The success of convolutional neural networks (CNNs) has made low-latency inference services on Graphic Processing Units (GPUs) a hot research topic. However, GPUs are hardware processors with high power consumption. To have the least energy consumption while meeting latency Service-Level-Objective (SLO), batching strategy and dynamic voltage frequency scaling (DVFS) are two important solutions. However, existing studies do not coordinate them and regard CNN as a black box, which makes inference services less energy-efficient. In this paper, we propose EALI, an energy-aware layer-level adaptive scheduling framework that is comprised of a power prediction model, a layer combination strategy, and an energy-aware layer-level scheduler. The power prediction model uses classic machine learning techniques to predict fine-grained layer-level power consumption. The layer combination strategy combines multiple layers into optimization units to lower scheduling overhead and complexity. The energy-aware layer-level scheduler adaptively coordinates batching strategy and layer-level DVFS according to workloads to minimize the energy consumption while meeting SLO. Our experimental results on NVIDIA Tesla M40 and V100 GPUs show that, compared to the state-of-the-art approaches, EALI decreases energy consumption by up to 36.24% while meeting SLO.

Recently, a large number of convolutional neural network (CNN) inference services have emerged on high-performance Graphic Processing Units (GPUs). However, GPUs are high power consumption units, and the energy consumption increases sharply along with the deployment of deep learning tasks. Although previous studies have considered the latency Service-Level-Objective (SLO) of inference services, they fail to directly take account of the energy consumption. Our investigation shows that coordinating batching and dynamic voltage frequency scaling (DVFS) settings can decrease the energy consumption of CNN inference. But it is affected by (i) larger configuration spaces; (ii) GPUs’ underutilization while data are transferred between CPUs and GPUs; (iii) fluctuating workloads. In this paper, we propose EAIS, an energy-aware adaptive scheduling framework that is comprised of a performance model, an asynchronous execution strategy, and an energy-aware scheduler. The performance model provides valid information about the performance characteristics of CNN inference services to shrink the feasible configuration space. The asynchronous execution strategy overlaps data upload and GPU execution to improve the system processing capacity. The energy-aware scheduler adaptively coordinates batching and DVFS according to fluctuating workloads to minimize energy consumption while meeting latency SLO. Our experimental results on NVIDIA Tesla M40 and V100 GPUs show that, compared to the state-of-the-art methods, EAIS decreases the energy consumption by up to 28.02% and improves the system processing capacity by up to 7.22% while meeting latency SLO. Besides, EAIS has been proved to have good versatility under different latency SLO constraints.

SummaryConvolutional neural network (CNN) inference usually runs on high‐performance graphic processing units (GPUs). Since GPU is a high power consumption unit, that makes the energy consumption increases sharply due to the deep learning tasks. The energy efficiency of CNN inference is not only related to the software and hardware configurations, but also closely related to the application requirements of inference tasks. However, it is not clear on GPUs at present. In this paper, we conduct a comprehensive study on the model‐level and layer‐level energy efficiency of popular CNN models. The results point out several opportunities for further optimization. We also analyze the parameter settings (i.e., batch size, dynamic voltage and frequency scaling) and propose a revenue model to allow an optimal trade‐off between energy efficiency and latency. Compared with the default settings, the optimal settings can improve revenue by up to 15.31×. We obtain the following main findings: (i) GPUs do not exploit the parallelism from the model depth and small convolution kernels, resulting in low energy efficiency. (ii) Convolutional layers are the most energy‐consuming CNN layers. However, due to the cache, the power consumption of all layers is relatively balanced. (iii) The energy efficiency of TensorRT is 1.53× than that of TensorFlow.

为了解决流程工厂模型拓扑相似度计算问题,提出基于概率分布函数的相似度计算方法.首先利用隐马尔可夫建模及状态序列推断,从工厂拓扑结构中挖掘隐含的拓扑语义;然后通过隐含狄利克雷分配模型将拓扑语义映射成长度固定的特征向量;最后用概率模型描述全体特征向量分布,并用分布函数计算2个拓扑结构间的相似度.依据流程工厂专家设计建模的专业知识及从业经验,验证了该方法的准确性和有效性.

