouts, which follow metrics for graph drawing aesthetics, these layouts have become the most common methods in the practical data visualization area. However, evaluating the performance of relevant algorithms remains a challenge, since graph layout quality is largely relying on aspects such as human intuition, personal judgment and methods’ pre-setting parameters. In addition, most aesthetics criteria of graph drawing conflict with each other. This study evaluated the performance measurements of four force-directed algorithms in terms of seven commonly applied aesthetic criteria based on practical raw data collected, and demonstrated the experimental framework. The early outcomes compared twenty final graph layouts and gave empirical evidences; the study may assist with future detailed force-directed algorithms selection based on users’ specific requirements.

Submitted on October 14, 2017; Revised on December 5, 2017; Accepted on December 20, 2017
References: 23

Image tactile interaction is a new interaction method between human and digital images, allowing users to know digital images by touch. In order to improve the authenticity of image tactile perception, this paper proposes a tactile perception model based on image region features. Aiming at the phenomenon of over-partitioning and computational complexity of JSEG algorithm, we propose an improved JSEG algorithm, which can effectively reduce the computational complexity, divide the image areas which are more in line with the subjective visual judgment, and can be used for region-based image tactile generation. The experimental results show that the proposed algorithm can correctly distinguish the image area and improve the accuracy of image tactile perception.

Submitted on October 20, 2017; Revised on November 23, 2017; Accepted on December 7, 2017
References: 15

Numerous experiments have shown that there are similar gestures in visual-based gesture recognition. In order to solve the problem, this paper proposes a new algorithm based on Convolutional Neural Network. According to the model test results, the confusion matrix is established. According to the correspondence between each gesture and the predicted result, probability matrix of misjudgement is established. Based on the probability matrix of misjudgement, we correct the gestures that have been incorrectly identified by the Convolutional Neural Network Model. After this algorithm, the recognition rate of similar gestures is increased from 5% to 12%. The innovation of this paper lies in the secondary error correction of the wrong gesture of Convolutional Neural Network structure.

Submitted on October 20, 2017; Revised on November 23, 2017; Accepted on December 7, 2017
References: 54

This paper aims to address the problem of joint segmentation and recognition of multiple actions in a long-term video. Since features obtained from a single frame cannot describe human motion in a period, some literatures initially divide a long-term video into many video clips with fixed length and represent a long-term video as a sequence of video clips. However, a fixed-length video clip may contain frames from two adjacent actions, which would significantly affect the performance of action segmentation and recognition. In this paper, we develop a double-layer framework for segmenting and recognizing multiple actions in a long-term video. In the first layer, a novel unsupervised method based on the directions of velocity is proposed to initially divide an input video into a series of clips with unfixed length. The second layer takes a sequence of video clips as input, and employs a joint segmentation and recognition method to group video clips into several segments while simultaneously labeling the action category for each segment. Experiments conducted on the IXMAS action dataset verify the effectiveness of the proposed approach.

Submitted on October 2, 2017; Revised on November 15, 2017; Accepted on December 10, 2017
References: 32

This paper proposed an algorithm that fused block-based algorithm (Radial Harmonic Fourier Moments, RHFMs) and SIFT algorithm (Scale Invariant Feature Transform, SIFT) for the duplicated forgery detection, also called copy-move forgery detection. It can effectively detect forgery image. Firstly, the SIFT algorithm is proposed to extract feature points from the pre-processing image. Subsequently, the nearest-neighbor (2NN) test, adaptive Euclidean distance and Random sample consensus (RANSAC) are applied to remove most of the mismatched feature points and get the candidate inlier matches. The affine matrixes based on the RANSAC are obtained by the candidate inlier matches. Then, Radial Harmonic Fourier Moments is proposed to extract invariances of the candidate inlier matches in circle blocks. The propagated criterion is calculated by affine matrix and circular feature of the inlier matches. The Simple Linear Iterative Clustering (SLIC) is presented to segment the host image into texture patches. Like pixel propagating, the circular block is slid in the corresponding texture blocks that the texture blocks contain the corresponding inlier matches to get more corresponding matches. Finally, some geometric image operations, such as dilation, are employed to eliminate the small holes or isolated pixels. A series of experiments showed that the proposed fusion algorithm can achieve superior performances than those of the moment invariant algorithms under various geometric transformations.

Submitted on October 17, 2017; Revised on November 23, 2017; Accepted on December 29, 2017
References: 23

Plant modeling based on point cloud is more difficult than others because of the noise-edge bonding. We present a method called the limited detail multi-density plants modeling to address these problems mentioned above. The core of the proposed method is a point-to-surface approach, which constantly refines the density of points with original RGB information to generate the fuzzy surface. Compared to conventional plants mesh models, the primary advantage of our approach is that it does not compute the steps of texture and illumination, and saves rendering time and storage space. We validate our method with both leaves and simple plants, and demonstrate that the limited detail multi-density point reconstruction is feasible, and it can generate good results with fast speed while using less storage space.

