三峡大学科技学院毕业设计外文翻译
我的翻译原文

ORIGINAL PAPEREggshell crack detection based on acoustic response and support vector data description algorithmHao Lin ÆJie-wen Zhao ÆQuan-sheng Chen ÆJian-rong Cai ÆPing ZhouReceived:21May 2009/Revised:27August 2009/Accepted:28August 2009/Published online:22September 2009ÓSpringer-Verlag 2009Abstract A system based on acoustic resonance and combined with pattern recognition was attempted to dis-criminate cracks in eggshell.Support vector data descrip-tion (SVDD)was employed to solve the classification problem due to the imbalanced number of training samples.The frequency band was between 1,000and 8,000Hz.Recursive least squares adaptive filter was used to process the response signal.Signal-to-noise ratio of acoustic impulse response was remarkably enhanced.Five charac-teristics descriptors were extracted from response fre-quency signals,and some parameters were optimized in building model.Experiment results showed that in the same condition SVDD got better performance than con-ventional classification methods.The performance of SVDD model was achieved with crack detection level of 90%and a false rejection level of 10%in the prediction set.Based on the results,it can be concluded that the acoustic resonance system combined with SVDD has significant potential in the detection of cracked eggs.Keywords Eggshell ÁCrack ÁDetection ÁAcoustic resonance ÁSupport vector data descriptionIntroductionIn the egg industry,the presence of cracks in eggshells is one of the main defects of physical quality.Cracked eggsare very vulnerable to bacterial infections leading to health hazards [1].It mostly results in significant economic loss in the egg industry.Recent research shows that it is possible to detect cracks in eggshells using acoustic response analysis [2–5].Supervised pattern recognition models were also employed to discriminate intact and cracked eggs [6].In these previous researches,training of discrimination models needs a considerable amount of intact egg samples and also corresponding defective ones.However,it is more difficult to acquire sufficient naturally cracked eggs samples than intact ones.Artificial infliction of cracking in eggs is time-consuming and a waste.Moreover,the artificially cracked eggs may not provide completely authentic information on naturally cracked ones.So,the traditional discrimination model shows poor performance when the numbers of sam-ples from the two classes are seriously unbalanced,because the samples of minority group cannot provide sufficient information to support the ultimate decision function.Support vector data description (SVDD),which is inspired by the theory of two-class support vector machine (SVM),is custom-tailored for one-class classification [7].One-class classification is always used to deal with a two-class classification problem,where each of the two classes has a special meaning [8].The two classes in SVDD are target class and outlier class,respectively.Target class is assumed to be sampled well,and many (training)example objects are available.The outlier class can be sampled very sparsely,or can be totally absent.The basic idea of SVDD is to define a boundary around samples of target with a volume as small as possible [9].SVDD has been used to solve the problem of unbalanced samples in the field of machine faults diagnosis,intrusion detection in the network,recog-nition of handwritten digits,face recognition,etc.[10–13].In this work,the algorithm of SVDD was employed to solve the classification problem of eggs due to imbalancedH.Lin ÁJ.Zhao (&)ÁQ.Chen ÁJ.Cai ÁP.ZhouSchool of Food and Biological Engineering,Jiangsu University,212013Zhenjiang,People’s Republic of Chinae-mail:zjw-205@;zhao_jiewen@ H.Line-mail:linhaolt794@Eur Food Res Technol (2009)230:95–100DOI 10.1007/s00217-009-1145-6number of samples.In addition,recursive least squares (RLS)adaptive filter was used to enhance the signal-to-noise ratio.Some excitation resonant frequency charac-teristics of signals were used as input vectors of SVDD model to discriminate intact and cracked eggs.Materials and methods Samples preparationAll barn egg samples were collected naturally from a poultry farm and they were intensively reared.These eggs were on maximum 3days old when they were measured.As much as 130eggs with intact shells and 30eggs with cracks were measured.The sizes of eggs ranged from peewee to jumbo.Irregular eggs were not incorporated into the data analysis.The cracks,which were 10–40mm long and less than 15-l m wide,were measured by a micrometer.Both,intact and cracked samples,were divided into two subsets.One of them called calibration set was used to build a model,and the other one called prediction set was used to test the robustness of the model.The calibration set contained 120samples;the number of intact and cracked samples were 110and 10,respectively.The remaining 40samples constituted the prediction set,with 20intact eggs and 20cracked ones.Experimental systemA system based on acoustic resonance was developed for the detection of crack in eggshell.The system consists of a product support,a light exciting mechanism,a microphone,signal amplifiers,a personal computer (PC)and software to acquire and analyze the results.A schematic diagram of the system is presented in Fig.1.A pair of rolls made of hard rubber was used to support the eggs,and the shape of the support was focused to normal eggshell surfaces.The excitation set included an electromagnetic driver,an adjustable volt DC power and a light metallic stick.The total mass of the stick was 6g,and its length 6cm.The excitation force is an important factor that affects the magnitude and width of the pulse.The adjustable volt DC power was used to control the excitation force.Based on previous test,the voltage of excitation was set at 30V.In this case,optimal signals were achieved without instrumentation overload.The impacting position was close to the crack in the cracked eggshells,which was placed randomly among intact eggshells.Data acquisition and analysisResponse signals obtained from the microphone were amplified,filtered and captured by a 16-bit data acquisition card.The program of data acquisition was compiled based on LabVIEW8.2software(National Instruments,USA)that allows a fast acquisition and processing of the response signal.The sampling rate was 22.05kHz.The time signal was transformed to a frequency signal by using a 512-point fast Fourier (FFT)transformation.The linear frequency spectrum accepted was transformed to a power spectrum.A band-pass filter was used to preserve the information of the frequency band between 1,000and 8,000Hz,because the features of response signals were legible in this frequency band and the signal-to-noise here was also favorable.Brief introduction of support vector data description (SVDD)SVDD is inspired by the idea of SVM [14,15].It is a method of data domain description also calledone-classFig.1Eggshell crackmeasurement system based on acoustic resonance analysisclassification.The basic idea of SVDD is to envelop samples or objects within a high-dimensional space with the volume as small as possible byfitting a hypersphere around the samples.The sketch map in two dimensions of SVDD is shown in Fig.2.By introducing kernels,this inflexible model becomes much more powerful and can give reliable results when a suitable kernel is used[16]. The problem of SVDD is tofind center a and radius R, which have the minimum volume of hypersphere contain-ing all samples X i.For a data set containing i normal data objects,when one or a few very remote objects are in it,a very large sphere is obtained,which will not represent the data very well.Therefore,we allow for some data points outside the sphere and introduce slack variable n i.As a result,the minimization problem can be denoted in thefollowing form:min LðRÞ¼R2þCX Ni¼1n i;s:t x iÀak k2R2þn i;n i!0ði¼1;2;...;NÞ;9>>>>>=>>>>>;ð1Þwhere the variable C gives the trade-off between simplicity (volume of the sphere)and the number of errors(number of target objects rejected).The above problem is usually solved by introducing Lagrange multipliers and can be transformed into maximizing the following function L with respect to the Lagrange multipliers.For an object x,we definef2ðxÞ¼xÀak k2¼ðxÁxÞÀ2X Ni¼1a iðzÁx iÞþX Ni¼1X Nj¼1a i a jðx iÁx jÞ:ð2ÞThe test objects x is accepted when the distance is smaller than the radius.These objects are called the support objects of the description or the SVs.Objects lying outside the sphere are also called bounded support vectors(BSVs). When a sphere is not always a goodfit for the boundary of data distribution,the inner product(x,y)is generalized by a kernel function k x;yðÞ¼/xðÞ;/yðÞf g;where a mapping/ of the data to a new feature space is applied.With such mapping,Eq.