The feature vector mapping of process plant’s topology structure is one of the key problem in process plant topology similarity calculation. High quality feature vector does not only need to convert the expression from graph to length fixed vector, but also should mine and express the topology semantics that under the connective relationships. A novel method has been presented in this paper to solve this problem. We convert the topology graph to observation sequences according to the reacting processes with path extracting algorithm; then reference the state sequences from observation sequences with HMM, this can remove the node dependencies and mine the latent topology semantics; at last, reduce the state sequences to a length fixed vector with LDA model. The background and basis definitions has been introduced in this paper firstly, some related work has been also described; then we analyzed the characteristics of process plants’ topology structure and present the details of the mapping method; finally, the results have been evaluated. According to the manually judgement by process industry professionals, the method is proved to be practical and effective.

To eliminate the drawbacks of traditional small feature culling algorithm which processing granularity are triangles and can't efficiently cope with the number of vertexes and triangles up to hundreds of millions in a certain time period,an improved small feature culling algorithm for large scale process plant based on octree was proposed.Based on the component primitive characteristics and spatial characteristics of the process plant model,the value of screen for quantizing the size of component was proposed,and the established octree and the value of screen were combined to estimate the upper limit of the number of pixels,so as to quickly determine whether the component would be culled or not.The experimental results show that the proposed algorithm is simple and effective.Compared with the current popular review software after loading the factory model with 10 000 pipelines,the frame rate is increased by at least 50％,which gready improves the platform's fluency.Process factory industry and graphics platform as a whole to enhance the level of design has a positive meaning.

Topology information plays the most important role in engineering CAD drawings. The efficiency and quality of engineering design often depend on the topology integrity of drawings. This paper proposes a general topology authentication algorithm, which is inspired by the semi-fragile watermarking scheme, to verify the topology integrity of heterogeneous engineering CAD drawings. Topology graph is firstly built according to the topology relation among objects with uniform codes. Then, the Laplacian matrix is con-structed and its Laplacian spectrum is computed to generate topology feature values. These values are then embedded into the geometric invariants of objects. Topology integrity verification is achieved through com-paring the extracted binary values with the topology feature values computed according to the current to-pology relation. Theoretical analysis and experiments have demonstrated that the proposed scheme can not only detect and locate malicious topological modifications, but also yield strong robustness against various topology-preserving modifications, such as global and local similarity transformations, copying, mirroring and file format converting. It's believed that the proposed scheme is a general algorithm and applicable to various kinds of heterogeneous engineering CAD drawings in industry practices.

This paper presents a novel unified framework for authenticating topology integrity of 2D heterogeneous engineering CAD drawings. Topology information, through which a variety of engineering reports are generated, plays the most important role in the engineering CAD field. However, topology integrity authentication for engineering CAD drawings is still in its infancy and few efforts were made in the literature. By efficiently extracting topological and geometric features regardless of their heterogeneity with respect to geometrical shape and topology representation, the proposed framework supports verifying topology integrity for various heterogeneous engineering CAD drawings, such as process flow diagrams (PFD), piping and instrument drawings (P&ID), piping isometric drawings (ISO) as well as sectional drawings. Topology authentication is achieved through embedding local topological features into geometric features by introducing a generic and effective semi-fragile watermarking scheme. A novel descriptor, called topology local binary patterns (T-LBP), is proposed to extract local topological features of heterogeneous drawings. Theoretical analysis and experiments have demonstrated the discrimination power, robustness and sensitivity of the proposed T-LBP descriptor. We also carry out further experiments to prove that the proposed framework can not only detect and locate malicious topological modifications, but also yield strong robustness against various topology preserving modifications.

Process plant models mainly include 3D models and 2D engineering drawings. Matching calculation between these CAD models has wide applicability in model consistency check and retrieval. In process plant, engineering design standards make 2D engineering drawing and 3D model differ in geometry, proportion and structure, leading to the inapplicability of current shape-feature based matching approaches. Since connection relationships between components are the core of a process plant, a topology based algorithm is proposed. Firstly, by exploiting components as vertices and relationships as edges, both 2D engineering drawing and 3D model are preprocessed into graph structures. Then each model's relationship types are extracted from the graph. Finally, regarding the extracted relationship types as primary feature, feature similarity is calculated to measure the matching degree between their corresponding models. The proposed algorithm is geometric deformation invariant. Experiments with industrial applications are presented, which demonstrates the effectiveness and feasibility of the proposed algorithm. (C) 2017 Elsevier Inc. All rights reserved.

Design review is important in collaborative design of process plants. To satisfy the actual work demands of design review, a 3-d design review system is developed and the key technologies such as information organization model and multi-resolution rendering approach are proposed. The information organization model combining scene tree and attribute tree can organize the information from different CAD systems with a unified structure, and optimize the information query speed. The multi-resolution rendering approach based on programmable graphics pipeline can improve rendering efficiency within less preprocessing time, without using extra hard-disk space. Examples show that the 3-d design review system can work on a general PC to review a large quantity of design information from different subjects, and ensure real-time interaction at the same time.