Submitted on November 12, 2017; Revised on November 30, 2017; Accepted on December 20, 2017
References: 22

Emotion contagion is an indispensable behavior in a dynamic crowd, especially in an evacuation situation. As a consequence, generating emotion contagion results is very useful in the crowd simulation field. However, because the topology of the crowd usually keeps changing dynamically, computing the contagion process is a challenge. In this paper, we represented our research about the emotion contagion effects on the virtual pedestrian dynamic aggregation process. First of all, we calculated individuals’ moving parameters based on their prefixed expectations according to the social force theory. After this, we made an adjacent test for each individual to generate nearer neighbors for further emotional contagions computing between neighbors. We then treated the emotional contagions between individuals and their neighbors as the information spreading process so that we can adopt the emotional information spreading model SIR (Susceptible Infective Removal) to calculate emotional influences, which are represented as their changing moving velocities during aggregation. Social force for computing low level moving parameters and SIR model for generating emotional influences were integrated by our method to simulate the dynamic pedestrian aggregation. Experimental results showed that the SIR model can effectively improve the fidelity of the emotional interaction process and crowd aggregation.

Submitted on November 20, 2017; Revised on December 10, 2017; Accepted on December 23, 2017
References: 26

In this paper, a non-contact measurement method with interactive functions is proposed based on the HoloLens, a mixed-reality product by Microsoft. By scanning the measurement space using the depth camera on the HoloLens, and by determining the measurement points using the Gaze and Gesture functions, this method enables non-contact measurement and automatic calculation of the distance, area and volume between the target points. Furthermore, it supports speech recognition for interaction. Compared with traditional measurement tools, the interactive measurement tool developed in this paper allows its users to take measurements in a more natural, contact-free way. Its broader measurement range, automatic calculation of area and volume while measuring distances, and the ability to measure places that are difficult for the traditional measuring tape, show its good application value.

Submitted on November 13, 2017; Revised on December 6, 2017; Accepted on December 22, 2017
References: 9

Hand motion capture systems usually use the anatomical skeleton statistics size as the skeleton length of the virtual hand. The difference in user hands leads to error in the delicate hand tracking process. By analyzing the skeleton structure and movement characteristics of the hand, a three-dimensional hand skeleton model is constructed. Based on two specific gestures, the virtual hand model is calibrated. A skeleton proportion estimation algorithm based on specific attitudes is proposed. It can estimate the exact proportion of hand bone and realize the initial deviation correction of sensor data. In the meantime, the attitudes and positions of the virtual hand skeleton is calculated based on kinematics. Hand movement tracking is realized with accurate reduction. Using this method, a hand tracking and interaction system works fine.

Submitted on October 25, 2017; Revised on November 30, 2017; Accepted on December 11, 2017
References: 8

Large display systems have been successfully applied in the virtual reality domain because they can provide a full immersion sense through large visual space and high display resolution. However, most of the previous studies on the interaction method of those systems focused on single or double users. In this paper, we study the effects of integrating group interaction in such systems and we propose a framework called “Groupnect”, which enables the unique experience of group interaction in a large display system. By using optical tracking and 3D gesture recognition technologies, our approach can automatically recognize gesture-based control signals for 20 users simultaneously and trigger corresponding real-time actions in a back-end system. Through a comparison experiment between standard interaction mode and group interaction mode, the results demonstrate that physical and mental participation of users could be promoted in group interaction mode. It has immense potential to design group interaction applications on entertainment, education and training areas.

Submitted on October 18, 2017; Revised on November 19, 2017; Accepted on December 17, 2017
References: 25

In this paper, we illustrate how to learn a general straightforward similarity function from raw image pairs, which is a fundamental task in computer vision. To encode the function, inspired by the recent achievements of deep learning methods, we explore several deep neural networks and adopt one of the suitable networks to our task encoding implementation with several models on benchmark datasets UKBench and Holidays. The adopted network achieves comparable overall results and especially presents the excellent learning ability for global-similar data. Compared to previous approaches, this function eliminates the complex handcrafted features extraction, and utilizes pairwise correlation information by the jointly processing.

Submitted on November 28, 2017; Revised on December 15, 2017; Accepted on December 23, 2017
References: 33

With the technical innovation of microprocessors, the application of cache memory prevents the main memory from limiting the development of microprocessors. However, the introduction of Cache also brings the cache coherence problem while the microprocessor performance is being improved. Based on this problem, this paper designs an improved CMP architecture model and proposes a new Cache Coherence Protocol Based on Virtual Bus (CCPBVB). Firstly, the D-Cache virtual bus is added to the architecture to realize the point-to-point consistency transaction transmission, avoiding the bus idle phenomenon caused by the polling query method that the broadcast consistency transaction must follow, while also improving the effective utilization rate of the bus. Then, in order to reduce the access delay caused by the ping-pong phenomenon in the cache memory, a protocol is designed to improve the C hit rate, which combines the write invalid strategy and write update strategy between the private cache to reduce the delay time of the system. The experimental results show that the proposed protocol not only improves the bus utilization, but also reduces the C access delay.

Submitted on October 22, 2017; Revised on November 19, 2017; Accepted on December 21, 2017
References: 16

This paper considers an advanced replacement model for a two-unit system. One is the key unit called unit 1 and the whole system is replaced once unit 1 fails. The system is preventively replaced in order to prevent failure of unit 1. The other unit referred to as unit 2, is cheaper and easily replaceable. Unit 2 is only replaced when it fails. However, if its failure is within a threshold of the planned preventive system replacement, the system replacement is done in advance to avoid over replacements of unit 2. A framework is proposed to jointly optimize the preventive replacement interval and the threshold for advanced replacement.

Submitted on October 1, 2017; Revised on November 7, 2017; Accepted on December 5, 2017
References: 8