(2)will then becomeL¼P Ni¼1a i kðx i;x iÞÀP Ni¼1P Nj¼1a i a j kðx i;x jÞ;s:t0a i C;P Ni¼1a i¼1and a¼Pia i/ðx iÞ:9>>>=>>>;ð3ÞIn brief,SVDDfirst maps the data which are not linearly separable into a high-dimensional feature space and then describe the data by the maximal margin hypersphere.SoftwareAll data-processing algorithms were implemented with the statistical software Matlab7.1(Mathworks,USA)under Windows XP.SVDD Matlab codes were downloaded from http://www-ict.ewi.tudelft.nl/*davidt/dd_tools.html free of charge.Result and discussionResponse signalsSince the acoustic response was an instantaneous impulse, it was difficult to discriminate between the different response signals of cracked and intact eggs in the time domain.The time domain signals were transformed by FFT to frequency domain signals for the next analysis.Typical power spectra of intact egg and cracked egg are shown in Fig.3,and the areas under the spectral envelope for the intact eggs were smaller than that of the cracked eggs.For the intact eggs,the peak frequencies were prominent, generally found in the middle place(3,500–5,000Hz).In contrast,the peak frequencies of cracked eggs were dis-perse and not prominent.Adaptive RLSfilteringSince the detection of cracked eggshells is based on acoustic response measurement,it is vulnerably interfered by the surrounding noise.This fact is reinforced by the much damped behaviors of agro-products[17].Therefore, response signal should be processed to remove noise in further analysis.Adaptive interference canceling is a standard approach to remove environmental noise[18,19].The RLS is a popular algorithm in thefield of adaptive signal processing. In adaptive RLSfiltering,the coefficients are adjusted from sample to sample to minimize the mean square error(MSE) between a measured noisy scalar signal and itsmodeledvalue from the filter [20,21].A scalar,real output signal,y k ,is measured at the discrete time k ,in response to a set of scalar input signals X k ði Þ;i ¼1;2;...;n ;where n is an arbitrary number of filter taps.For this research,n is set to the number of degrees of freedom to ensure conformity of the resulting filter matrices.The input and the output sig-nals are related by the simple regression model:y k ¼X n À1i ¼0w ði ÞÁx k ði Þþe k :ð4Þwhere e k represents measurement error and w (i )represents the proportion that is contained in the primary scalar signal y k .The implementation of the RLS algorithm is optimized by exploiting the inversion matrix lemma and provides fast convergence and small error rates [22].System identification of a 32-coefficient FIR filter combined with adaptive RLS filtering was used to process the signals.The forgetting factor was 1,and the vector of initial filter coefficients was 0.Figure 4shows the fre-quency signals before and after adaptive RLS filtering.Variable selectionBased on the differences of frequency domain response signals from intact and cracked eggs,five characteristic descriptors were extracted from the response frequency signals as the inputs of the discrimination model.These are shown in Table 1.Parameter optimization in SVDD modelThe basic concept of SVDD is to map nonlinearly the original data X into a higher-dimensional feature space.The transformation into a higher-dimensional space is implemented by a kernel function [23].So,selection of kernel function has a high influence on the performance of the SVDD model.Several kernel functions have been proposed for the SVDD classifier.Not all kernel functions are equally useful for the SVDD.It has been demonstrated that Gaussian kernel results in tighter description and gives a good performance under general smoothness assumptions [24].Thus,Gaussian kernel was adopted in this study.To obtain a good performance,the regularization parameter C and the kernel function r have to be opti-mized.Parameter C determines the trade-off between minimizing the training error and minimizing model complexity.By using Gaussian kernel,the data description transforms from a solid hyper-sphere to a Parzen density estimator.An appropriate selection with width parameter r of Gaussian kernel is important to the density estimation of target objects.There is no systematic methodology for the optimization of these parameters.In this study,the procedure of opti-mization was carried out in two search steps.First,a comparatively large step length was attempted to search optimal value of parameters.The favorable results of the model were found with values of C between 0.005and 0.1,and values of r between 10and 500.Therefore,a much smaller step length was employed for further searching these parameters.In the second search step,50parameter r values with the step of 10(r =10,20–500)and 20parameter C values with the step of 0.005(C =0.005,0.01–1)were tested simultaneously in the building model.Identification results of SVDD model influenced by values of r and C are shown in Fig.5.The optimal model was achieved when r was equal to 420and C was equal to 0.085or 0.09.Here,the identification rates of intactandFig.3Typical response frequency signal ofeggsFig.4Frequency signals before and after adaptive RLS filteringcracked eggs were both 90%in the prediction set.Fur-thermore,it was found that the performance of the SVDD model could not be improved by smaller search parison of discrimination modelsConventional two-class linear discrimination analysis (LDA)model and SVM model were used comparatively to classify intact and cracked eggs.Gaussian kernel was recommended as the kernel function of the SVM model.Parameters of SVM model were also optimized as in SVDD.Table 2shows the optimal results from three dis-crimination models in the prediction set.Identification rates of intact eggs were both 100%in the LDA and SVM models,but 50and 35%for cracked eggs,respectively.In other words,at least 50%of cracked eggs could not be identified in conventional discrimination model.However,detection of cracked eggs is the task we focus on.The identification rates of intact and cracked eggs were both 90%in the SVDD pared with conventional two-class discrimination models,SVDD model showed its superior performance in the discrimination of cracked eggs.LDA is a linear and parametric method with discrimi-nating character.In terms of a set of discriminant functions,the classifier is said to assign an unknown example X to thecorresponding class [25].In the case of conventional LDA classification,the ultimate decision function is based on sufficient information support from two-class training samples.In general,such classification does not pay enough attention to the samples in minority class in building model.It is possible to obtain an inaccurate estimation of the centroid between the two classes.Conventional LDA clas-sification always poorly describes the specific class with scarce training samples.Therefore,it is often unpractical to solve the classification problem using tradition LDA clas-sifier,in case of imbalanced number in training samples.The basic concept of SVM is to map the original data X into a higher-dimensional feature space and find the ‘optimal’hyperplane boundary to separate the two classes [26].In SVM classification,the ‘optimal’boundary is defined as the most distant hyperplane from both sets,which is also called the ‘middle point’between the clas-sification sets.This boundary is expected to be the optimal classification of the sets,since it is the best isolated from the two sets [27].The margin is the minimal distance from the separating hyperplane to the closest data points [28].In general,when the information support from both positive and negative training sets are sufficient and equal,an appropriate separating hyperplane can be obtained.How-ever,when the samples from one class are insufficient to support the separating hyperplane,it will result in the hyperplane being excessively close to this class.As a result,most of the unknown sets may be recognized as the other class.Therefore,compared with other discrimination models,SVM showed poorest performance in discrimi-nating cracked eggs.Differing from conventional classification-based app-roach,SVDD is an approach for one-class classification.ItTable 1Frequencycharacteristics selection and expressionSome Low frequency band:1,000–3,720Hz,Middlefrequency band:3,720–7,440HzVariables Resonance frequency characteristics Expression X1Value of the area of amplitudeX 1¼P512i ¼0PiX2Value of the standard deviation of amplitude X 2¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðPi ÀP Þq=nX3Value of the frequency band of maximum amplitude X 3¼Index max ðPi ÞX4Mean of top three frequency amplitude values X 4¼Max 1:3ðPi Þ=3X5Ratio of amplitude values of middle frequency bands to low frequency bandX 5¼P 200i ¼1Pi P 400i ¼201Pi200Fig.5Identification rates of SVDD models with different values ofparameter r and CTable 2Comparison of results from three discrimination models ModelIdentification rates in the prediction set (%)Intact eggsCracked eggs LDA 10050SVM 10035SVDD9090focuses mainly on normal or target objects.SVDD can handle cases with only a few outlier objects.The advantage of SVDD is that the target class can be any one of two training classes.The selection of the target class depends on the reliability of the information provided from training samples.In general,the class containing more samples may provide sufficient