为提高流程工厂的设计效率,针对流程工厂的领域特性,提出一种基于编辑距离的流程工厂模型局部检索算法。算法将图相似性的相关思想与流程工厂模型结合,首先将待检索模型和历史流程工厂模型转化为属性图结构;然后设定阈值,并计算待检索模型与历史流程工厂模型对应的属性图之间的最小编辑距离:若计算结果大于阈值,则认为当前模型不包含待检索模型;否则返回检索到的局部模型。该算法将流程工厂模型的局部检索问题简化为属性图的子图查询问题,实现了流程工厂模型的局部检索。通过实验证明了算法的有效性和可行性。

While machining width is an important factor of the machining time of freeform surface finishing operations, in reality the kinematic capability of the machine tool is usually the bottleneck of achieving higher feed speed and optimal machining time.The purpose of this paper is to conveniently (and approximately) determine the optimal cut direction considering the speed kinematic capability of the machine tool, without having to compute the actual tool path.We propose a mathematical instrument, called Machine Kinematic Metric (MKM), to easily evaluate infinitesimal machining time on a freeform surface based on machine kinematic consideration.It's a tensor field similar to the metric tensor in differential geometry.MKM is integrated over the part surface to approximate the cut-direction-dependent total machining time, and used to determine the optimal cut direction that minimizes the machining time.To validate the accuracy of the prediction using MKM, we apply the method and compute the machining time at every direction with one degree apart and derive the optimal cut-direction.The computation is performed on two examples: a simple freeform surface and a complex die face model.We then use a commercial CNC emulator software from Huazhong CNC to precisely simulate the machining time in distributed cut directions (five degree apart) for the two models.We find that the optimal cut direction determined from CNC simulation is consistent with the prediction from the proposed method.It validates that the proposed method is a convenient and economical tool to approximately determine the optimal cut direction based on machine speed kinematic capability.

Process plant Computer-Aided Design (CAD) models can be roughly divided into 3D models and 2D engineering drawings. Matching calculation of these CAD models not only contributes to their coherency verification, but also benefits for the development of model retrieval. However, in process plant area, 3D models and 2D engineering drawings are different in both graphical representations and content structures, which leads to the inapplicability of current shape-feature based 2D & 3D matching approaches. To resolve this problem, a topological structure based algorithm is proposed. Exploiting component as the basic unit, we firstly transform 2D engineering drawings and 3D models into attribute graphs. Then, by introducing related researches of graph similarity, we calculate attribute graph similarity to measure the matching degree between their corresponding CAD models. The proposed algorithm is translation, rotation and similarity transformation consistent. Experimental results demonstrate its effectiveness and feasibility.

To better understand the evolution process of emotion in crowd situation,this paper provided a method of building the emotion model based on interpersonal relationship.It constructed interpersonal relationship on the basis of the small world network,and represented the emotional relationship among persons in the real life by parameter of mood introduced in the model.Based on the Bosses' group emotion model and the relationship network as the medium of emotion contagion,it simulated the evolution process of emotion under the different relationship situations.It studied the influence of the adjacent neighbors number K,the rewiring probability P and the affection relationship R on the result of emotion contagion.The numerical simulation results confirm that the closer the affection relationship of crowd is and the more the neighbor number of K,the stronger the emotion strength of crowd and the shorter the time is.Rewiring probability P also has slightly effect on the intensity of group emotion.

Piping isometric drawings, which feature their intrinsical topological relation rather than just geometrical shape, are important industrial art works in the field of Computer-Aided Design (CAD). This paper takes a fresh look at the topology integrity authentication of piping isometric drawings, which has not been mentioned before in the literature, from the digital watermarking perspective. A blind and semi-fragile watermarking based algorithm is proposed to address the referred interesting issue. The topology authentication problem of piping isometric drawings is investigated. In addition to the stretching operation, both global and local similarity transformation operations, which are critical problems in the case of watermarking embedding and extraction, are analyzed in detail. The topological graph is extracted and constructed from the drawing firstly. Then, similarity transformation invariants are constructed as watermarks carriers for each node. After that, the topological relation among joint components is encoded into singular watermarks for each node of the graph. These generated topology sensitive watermarks are embedded into geometrical invariants of each node via quantization index modulation. Theoretical analysis and experimental results demonstrate that our approach yields a strong ability in detecting and locating unauthorized topology attacks while achieves robustness against both global and local similarity transformations especially the stretching operation. The proposed scheme can be employed to authenticate topology integrity for each of the drawings derived from the model individually in industry practices.