information,and it can be selected as target class[29].Furthermore,SVDD can adapt to the real shape of samples andfindflexible boundary with a mini-mum volume by introducing kernel function.The boundary is described by a few training objects,the support vectors. It is possible to replace normal inner products with kernel functions and obtain moreflexible data descriptions[30]. Width parameter r can be set to give the desired number of support vectors.In addition,extra data on the form of outlier objects can be helpful to improve the performance of the SVDD model.ConclusionsDetection of crack in eggshell based on acoustic impulse resonance was attempted in this work.The SVDD method was employed for solving classification problem where the samples of cracked eggs were not sufficient.The results indicated that detection of crack in eggshell based on the acoustic impulse resonance was feasible,and the SVDD model showed its superior performance in contrast to conventional two-class discrimination models.It can be concluded that SVDD is an excellent method of classifi-cation problem with imbalanced numbers.It is a promising method that uses acoustic resonance technique combined with SVDD to detect cracked eggs.Some relative ideas would be attempted for further improvement of the per-formance of SVDD model in our future work,such as follows:(1)introduce new kernel functions,which can help to obtain a moreflexible boundary;(2)try more methods for selection of parameters to obtain the optimal ones,since parameters of kernel functions are closely related to the tightness of the constructed boundary and the target rejection rate,and appropriate parameters are important to improve the performance of SVDD models;(3)investigate the contribution of abnormal targets to the calibration model and develop a robust model,which has an excellent ability to deal with abnormal targets.Acknowledgments This work is a part of the National Key Tech-nology R&D Program of China(Grant No.2006BAD11A12).We are grateful to the Web site http://www-ict.ewi.tudelft.nl/*davidt/ dd_tools.html,where we downloaded SVDD Matlab codes free of charge.References1.Lin J,Puri VM,Anantheswaran RC(1995)Trans ASAE38(6):1769–17762.Cho HK,Choi WK,Paek JK(2000)Trans ASAE43(6):1921–19263.De Ketelaere B,Coucke P,De Baerdemaeker J(2000)J Agr EngRes76:157–1634.Coucke P,De Ketelaere B,De Baerdemaeker J(2003)J SoundVib266:711–7215.Wang J,Jiang RS(2005)Eur Food Res Technol221:214–2206.Jindal VK,Sritham E(2003)ASAE Annual International Meet-ing,USA7.Tax DMJ,Duin RPW(1999)Pattern Recognit Lett20:1191–11998.Pan Y,Chen J,Guo L(2009)Mech Syst Signal Process23:669–6819.Lee SW,Park JY,Lee SW(2006)Patten Recognit39:1809–181210.Podsiadlo P,Stachowiak GW(2006)Tribol Int39:1624–163311.Sanchez-Hernandeza C,Boyd DS,Foody GM(2007)Ecol Inf2:83–8812.Liu YH,Lin SH,Hsueh YL,Lee MJ(2009)Expert Syst Appl36:1978–199813.Cho HW(2009)Expert Syst Appl36:434–44114.Tax DMJ,Duin RPW(2001)J Mach Learn Res2:155–17315.Tax DMJ,Duin RPW(2004)Mach Learn54:45–6616.Guo SM,Chen LC,Tsai JHS(2009)Pattern Recognit42:77–8317.De Ketelaere B,Maertens K,De Baerdemaeker J(2004)MathComput Simul65:59–6718.Adall T,Ardalan SH(1999)Comput Elect Eng25:1–1619.Madsen AH(2000)Signal Process80:1489–150020.Chase JG,Begoc V,Barroso LR(2005)Comput Struct83:639–64721.Wang X,Feng GZ(2009)Signal Process89:181–18622.Djigan VI(2006)Signal Process86:776–79123.Bu HG,Wang J,Huang XB(2009)Eng Appl Artif Intell22:224–23524.Tao Q,Wu GW,Wang J(2005)Pattern Recognit38:1071–107725.Xie JS,Qiu ZD(2007)Pattern Recognit40:557–56226.Devos O,Ruckebusch C,Durand A,Duponchel L,Huvenne JP(2009)Chemom Intell Lab Syst96:27–3327.Liu X,Lu WC,Jin SL,Li YW,Chen NY(2006)Chemom IntellLab Syst82:8–1428.Chen QS,Zhao JW,Fang CH,Wang DM(2007)SpectrochimActa Pt A Mol Biomol Spectrosc66:568–57429.Huang WL,Jiao LC(2008)Prog Nat Sci18:455–46130.Foody GM,Mathur A,Sanchez-Hernandez C,Boyd DS(2006)Remote Sens Environ104:1–14。
毕业设计外文翻译_英文版

A Design and Implementation of Active NetworkSocket ProgrammingK.L. Eddie Law, Roy LeungThe Edward S. Rogers Sr. Department of Electrical and Computer EngineeringUniversity of TorontoToronto, Canadaeddie@, roy.leung@utoronto.caAbstract—The concept of programmable nodes and active networks introduces programmability into communication networks. Code and data can be sent and modified on their ways to destinations. Recently, various research groups have designed and implemented their own design platforms. Each design has its own benefits and drawbacks. Moreover, there exists an interoperability problem among platforms. As a result, we introduce a concept that is similar to the network socket programming. We intentionally establish a set of simple interfaces for programming active applications. This set of interfaces, known as Active Network Socket Programming (ANSP), will be working on top of all other execution environments in future. Therefore, the ANSP offers a concept that is similar to “write once, run everywhere.” It is an open programming model that active applications can work on all execution environments. It solves the heterogeneity within active networks. This is especially useful when active applications need to access all regions within a heterogeneous network to deploy special service at critical points or to monitor the performance of the entire networks. Instead of introducing a new platform, our approach provides a thin, transparent layer on top of existing environments that can be easily installed for all active applications.Keywords-active networks; application programming interface; active network socket programming;I. I NTRODUCTIONIn 1990, Clark and Tennenhouse [1] proposed a design framework for introducing new network protocols for the Internet. Since the publication of that position paper, active network design framework [2, 3, 10] has slowly taken shape in the late 1990s. The active network paradigm allows program code and data to be delivered simultaneously on the Internet. Moreover, they may get executed and modified on their ways to their destinations. At the moment, there is a global active network backbone, the ABone, for experiments on active networks. Apart from the immaturity of the executing platform, the primary hindrance on the deployment of active networks on the Internet is more on the commercially related issues. For example, a vendor may hesitate to allow network routers to run some unknown programs that may affect their expected routing performance. As a result, alternatives were proposed to allow active network concept to operate on the Internet, such as the application layer active networking (ALAN) project [4] from the European research community. In the ALAN project, there are active server systems located at different places in the networks and active applications are allowed to run in these servers at the application layer. Another potential approach from the network service provider is to offer active network service as the premium service class in the networks. This service class should provide the best Quality of Service (QoS), and allow the access of computing facility in routers. With this approach, the network service providers can create a new source of income.The research in active networks has been progressing steadily. Since active networks introduce programmability on the Internet, appropriate executing platforms for the active applications to execute should be established. These operating platforms are known as execution environments (EEs) and a few of them have been created, e.g., the Active Signaling Protocol (ASP) [12] and the Active Network Transport System (ANTS) [11]. Hence, different active applications can be implemented to test the active networking concept.With these EEs, some experiments have been carried out to examine the active network concept, for example, the mobile networks [5], web proxies [6], and multicast routers [7]. Active networks introduce a lot of program flexibility and extensibility in networks. Several research groups have proposed various designs of execution environments to offer network computation within routers. Their performance and potential benefits to existing infrastructure are being evaluated [8, 9]. Unfortunately, they seldom concern the interoperability problems when the active networks consist of multiple execution environments. For example, there are three EEs in ABone. Active applications written for one particular EE cannot be operated on other platforms. This introduces another problem of resources partitioning for different EEs to operate. Moreover, there are always some critical network applications that need to run under all network routers, such as collecting information and deploying service at critical points to monitor the networks.In this paper, a framework known as Active Network Socket Programming (ANSP) model is proposed to work with all EEs. It offers the following primary objectives.• One single programming interface is introduced for writing active applications.• Since ANSP offers the programming interface, the design of EE can be made independent of the ANSP.This enables a transparency in developing andenhancing future execution environments.• ANSP addresses the interoperability issues among different execution environments.• Through the design of ANSP, the pros and cons of different EEs will be gained. This may help design abetter EE with improved performance in future.The primary objective of the ANSP is to enable all active applications that are written in ANSP can operate in the ABone testbed . While the proposed ANSP framework is essential in unifying the network environments, we believe that the availability of different environments is beneficial in the development of a better execution environment in future. ANSP is not intended to replace all existing environments, but to enable the studies of new network services which are orthogonal to the designs of execution environments. Therefore, ANSP is designed to be a thin and transparent layer on top of all execution environments. Currently, its deployment relies on automatic code loading with the underlying environments. As a result, the deployment of ANSP at a router is optional and does not require any change to the execution environments.II. D ESIGN I SSUES ON ANSPThe ANSP unifies existing programming interfaces among all EEs. Conceptually, the design of ANSP is similar to the middleware design that offers proper translation mechanisms to different EEs. The provisioning of a unified interface is only one part of the whole ANSP platform. There are many other issues that need to be considered. Apart from translating a set of programming interfaces to other executable calls in different EEs, there are other design issues that should be covered, e.g., • a unified thread library handles thread operations regardless of the thread libraries used in the EEs;• a global soft-store allows information sharing among capsules that may execute over different environmentsat a given router;• a unified addressing scheme used across different environments; more importantly, a routing informationexchange mechanism should be designed across EEs toobtain a global view of the unified networks;• a programming model that should be independent to any programming languages in active networks;• and finally, a translation mechanism to hide the heterogeneity of capsule header structures.A. Heterogeneity in programming modelEach execution environment provides various abstractions for its services and resources in the form of program calls. The model consists of a set of well-defined components, each of them has its own programming interfaces. For the abstractions, capsule-based programming model [10] is the most popular design in active networks. It is used in ANTS [11] and ASP [12], and they are being supported in ABone. Although they are developed based on the same capsule model, their respective components and interfaces are different. Therefore, programs written in one EE cannot run in anther EE. The conceptual views of the programming models in ANTS and ASP are shown in Figure 1.There are three distinct components in ANTS: application, capsule, and execution environment. There exist user interfaces for the active applications at only the source and destination routers. Then the users can specify their customized actions to the networks. According to the program function, the applications send one or more capsules to carry out the operations. Both applications and capsules operate on top of an execution environment that exports an interface to its internal programming resources. Capsule executes its program at each router it has visited. When it arrives at its destination, the application at destination may either reply it with another capsule or presents this arrival event to the user. One drawback with ANTS is that it only allows “bootstrap” application.Figure 1. Programming Models in ASP and ANTS.In contrast, ASP does not limit its users to run “bootstrap” applications. Its program interfaces are different from ANTS, but there are also has three components in ASP: application client, environment, and AAContext. The application client can run on active or non-active host. It can start an active application by simply sending a request message to the EE. The client presents information to users and allows its users to trigger actions at a nearby active router. AAContext is the core of the network service and its specification is divided into two parts. One part specifies its actions at its source and destination routers. Its role is similar to that of the application in ANTS, except that it does not provide a direct interface with the user. The other part defines its actions when it runs inside the active networks and it is similar to the functional behaviors of a capsule in ANTS.In order to deal with the heterogeneity of these two models, ANSP needs to introduce a new set of programming interfaces and map its interfaces and execution model to those within the routers’ EEs.B. Unified Thread LibraryEach execution environment must ensure the isolation of instance executions, so they do not affect each other or accessThe authors appreciate the Nortel Institute for Telecommunications (NIT) at the University of Toronto to allow them to access the computing facilitiesothers’ information. There are various ways to enforce the access control. One simple way is to have one virtual machine for one instance of active applications. This relies on the security design in the virtual machines to isolate services. ANTS is one example that is using this method. Nevertheless, the use of multiple virtual machines requires relatively large amount of resources and may be inefficient in some cases. Therefore, certain environments, such as ASP, allow network services to run within a virtual machine but restrict the use of their services to a limited set of libraries in their packages. For instance, ASP provides its thread library to enforce access control. Because of the differences in these types of thread mechanism, ANSP devises a new thread library to allow uniform accesses to different thread mechanisms.C. Soft-StoreSoft-store allows capsule to insert and retrieve information at a router, thus allowing more than one capsules to exchange information within a network. However, problem arises when a network service can execute under different environments within a router. The problem occurs especially when a network service inserts its soft-store information in one environment and retrieves its data at a later time in another environment at the same router. Due to the fact that execution environments are not allowed to exchange information, the network service cannot retrieve its previous data. Therefore, our ANSP framework needs to take into account of this problem and provides soft-store mechanism that allows universal access of its data at each router.D. Global View of a Unified NetworkWhen an active application is written with ANSP, it can execute on different environment seamlessly. The previously smaller and partitioned networks based on different EEs can now be merging into one large active network. It is then necessary to advise the network topology across the networks. However, different execution environments have different addressing schemes and proprietary routing protocols. In order to merge these partitions together, ANSP must provide a new unified addressing scheme. This new scheme should be interpretable by any environments through appropriate translations with the ANSP. Upon defining the new addressing scheme, a new routing protocol should be designed to operate among environments to exchange topology information. This allows each environment in a network to have a complete view of its network topology.E. Language-Independent ModelExecution environment can be programmed in any programming language. One of the most commonly used languages is Java [13] due to its dynamic code loading capability. In fact, both ANTS and ASP are developed in Java. Nevertheless, the active network architecture shown in Figure 2 does not restrict the use of additional environments that are developed in other languages. For instance, the active network daemon, anted, in Abone provides a workspace to execute multiple execution environments within a router. PLAN, for example, is implemented in Ocaml that will be deployable on ABone in future. Although the current active network is designed to deploy multiple environments that can be in any programming languages, there lacks the tool to allow active applications to run seamlessly upon these environments. Hence, one of the issues that ANSP needs to address is to design a programming model that can work with different programming languages. Although our current prototype only considers ANTS and ASP in its design, PLAN will be the next target to address the programming language issue and to improve the design of ANSP.Figure 2. ANSP Framework Model.F. Heterogeneity of Capsule Header StructureThe structures of the capsule headers are different in different EEs. They carries capsule-related information, for example, the capsule types, sources and destinations. This