针对港口码头结构CAD几何模型生成CAE单元模型,工程荷载CAD模型创建CAE荷载模型,和CAE分析结果与CAD模型的关联关系等问题,利用模板的模型独立性、规则性和可执行性等特性,研究了基于模板的港口码头三维结构CAD/CAE集成技术,定义了截面草图、三维实体和构件等多种CAD建模模板,定义了机械草图和荷载等多种荷载建模模板,定义了CAE分析规则,实现了CAD几何模型自动生成CAE单元模型的功能,定义了荷载转换规则,实现了工程荷载CAD模型自动创建CAE荷载模型的功能,定义了整体模型模板,建立CAE分析结果与CAD模型的多对一关系,以此实现了港口码头三维结构CAD/CAE集成系统,大大提高了港口码头的设计效率,降低了设计人员的设计难度,减少了设计错误.

The Computer-Aided Plant Design (CAPD) model characterizes its peculiar complex topology among a tremendous number of plant components under complex constraints rather than just geometrical shapes. Lots of watermarking schemes for CAD models have been proposed for geometrical information protection or authentication. However, in the literature, none of them has mentioned the problem of topology authentication for CAPD models yet. In this paper, a semi-fragile watermarking algorithm for topology authentication of CAPD models is presented. We first discuss the problem of topology authentication of CAPD models. Then a subset of the model’s connection points are selected as mark points, also called watermark carriers, according to the mark point selecting principle. We encode the topological relation among components into watermarks. Afterwards, the topology based watermarks are embedded in mark points to keep them in a predefined relationship with neighboring connection points so that any changes will ruin the relationship between marked connection points and neighboring connection points. To the best of our knowledge, our algorithm is the first semi-fragile and blind scheme that can authenticate and verify the topology of CAPD models. Experimental results show that our approach not only can detect and locate malicious topology attacks such as components modification and joint ends modification, but also is robust against various non-malicious attacks such as similarity transformations and simplification.

To express the design process effectively by establishing the process model and framework is the foundation of collaborative design.According to the features of design process for complex product,a multi-level process model of collaborative design for complex product based on the dimensions of time,particle size and parallelism is established,combining with meta-synthesis.This process model emphasizes on the combinations of multi-stage in time dimension,combinations of multi-specialty in parallelism dimension and iteration of decomposition and integration in particle dimension.Then,on the base of the process model,a process framework of collaborative design for complex product based on meta-synthesis is proposed,combining with expert system,knowledge system and machinery system.Machinery system helps the expert group with knowledge from knowledge system.Expert system fulfills the collaborative design process of combinations of human-computer,multi-specialty and multi-stage,with the services provided by machinery system,e.g.project management,group argumentation,collaborative decision and 3D visualization.Finally the prototype system which has been developed based on the idea of this paper is introduced.

The research of computable emotional contagion has important significant for understanding the behavior choice and decision-making of individuals in crowd.In order to make the behavior of virtual individual more believable,we proposed a method to simulate the process of emotional contagion happened in crowd based on the psychological theory.The proposed emotion model can reflect the dynamic process of individual in group on three factors: personality,the capability of emotional influence upon others,and the size of group,and can provide the way to calculated the level of the emotion induced by external stimulus,the emotional change caused by emotional contagion among individuals and the degree of emotional decay.Based on the individual's emotion model provided by us,some experiments have been carried out.And the results show that the emotion model is efficient to simulate the emotion of individuals in crowd.

Topology authentication for computer-aided plant design (CAPD) models features intrinsically complex topological relations. This study investigates a semi-fragile watermarking scheme for CAPD models represented by parametric solids, which offers a solution to the problem of topology authentication. We first analyze the geometrical and topological structures of CAPD models. Then, we propose an effective semi-fragile watermarking method for topology authentication, which is based on Laplacian coordinates and quantization index modulation (QIM), against several attacks. We compute the custom Laplacian coordinate vector for each marked connection point according to the topological relation among joint plant components. The topology-based watermark for each marked connection point is generated from selected attributes of its joint plant component. Watermarks are inserted into the coordinates of marked connection points by adjusting the lengths of their Laplacian coordinate vectors. Both experimental results and theoretical analysis demonstrate that our approach can not only detect and locate malicious topology attacks, such as component modification and joint ends modification, but is also robust against various non-malicious attacks, such as similarity transformations and level-of-detail (LOD).