information is important when certain decision needs to be made within its target environment. A unified model should allow its program code to be executed on different environments. However, the capsule header prevents different environments to interpret its information successfully. Therefore, ANSP should carry out appropriate translation to the header information before the target environment receives this capsule.III. ANSP P ROGRAMMING M ODELWe have outlined the design issues encountered with the ANSP. In the following, the design of the programming model in ANSP will be discussed. This proposed framework provides a set of unified programming interfaces that allows active applications to work on all execution environments. The framework is shown in Figure 2. It is composed of two layers integrated within the active network architecture. These two layers can operate independently without the other layer. The upper layer provides a unified programming model to active applications. The lower layer provides appropriate translation procedure to the ANSP applications when it is processed by different environments. This service is necessary because each environment has its own header definition.The ANSP framework provides a set of programming calls which are abstractions of ANSP services and resources. A capsule-based model is used for ANSP, and it is currently extended to map to other capsule-based models used in ANTSand ASP. The mapping possibility to other models remains as our future works. Hence, the mapping technique in ANSP allows any ANSP applications to access the same programming resources in different environments through a single set of interfaces. The mapping has to be done in a consistent and transparent manner. Therefore, the ANSP appears as an execution environment that provides a complete set of functionalities to active applications. While in fact, it is an overlay structure that makes use of the services provided from the underlying environments. In the following, the high-level functional descriptions of the ANSP model are described. Then, the implementations will be discussed. The ANSP programming model is based upon the interactions between four components: application client , application stub , capsule , and active service base.Figure 3. Information Flow with the ANSP.•Application Client : In a typical scenario, an active application requires some means to present information to its users, e.g., the state of the networks. A graphical user interface (GUI) is designed to operate with the application client if the ANSP runs on a non-active host.•Application Stub : When an application starts, it activates the application client to create a new instance of application stub at its near-by active node. There are two responsibilities for the application stub. One of them is to receive users’ instructions from the application client. Another one is to receive incoming capsules from networks and to perform appropriate actions. Typically, there are two types of actions, thatare, to reply or relay in capsules through the networks, or to notify the users regarding the incoming capsule. •Capsule : An active application may contain several capsule types. Each of them carries program code (also referred to as forwarding routine). Since the application defines a protocol to specify the interactions among capsules as well as the application stubs. Every capsule executes its forwarding routine at each router it visits along the path between the source and destination.•Active Service Base : An active service base is designed to export routers’ environments’ services and execute program calls from application stubs and capsules from different EEs. The base is loaded automatically at each router whenever a capsule arrives.The interactions among components within ANSP are shown in Figure 3. The designs of some key components in the ANSP will be discussed in the following subsections. A. Capsule (ANSPCapsule)ANSPXdr decode () ANSPXdr encode () int length ()Boolean execute ()New types of capsule are created by extending the abstract class ANSPCapsule . New extensions are required to define their own forwarding routines as well as their serialization procedures. These methods are indicated below:The execution of a capsule in ANSP is listed below. It is similar to the process in ANTS.1. A capsule is in serial binary representation before it issent to the network. When an active router receives a byte sequence, it invokes decode() to convert the sequence into a capsule. 2. The router invokes the forwarding routine of thecapsule, execute(). 3. When the capsule has finished its job and forwardsitself to its next hop by calling send(), this call implicitly invokes encode() to convert the capsule into a new serial byte representation. length() isused inside the call of encode() to determine the length of the resulting byte sequence. ANSP provides a XDR library called ANSPXdr to ease the jobs of encoding and decoding.B. Active Service Base (ANSPBase)In an active node, the Active Service Base provides a unified interface to export the available resources in EEs for the rest of the ANSP components. The services may include thread management, node query, and soft-store operation, as shown in Table 1.TABLE I. ACTIVE SERVICE BASE FUNCTION CALLSFunction Definition Descriptionboolean send (Capsule, Address) Transmit a capsule towards its destination using the routing table of theunderlying environment.ANSPAddress getLocalHost () Return address of the local host as an ANSPAddress structure. This isuseful when a capsule wants to check its current location.boolean isLocal (ANSPAddress) Return true if its input argument matches the local host’s address andreturn false otherwise.createThread () Create a new thread that is a class ofANSPThreadInterface (discussed later in Section VIA “Unified Thread Abstraction”).putSStore (key, Object) Object getSStore (key) removeSStore (key)The soft-store operations are provided by putSStore(), getSSTore(), and removeSStore(), and they put, retrieve, and remove data respectively. forName (PathName) Supported in ANSP to retrieve a classobject corresponding to the given path name in its argument. This code retrieval may rely on the code loading mechanism in the environment whennecessary.C. Application Client (ANSPClient)boolean start (args[])boolean start (args[],runningEEs) boolean start (args[],startClient)boolean start (args[],startClient, runningEE)Application Client is an interface between users and the nearby active source router. It does the following responsibilities.1. Code registration: It may be necessary to specify thelocation and name of the application code in some execution environments, e.g., ANTS. 2. Application initialization: It includes selecting anexecution environment to execute the application among those are available at the source router. Each active application can create an application client instance by extending the abstract class, ANSPClient . The extension inherits a method, start(), to automatically handle both the registration and initialization processes. All overloaded versions of start() accept a list of arguments, args , that are passed to the application stub during its initialization. An optional argument called runningEEs allows an application client to select a particular set of environment variables, specified by a list of standardized numerical environment ID, the ANEP ID, to perform code registration. If this argument is not specified, the default setting can only include ANTS and ASP. D. Application Stub (ANSPApplication)receive (ANSPCapsule)Application stubs reside at the source and destination routers to initialize the ANSP application after the application clients complete the initialization and registration processes. It is responsible for receiving and serving capsules from the networks as well as actions requested from the clients. A new instance is created by extending the application client abstract class, ANSPApplication . This extension includes the definition of a handling routine called receive(), which is invoked when a stub receives a new capsule.IV. ANSP E XAMPLE : T RACE -R OUTEA testbed has been created to verify the design correctnessof ANSP in heterogeneous environments. There are three types of router setting on this testbed:1. Router that contains ANTS and a ANSP daemonrunning on behalf of ASP; 2. Router that contains ASP and a ANSP daemon thatruns on behalf of ANTS; 3. Router that contains both ASP and ANTS.The prototype is written in Java [11] with a traceroute testing program. The program records the execution environments of all intermediate routers that it has visited between the source and destination. It also measures the RTT between them. Figure 4 shows the GUI from the application client, and it finds three execution environments along the path: ASP, ANTS, and ASP. The execution sequence of the traceroute program is shown in Figure 5.Figure 4. The GUI for the TRACEROUTE Program.The TraceCapsule program code is created byextending the ANSPCapsule abstract class. When execute() starts, it checks the Boolean value of returning to determine if it is returning from the destination. It is set to true if TraceCapsule is traveling back to the source router; otherwise it is false . When traveling towards the destination, TraceCapsule keeps track of the environments and addresses of the routers it has visited in two arrays, path and trace , respectively. When it arrives at a new router, it calls addHop() to append the router address and its environment to these two arrays. When it finally arrives at the destination, it sets returning to false and forwards itself back to the source by calling send().When it returns to source, it invokes deliverToApp() to deliver itself to the application stub that has been running at the source. TraceCapsule carries information in its data field through the networks by executing encode() and decode(), which encapsulates and de-capsulates its data using External Data Representation (XDR) respectively. The syntax of ANSP XDR follows the syntax of XDR library from ANTS. length() in TraceCapsule returns the data length, or it can be calculated by using the primitive types in the XDRlibrary.Figure 5. Flow of the TRACEROUTE Capsules.V. C ONCLUSIONSIn this paper, we present a new unified layered architecture for active networks. The new model is known as Active Network Socket Programming (ANSP). It allows each active application to be written once and run on multiple environments in active networks. Our experiments successfully verify the design of ANSP architecture, and it has been successfully deployed to work harmoniously with ANTS and ASP without making any changes to their architectures. In fact, the unified programming interface layer is light-weighted and can be dynamically deployable upon request.R EFERENCES[1] D.D. Clark, D.L. Tennenhouse, “Architectural Considerations for a NewGeneration of Protocols,” in Proc. ACM Sigcomm’90, pp.200-208, 1990. [2] D. Tennenhouse, J. M. Smith, W. D. Sicoskie, D. J. Wetherall, and G. J.Minden, “A survey of active network research,” IEEE Communications Magazine , pp. 80-86, Jan 1997.[3] D. Wetherall, U. Legedza, and J. Guttag, “Introducing new internetservices: Why and how,” IEEE Network Magazine, July/August 1998. [4] M. Fry, A. Ghosh, “Application Layer Active Networking,” in ComputerNetworks , Vol.31, No.7, pp.655-667, 1999.[5] K. W. Chin, “An Investigation into The Application of Active Networksto Mobile Computing Environments”, Curtin University of Technology, March 2000.[6] S. Bhattacharjee, K. L. Calvert, and E. W. Zegura, “Self OrganizingWide-Area Network Caches”, Proc. IEEE INFOCOM ’98, San Francisco, CA, 29 March-2 April 1998.[7] L. H. Leman, S. J. Garland, and D. L. Tennenhouse, “Active ReliableMulticast”, Proc. IEEE INFOCOM ’98, San Francisco, CA, 29 March-2 April 1998.[8] D. Descasper, G. Parulkar, B. Plattner, “A Scalable, High PerformanceActive Network Node”, In IEEE Network, January/February 1999.[9] E. L. Nygren, S. J. Garland, and M. F. Kaashoek, “PAN: a high-performance active network node supporting multiple mobile code system”, In the Proceedings of the 2nd IEEE Conference on Open Architectures and Network Programming (OpenArch ’99), March 1999. [10] D. L. Tennenhouse, and D. J. Wetherall. “Towards an Active NetworkArchitecture”, In Proceeding of Multimedia Computing and Networking , January 1996.[11] D. J. Wetherall, J. V. Guttag, D. L. Tennenhouse, “ANTS: A toolkit forBuilding and Dynamically Deploying Network Protocols”, Open Architectures and Network Programming, 1998 IEEE , 1998 , Page(s): 117 –129.[12] B. Braden, A. Cerpa, T. Faber, B. Lindell, G. Phillips, and J. Kann.“Introduction to the ASP Execution Environment”: /active-signal/ARP/index.html .[13] “The java language: A white paper,” Tech. Rep., Sun Microsystems,1998.。
毕业设计外文翻译英文加中文

A Comparison of Soft Start Mechanisms for Mining BeltConveyors1800 Washington Road Pittsburgh, PA 15241 Belt Conveyors are an important method for transportation of bulk materials in the mining industry. The control of the application of the starting torque from the belt drive system to the belt fabric affects the performance, life cost, and reliability of the conveyor. This paper examines applications of each starting method within the coal mining industry.INTRODUCTIONThe force required to move a belt conveyor must be transmitted by the drive pulley via friction between the drive pulley and the belt fabric. In order to transmit power there must be a difference in the belt tension as it approaches and leaves the drive pulley. These conditions are true for steady state running, starting, and stopping. Traditionally, belt designs are based on static calculations of running forces. Since starting and stopping are not examined in detail, safety factors are applied to static loadings (Harrison, 1987). This paper will primarily address the starting or acceleration duty of the conveyor. The belt designer must control starting acceleration to prevent excessive tension in the belt fabric and forces in the belt drive system (Suttees, 1986). High acceleration forces can adversely affect the belt fabric, belt splices, drive pulleys, idler pulleys, shafts, bearings, speed reducers, and couplings. Uncontrolled acceleration forces can cause belt conveyor system performance problems with vertical curves, excessive belt take-up movement, loss of drive pulley friction, spillage of materials, and festooning of the belt fabric. The belt designer is confronted with two problems, The belt drive system must produce a minimum torque powerful enough to start the conveyor, and controlled such that the acceleration forces are within safe limits. Smooth starting of the conveyor can be accomplished by the use of drive torque control equipment, either mechanical or electrical, or a combination of the two (CEM, 1979).SOFT START MECHANISM EVALUATION CRITERIONWhat is the best belt conveyor drive system? The answer depends on many variables. The best system is one that provides acceptable control for starting, running, and stopping at a reasonable cost and with high reliability (Lewdly and Sugarcane, 1978). Belt Drive System For the purposes of this paper we will assume that belt conveyors are almost always driven byelectrical prime movers (Goodyear Tire and Rubber, 1982). The belt "drive system" shall consist of multiple components including the electrical prime mover, the electrical motor starter with control system, the motor coupling, the speed reducer, the low speed coupling, the belt drive pulley, and the pulley brake or hold back (Cur, 1986). It is important that the belt designer examine the applicability of each system component to the particular application. For the purpose of this paper, we will assume that all drive system components are located in the fresh air, non-permissible, areas of the mine, or in non-hazardous, National Electrical Code, Article 500 explosion-proof, areas of the surface of the mine.Belt Drive Component Attributes SizeCertain drive components are available and practical in different size ranges. For this discussion, we will assume that belt drive systems range from fractional horsepower to multiples of thousands of horsepower. Small drive systems are often below 50 horsepower. Medium systems range from 50 to 1000 horsepower. Large systems can be considered above 1000 horsepower. Division of sizes into these groups is entirely arbitrary. Care must be taken to resist the temptation to over motor or under motor a belt flight to enhance standardization. An over motored drive results in poor efficiency and the potential for high torques, while an under motored drive could result in destructive overspending on regeneration, or overheating with shortened motor life (Lords, et al., 1978).Torque ControlBelt designers try to limit the starting torque to no more than 150% of the running torque (CEMA, 1979; Goodyear, 1982). The limit on the applied starting torque is often the limit of rating of the belt carcass, belt splice, pulley lagging, or shaft deflections. On larger belts and belts with optimized sized components, torque limits of 110% through 125% are common (Elberton, 1986). In addition to a torque limit, the belt starter may be required to limit torque increments that would stretch belting and cause traveling waves. An ideal starting control system would apply a pretension torque to the belt at rest up to the point of breakaway, or movement of the entire belt, then a torque equal to the movement requirements of the belt with load plus a constant torque to accelerate the inertia of the system components from rest to final running speed. This would minimize system transient forces and belt stretch (Shultz, 1992). Different drive systems exhibit varying ability to control the application of torques to the belt at rest and at different speeds. Also, the conveyor itself exhibits two extremes of loading. An empty belt normally presents the smallest required torque for breakaway and acceleration, while a fully loaded belt presents the highest required torque. A mining drive system must be capable of scaling the applied torque from a 2/1 ratio for a horizontal simple belt arrangement, to a 10/1 ranges for an inclined or complex belt profile.Thermal RatingDuring starting and running, each drive system may dissipate waste heat. The waste heat may be liberated in the electrical motor, the electrical controls,, the couplings, the speed reducer, or the belt braking system. The thermal load of each start Is dependent on the amount of belt load and the duration of the start. The designer must fulfill the application requirements for repeated starts after running the conveyor at full load. Typical mining belt starting duties vary from 3 to 10 starts per hour equally spaced, or 2 to 4 starts in succession. Repeated starting may require the dreading or over sizing of system components. There is a direct relationship between thermal rating for repeated starts and costs. Variable Speed. Some belt drive systems are suitable for controlling the starting torque and speed, but only run at constant speed. Some belt applications would require a drive system capable of running for extended periods at less than full speed. This is useful when the drive load must be shared with other drives, the belt is used as a process feeder for rate control of the conveyed material, the belt speed is optimized for the haulage rate, the belt is used at slower speeds to transport men or materials, or the belt is run a slow inspection or inching speed for maintenance purposes (Hager, 1991). The variable speed belt drive will require a control system based on some algorithm to regulate operating speed. Regeneration or Overhauling Load. Some belt profiles present the potential for overhauling loads where the belt system supplies energy to the drive system. Not all drive systems have the ability to accept regenerated energy from the load. Some drives can accept energy from the load and return it to the power line for use by other loads. Other drives accept energy from the load and dissipate it into designated dynamic or mechanical braking elements. Some belt profiles switch from motoring to regeneration during operation. Can the drive system accept regenerated energy of a certain magnitude for the application? Does the drive system have to control or modulate the amount of retarding force during overhauling? Does the overhauling occur when running and starting? Maintenance and Supporting Systems. Each drive system will require periodic preventative maintenance. Replaceable items would include motor brushes, bearings, brake pads, dissipation resistors, oils, and cooling water. If the drive system is conservatively engineered and operated, the lower stress on consumables will result in lower maintenance costs. Some drives require supporting systems such as circulating oil for lubrication, cooling air or water, environmental dust filtering, or computer instrumentation. The maintenance of the supporting systems can affect the reliability of the drive system.CostThe drive designer will examine the cost of each drive system. The total cost is the sum of the first capital cost to acquire the drive, the cost to install and commission the drive, thecost to operate the drive, and the cost to maintain the drive. The cost for power to operate the drive may vary widely with different locations. The designer strives to meet all system performance requirements at lowest total cost. Often more than one drive system may satisfy all system performance criterions at competitive costs.ComplexityThe preferred drive arrangement is the simplest, such as a single motor driving through a single head pulley.However,mechanical, economic,and functional requirements often necessitate the use of complex drives.The belt designer must balance the need for sophistication against the problems that accompany complex systems. Complex systems require additional design engineering for successful deployment. An often-overlooked cost in a complex system is the cost of training onsite personnel, or the cost of downtime as a result of insufficient training.SOFT START DRIVE CONTROL LOGICEach drive system will require a control system to regulate the starting mechanism. The most common type of control used on smaller to medium sized drives with simple profiles is termed "Open Loop Acceleration Control". In open loop, the control system is previously configured to sequence the starting mechanism in a prescribed manner, usually based on time. In open loop control, drive-operating parameters such as current, torque, or speed do not influence sequence operation. This method presumes that the control designer has adequately modeled drive system performance on the conveyor. For larger or more complex belts, "Closed Loop" or "Feedback" control may he utilized. In closed loop control, during starting, the control system monitors via sensors drive operating parameters such as current level of the motor, speed of the belt, or force on the belt, and modifies the starting sequence to control, limit, or optimize one or wore parameters. Closed loop control systems modify the starting applied force between an empty and fully loaded conveyor. The constants in the mathematical model related to the measured variable versus the system drive response are termed the tuning constants. These constants must be properly adjusted for successful application to each conveyor. The most common schemes for closed loop control of conveyor starts are tachometer feedback for speed control and load cell force or drive force feedback for torque control. On some complex systems, It is desirable to have the closed loop control system adjust itself for various encountered conveyor conditions. This is termed "Adaptive Control". These extremes can involve vast variations in loadings, temperature of the belting, location of the loading on the profile, or multiple drive options on the conveyor. There are three commonadaptive methods. The first involves decisions made before the start, or 'Restart Conditioning'. If the control system could know that the belt is empty, it would reduce initial force and lengthen the application of acceleration force to full speed. If the belt is loaded, the control system would apply pretension forces under stall for less time and supply sufficient torque to adequately accelerate the belt in a timely manner. Since the belt only became loaded during previous running by loading the drive, the average drive current can be sampled when running and retained in a first-in-first-out buffer memory that reflects the belt conveyance time. Then at shutdown the FIFO average may be use4 to precondition some open loop and closed loop set points for the next start. The second method involves decisions that are based on drive observations that occur during initial starting or "Motion Proving'. This usually involves a comparison In time of the drive current or force versus the belt speed. if the drive current or force required early in the sequence is low and motion is initiated, the belt must be unloaded. If the drive current or force required is high and motion is slow in starting, the conveyor must be loaded. This decision can be divided in zones and used to modify the middle and finish of the start sequence control. The third method involves a comparison of the belt speed versus time for this start against historical limits of belt acceleration, or 'Acceleration Envelope Monitoring'. At start, the belt speed is measured versus time. This is compared with two limiting belt speed curves that are retained in control system memory. The first curve profiles the empty belt when accelerated, and the second one the fully loaded belt. Thus, if the current speed versus time is lower than the loaded profile, it may indicate that the belt is overloaded, impeded, or drive malfunction. If the current speed versus time is higher than the empty profile, it may indicate a broken belt, coupling, or drive malfunction. In either case, the current start is aborted and an alarm issued.CONCLUSIONThe best belt starting system is one that provides acceptable performance under all belt load Conditions at a reasonable cost with high reliability. No one starting system meets all needs. The belt designer must define the starting system attributes that are required for each belt. In general, the AC induction motor with full voltage starting is confined to small belts with simple profiles. The AC induction motor with reduced voltage SCR starting is the base case mining starter for underground belts from small to medium sizes. With recent improvements, the AC motor with fixed fill fluid couplings is the base case for medium to large conveyors with simple profiles. The Wound Rotor Induction Motor drive is the traditional choice for medium to large belts with repeated starting duty or complex profilesthat require precise torque control. The DC motor drive, Variable Fill Hydrokinetic drive, and the Variable Mechanical Transmission drive compete for application on belts with extreme profiles or variable speed at running requirements. The choice is dependent on location environment, competitive price, operating energy losses, speed response, and user familiarity. AC Variable Frequency drive and Brush less DC applications are limited to small to medium sized belts that require precise speed control due to higher present costs and complexity. However, with continuing competitive and technical improvements, the use of synthesized waveform electronic drives will expand.REFERENCES[1]Michael L. Nave, P.E.1989.CONSOL Inc.煤矿业带式输送机几种软起动方式的比较1800 年华盛顿路匹兹堡, PA 15241带式运送机是采矿工业运输大批原料的重要方法。
毕业设计外文翻译-中文版

本科生毕业设计(论文)外文科技文献译文译文题目(外文题目)学院(系)Socket网络编程的设计与实现A Design andImplementation of Active Network Socket Programming机械与能源工程学院专学业号机械设计制造及其自动化071895学生姓名李杰林日期2012年5月27日指导教师签名日期摘要:编程节点和活跃网络的概念将可编程性引入到通信网络中,并且代码和数据可以在发送过程中进行修改。
最近,多个研究小组已经设计和实现了自己的设计平台。
每个设计都有其自己的优点和缺点,但是在不同平台之间都存在着互操作性问题。
因此,我们引入一个类似网络socket编程的概念。
我们建立一组针对应用程序进行编程的简单接口,这组被称为活跃网络Socket编程(ANSP)的接口,将在所有执行环境下工作。
因此,ANSP 提供一个类似于“一次性编写,无限制运行”的开放编程模型,它可以工作在所有的可执行环境下。
它解决了活跃网络中的异构性,当应用程序需要访问异构网络内的所有地区,在临界点部署特殊服务或监视整个网络的性能时显得相当重要。
我们的方案是在现有的环境中,所有应用程序可以很容易地安装上一个薄薄的透明层而不是引入一个新的平台。
关键词:活跃网络;应用程序编程接口;活跃网络socket编程1 导言1990年,为了在互联网上引入新的网络协议,克拉克和藤农豪斯[1]提出了一种新的设计框架。
自公布这一标志性文件,活跃网络设计框架[2,3,10]已经慢慢在20世纪90 年代末成形。
活跃网络允许程序代码和数据可以同时在互联网上提供积极的网络范式,此外,他们可以在传送到目的地的过程中得到执行和修改。
ABone作为一个全球性的骨干网络,开始进行活跃网络实验。
除执行平台的不成熟,商业上活跃网络在互联网上的部署也成为主要障碍。
例如,一个供应商可能不乐意让网络路由器运行一些可能影响其预期路由性能的未知程序,。
毕业设计外文文献翻译范文

毕业设计外文文献翻译专业学生姓名班级学号指导教师优集学院外文资料名称:Knowledge-Based Engineeri--ng Design Methodology外文资料出处:Int.J.Engng Ed.Vol.16.No.1附件: 1.外文资料翻译译文2.外文原文基于知识工程(KBE)设计方法D. E. CALKINS1.背景复杂系统的发展需要很多工程和管理方面的知识、决策,它要满足很多竞争性的要求。
设计被认为是决定产品最终形态、成本、可靠性、市场接受程度的首要因素。
高级别的工程设计和分析过程(概念设计阶段)特别重要,因为大多数的生命周期成本和整体系统的质量都在这个阶段。
产品成本的压缩最可能发生在产品设计的最初阶段。
整个生命周期阶段大约百分之七十的成本花费在概念设计阶段结束时,缩短设计周期的关键是缩短概念设计阶段,这样同时也减少了工程的重新设计工作量。
工程权衡过程中采用良好的估计和非正式的启发进行概念设计。
传统CAD工具对概念设计阶段的支持非常有限。
有必要,进行涉及多个学科的交流合作来快速进行设计分析(包括性能,成本,可靠性等)。
最后,必须能够管理大量的特定领域的知识。
解决方案是在概念设计阶段包含进更过资源,通过消除重新设计来缩短整个产品的时间。
所有这些因素都主张采取综合设计工具和环境,以在早期的综合设计阶段提供帮助。
这种集成设计工具能够使由不同学科的工程师、设计者在面对复杂的需求和约束时能够对设计意图达成共识。
那个设计工具可以让设计团队研究在更高级别上的更多配置细节。
问题就是架构一个设计工具,以满足所有这些要求。
2.虚拟(数字)原型模型现在需要是一种代表产品设计为得到一将允许一产品的早发展和评价的真实事实上原型的过程的方式。
虚拟样机将取代传统的物理样机,并允许设计工程师,研究“假设”的情况,同时反复更新他们的设计。
真正的虚拟原型,不仅代表形状和形式,即几何形状,它也代表如重量,材料,性能和制造工艺的非几何属性。
三峡大学科技学院外文翻译样式

三峡大学科技学院
毕业设计(论文)
译文
译文题目楷体2号
学生姓名:(宋体3号)学号:
专业:班级:
指导教师:
评阅教师:
完成日期二○○八年月日
(黑体3号)
说明
一、外文翻译是毕业论文的一个重要组成部分,各教学单位可根据学生的实际情况确定做外文翻译的时间,原则上要求在毕业前一学期末进行,完成时间最迟于毕业学期初。
二、通过文献查阅与翻译,进一步提高掌握使用外文的能力,熟悉本专业的几种主要外文书刊,了解毕业设计(论文)课题的国内外信息与动向。
内容要求如下: 1)阅读
每位学生在文献查阅环节中,必须阅读5~10万个印刷符号的与本专业本课题相关的外文文献资料,选择其主要的翻译1~2万个印刷符号。
(约3000汉字) 2)翻译
标题应真实的反映出翻译外文的主体内容或原外文标题内容,一般控制在20个字之内;外文翻译成中文的内容应能忠实的反映原文内容。
3)中外文正文格式按《三峡大学科技学院毕业论文印制规格的规定》执行。
格式要求:
外文翻译正文参照毕业设计(论文)正文格式要求。
毕业设计(论文)外文资料翻译

毕业设计(论文)外文资料翻译学院:艺术学院专业:环境设计姓名:学号:外文出处: The Swedish Country House附件: 1.外文资料翻译译文;2.外文原文附件1:外文资料翻译译文室内装饰简述一室内装饰设计要素1 空间要素空间的合理化并给人们以美的感受是设计基本的任务。
要勇于探索时代、技术赋于空间的新形象,不要拘泥于过去形成的空间形象。
2 色彩要求室内色彩除对视觉环境产生影响外,还直接影响人们的情绪、心理。
科学的用色有利于工作,有助于健康。
色彩处理得当既能符合功能要求又能取得美的效果。
室内色彩除了必须遵守一般的色彩规律外,还随着时代审美观的变化而有所不同。
3 光影要求人类喜爱大自然的美景,常常把阳光直接引入室内,以消除室内的黑暗感和封闭感,特别是顶光和柔和的散射光,使室内空间更为亲切自然。
光影的变换,使室内更加丰富多彩,给人以多种感受。
4 装饰要素室内整体空间中不可缺少的建筑构件、如柱子、墙面等,结合功能需要加以装饰,可共同构成完美的室内环境。
充分利用不同装饰材料的质地特征,可以获得千变完化和不同风格的室内艺术效果,同时还能体现地区的历史文化特征。
5 陈设要素室内家具、地毯、窗帘等,均为生活必需品,其造型往往具有陈设特征,大多数起着装饰作用。
实用和装饰二者应互相协调,求的功能和形式统一而有变化,使室内空间舒适得体,富有个性。
6 绿化要素室内设计中绿化以成为改善室内环境的重要手段。
室内移花栽木,利用绿化和小品以沟通室内外环境、扩大室内空间感及美化空间均起着积极作用。
二室内装饰设计的基本原则1 室内装饰设计要满足使用功能要求室内设计是以创造良好的室内空间环境为宗旨,使室内环境合理化、舒适化、科学化;要考虑人们的活动规律处理好空间关系,空间尺寸,空间比例;合理配置陈设与家具,妥善解决室内通风,采光与照明,注意室内色调的总体效果。
2 室内装饰设计要满足精神功能要求室内设计的精神就是要影响人们的情感,乃至影响人们的意志和行动,所以要研究人们的认识特征和规律;研究人的情感与意志;研究人和环境的相互作用。
毕业设计(论文)外文资料翻译(学生用)

毕业设计外文资料翻译学院:信息科学与工程学院专业:软件工程姓名: XXXXX学号: XXXXXXXXX外文出处: Think In Java (用外文写)附件: 1.外文资料翻译译文;2.外文原文。
附件1:外文资料翻译译文网络编程历史上的网络编程都倾向于困难、复杂,而且极易出错。
程序员必须掌握与网络有关的大量细节,有时甚至要对硬件有深刻的认识。
一般地,我们需要理解连网协议中不同的“层”(Layer)。
而且对于每个连网库,一般都包含了数量众多的函数,分别涉及信息块的连接、打包和拆包;这些块的来回运输;以及握手等等。
这是一项令人痛苦的工作。
但是,连网本身的概念并不是很难。
我们想获得位于其他地方某台机器上的信息,并把它们移到这儿;或者相反。
这与读写文件非常相似,只是文件存在于远程机器上,而且远程机器有权决定如何处理我们请求或者发送的数据。
Java最出色的一个地方就是它的“无痛苦连网”概念。
有关连网的基层细节已被尽可能地提取出去,并隐藏在JVM以及Java的本机安装系统里进行控制。
我们使用的编程模型是一个文件的模型;事实上,网络连接(一个“套接字”)已被封装到系统对象里,所以可象对其他数据流那样采用同样的方法调用。
除此以外,在我们处理另一个连网问题——同时控制多个网络连接——的时候,Java内建的多线程机制也是十分方便的。
本章将用一系列易懂的例子解释Java的连网支持。
15.1 机器的标识当然,为了分辨来自别处的一台机器,以及为了保证自己连接的是希望的那台机器,必须有一种机制能独一无二地标识出网络内的每台机器。
早期网络只解决了如何在本地网络环境中为机器提供唯一的名字。
但Java面向的是整个因特网,这要求用一种机制对来自世界各地的机器进行标识。
为达到这个目的,我们采用了IP(互联网地址)的概念。
IP以两种形式存在着:(1) 大家最熟悉的DNS(域名服务)形式。
我自己的域名是。
所以假定我在自己的域内有一台名为Opus的计算机,它的域名就可以是。
- 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。
普罗尼法:分析在消弧线圈保护网络的接地故障电流的有效工具Oinis CHAARI ;Patrick BASTARD ;Michel MEUNIER摘要:普罗尼法是一种信号模态分量估计技术。
每一个模态分量被定义为四个参数:频率、幅度、相位、和阻尼。
这种方法被用来分析在消弧线圈保护的20千伏网络接地故障电流。
普罗尼法参数中的一些电力系统特性方面的变化(在母线和所述故障,故障电阻和整个网络的电容电流之间的距离)呈现。
这些与普罗尼法参数有关的暂态故障电流可用于确定发生了什么样的故障,以及在哪里发生。
关键词:普罗尼法、信号分析、奇异值分解、消弧线圈保护网络、瞬时接地故障1.简介在所有可能发生在电力网络的故障,接地相故障是最常见的。
该接地故障的影响取决于中性点接地方式。
在本文中,我们把消弧线圈接地,其中一个消弧线圈被连接在电力系统中性点和地面之间。
对种电抗进行调整来匹配电网系统基波频率的值和零序电容的值。
结果是,接地故障期间流过故障点的电流不足以产生电弧。
然而,使用中性点经消弧线圈接地有利于系统稳定。
消弧线圈接地已大量用于欧洲和EDF(法国电力)并决定将其应用到整个法国电力系统。
这种中性点接地方式很少使用在中性点直接接地占优势的北美。
然而,也有一些系统使用中性点经消弧线圈接地更有利。
数字继电器应快速检测通过线路的任何故障信号并实时分析。
大多数时间,用50赫兹的分析来诊断故障。
但是,在一个补偿功率网络中,50赫兹的零序电流分量可能非常弱,因此很危险。
此外,在接地故障后的几毫秒,电流的暂态分量比稳态分量高的多。
首先,接地故障往往是由一系列的短时间瞬态持续形成的故障。
在故障暂态信号中的信息应该用于零序保护以提高保护的速度和准确性。
需要一种精确分析电流信号的工具分析当前的信号,并找到一个小数量的参数,定义波形。
几种算法已被用于在电信号的分析。
最广泛使用的方法是傅立叶变换用来分析信号的频谱分量,这种实时分析能产生一个完整的平稳信号。
这种方法符合严格的限制,分析信号时有强烈依赖于时间的特性。
此外,傅里叶变换对信号的非周期性分量非常敏感。
卡尔曼滤波理论的技术已被应用于从信号中除去干扰信号。
然而,卡尔曼滤波模拟非周期分量的能力有限。
在电力系统继电保护中提出了最小二乘线性拟合方法。
但线性拟合需要信号模拟的先进知识。
问题是,故障信号的暂态分量研究相当困难。
此外,很难建立故障本身和暂态信号的特性之间的关系。
因此,需要一个能分析非周期分量的方法,必须建立暂态信号处理方法。
普罗尼法便是其中之一。
这是近两个世纪以来加斯帕德里奇,男爵普罗尼,提出了模拟采样数据的方法,在他的实验中对气体阻尼指数函数的线性组合。
普罗尼法原始方法已经被改进了很多次从而普罗尼法的现代版本推广到阻尼正弦模型:普罗尼法现在用来分析信号中暂态分量的频率、阻尼、大小,和相位。
普罗尼法适用于成指数衰减的正弦信号的分析,来自于电力系统的信号只要是与时间成线性关系的动态信号。
文献[ 12 ]、[ 13 ]普罗尼法用于分析美国西北部电力系统的振荡。
文献[ 14 ]普罗尼法信号分析方法应用于多机系统的电力系统稳定的设计。
文献[15] 验证普罗尼法分析数字模型高压直流输电系统动态监测系统扰动。
其他显著作品包含在[16]〜[20]。
在本文中,普罗尼法应用于分析中性点经消弧线圈接地系统的故障电流。
接地故障发生在电力系统的馈线上,它是一个地相或者地球相-相故障。
我们的目的是将故障电流分解成阻尼正弦分量。
每组有四个特征参数:频率,阻尼,幅度和相位。
这些参数有利于对电力系统的特点和故障本身的研究,以确定任何接地故障。
在本文的第Ⅱ部分,提出了补偿电网。
通过EMTP(电磁暂态程序)产生接地故障电流。
第Ⅲ部分,普罗尼法的提出及简要描述。
第Ⅳ部分,对模拟的结果进行说明。
2.消弧线圈保护电路一、最先进的单相接地故障被检测出来取决于接地系统和系统电流中的零序电流。
消弧线圈等效为一个电感值可变的电感,它对于基波电容电流是可以完全补偿的,但对于谐波电流就不能完全补偿了,因此要想实现无残流,需要加入有源补偿电路。
发生接地故障时小电流接地系统能够有效减少绝缘的劣化。
然而,在消弧线圈保护的网络中,最常见的接地故障是间歇性的电弧故障,它们是一系列的自熄故障。
不同的方法已被用于检测和定位接地故障。
一般来说,它们都是基于固定的值,如系统的基频分量和第五次谐波。
因此,我们可以猜测,在瞬态条件下,这种继电器的精度问题。
在瞬态故障检测技术领域,可以引用的:首先,在一个大的频率通带零序有功功率测量;其次,基于基频分量的数字量计算保护。
能够保护大部分的接地故障。
然而,他们不利用高频模态分量。
为了改善现有的继电器暂态电流必须进行分析。
我们认为图1所描述的径向网络。
该接地系统是一个消弧线圈、电感X,和一个并联的电阻R。
电感X n在一个直接接地故障的情况下,流入消弧线圈的电流与电容电流的总电流是相同的。
图1、消弧线圈保护网络二、EMTP模拟在图一的网络中模拟EMTP(电磁暂态程序)。
电源变压器表示由[SL]矩阵计算与BCTRAN子程序计算,分布参数电路模型来模拟从母线径向方式的七条线路。
单相接地短路,电阻值R d等于2Ω在A相B相都是2Ω在C相是2~16Ω。
(图2a)。
另一方面,A、B两相接地短路时。
在这种情况下,X(t)是A相对地电流(图2b)。
注意:在这种情况下,X(t)是A、B两相电流的总和。
EMTP仿真的各种参数:.L T:这7个输出线路的总长度.D:故障与母线之间的距离.R d:接地电阻让我们来模拟一个例子,单相接地故障发生在时间0时。
我们假设R d = 2Ω,D = 5Km ,X n = 60Ω,R n = 600Ω和L T = 70Km 。
图3显示R d 上的故障电流。
图2、短路电流x (t ),(a )单相接地短路,(b )两相短路接地图3、短路电流X(t),单相接地短路时接地电阻上电流与时间的变化3.信号分析法一、基本假设我们认为,接地故障发生在时间零点。
因此,随时间变化的信号,X(t),相对地的电阻上故障电流与时间的关系分析。
我们假设在图一中所描述的电力系统中的非线性关系是可以忽略不计的。
我们可以认为故障电流信号是一个线性的正弦激励波形。
因此,对信号进行分析,X(t)是共轭复数和实指数函数之和。
即,X(t)为指数衰减和纯正弦波的总和,它可以表示为:X(t)= A K q k =1e−αk t cos (2πf k t +θk )(1) 其中q 是初等函数的数量,A k 是一个量级,a k 是阻尼因子,f k 是赫兹,θk 是相位弧度。
f k = O ,θk = O 或者π,和αk = O 纯正弦波。
我们假设X(t)是由q 1纯阻尼指数函数和q 2正弦(q =q 1+q 2)。
设X 为N 个等距样本的实际测量数据。
可以这样写:x = x 0,x 1,·····,x N−1 T其中“T”表示复共轭转置矩阵。
从公式(1)我们得出,n= 0,1,…,N-1:x n =X(t n =n∆t )= A k q k =1e −αk n∆t csc(2πf k n∆t +θk ) (2) 其中t 是以秒为单位的采样周期。
由方程(2)可以得到达由q 1、q 2,组成的复杂表达式:x n = βk z k n p k =1 n=0,1,…..,N -1 (3)其中p 表示(p = q 1 + 2q 2),βk 是复幅度,z k 是复频率。
βk 和z k 实际参数范围如下:βk =A K e jθkβk =12A k e jθk z k =e −a k +j 2πf k ∆t (4)二、数值计算方法分析信号X(t),我们选择一种合适的方法,该方法非常适合呈指数衰减的正弦信号。
它是由Tufts 和 Kumaresan 改进的普罗尼法[18]。
此方法可以找到复杂的参数, βk ,z k k =1→p ,因此四的实际参数, A K ,a k ,f k ,θk k =1→q 每个基本功能。
第一步,我们选择一个整数L 使得L>>p 。
事实上,L 值在n/3和n/2之间。
然后,多项式ΨL (z)的定义是:ΨL (z)= (1−z k −1z −1)L k =1= a k z −k L k =0,α0=1. (5)注意: z k −1 k =1→p , z k −1 k =p +1→L 是ΨL (z)的根。
其结果要测定[z k −1]k =1→L 。
系数的计算[a k ]k =1→L ,由(5)得到递归方程: x n =- a k x n +k ,L k =1 n=0,…..,N -L-1. (6)方程(6)可被视为线性方程组其未知数是[a k ]k =1→L 。
因此,他们可以写成矩阵:X CR .a =−x C R X CR= x 1⋯x L ⋮⋱⋮x N−L ⋯x N−1 ,a = a 1⋮a L 和x C R = x 0⋮x N−L−1 (7) 方程(7)中的L>>p ,它有一个以上的解。
在所有的解决方案中,我们试图确定的一个,最大限度地减少了以下公式的量:a 2= a 1 2+ a 2 2+⋯+ a L 2一旦[a k ]k =1→L 被计算, z k −1 k =1→L 从多项式的根被确定。
其结果是p 为零, z k −1 k =1→P ,ΨL (z)在(L-P )的外部, z k −1 k =P +1→L 落在园内[ 19 ]。
然而,p 的值是未知的。
这导致解决一个临时命令优于P 的期望值。
确定[a k ]k =1→L ,我们计算矩阵X CR 的奇异值分解。
然后,我们在三个步骤确定奇异值分解的解决方案。
第一次计算σ,在(L→m 1)最小的奇异值。
第二步,所有这些(L→m 1)最小奇异值不断变小直到为零。
σ是用来计算[a k ]k =1→L 如下:a =− 1λk 2−σ2 m 1k =1 v k T X CR T x C R v k (8) 在[λk ]k =1→L 是X CR 的奇异值和[v k ]k =1→L 是特征向量X CR T X CR 。
一旦得到a ,我们提取多项式的根ΨL (z),那么我们只取最优的一个根。
他们的数量等于m2。
如果m2≠m1再次计算a。
带入(8)计算得到m1,m2的值。
我们重新计算a直到m2=m1。
因此,我们可以得到p=m2=m1。
其结果是,在p得到根是在p复频率的倒数[z k]k=1→p。
联立f k和a k得到z K计算:αk=−log z kΔtf k=angle z k2πΔtk=1,….,q. (9)下一步是编写线性方程组(3)在下面的矩阵形式:x =V.β(10)β=β1⋮βp和V=11z1z2…1…z p⋮⋮z1N−1z2N−1⋮…z p N−1V是一个范德蒙矩阵,用最小二乘法解(10)得到结果:β=V T−1V T.x (11)A k=βkA k=2βkθk=angleβkk=1,…,q (12)普罗尼法完成给定的指数参数的计算(9)和(12),该方法提供了很好的效果,因为信号的信噪比(SNR)足够高。