外文翻譯--以知識(shí)為基礎(chǔ)的方法在機(jī)械產(chǎn)品設(shè)計(jì)任務(wù)中的實(shí)施

1、<p>　　A knowledge-based approach for the task implementation in mechanical product design</p><p>　　Abstract Although advances have been made to integrate the CAD system with design knowledge, there are

2、 still some barriers to apply the knowledge-based design approach to wide practice.A task implementation method is proposed for the mechanical product design process.The design task implementation model is constructed ba

3、sed on its organized-mode and modularity. Declarative knowledge primitives make up of the kernel composition of design task. It makes the design task more controllable and autom</p><p>　　1 Introduction</p

4、><p>　　Increasing competition of the market forces enterprises to seek ways to be leading in product design and manufacturing. Design in early stages of development is advocated for application to engineering d

5、esign to shorten and automate the development process. Computer-aided design systems are now being applied to "up-stream" design activities Integration of high-end mechanical computer-aided design (MCAD) system

6、s with design activities and design knowledge would allow manufacturers to automate the</p><p>　　2 Related research</p><p>　　This section describes related research that have been devoted to int

7、egration of the MCAD system with knowledge for the enhancement of productivity and reusability of KBE methodologies. In these research and developments, multilayered modeling approaches and the objectoriented method are

8、 commonly used to construct knowledge,design process, and product structure. Lee et al. [6] have suggested an integrated inference architecture of an intelligent CAD system for machine tools design. In this appr</p>

9、;<p>　　Wong et al. [ 1 ] proposed four types of knowledge cells, i.e.,function cell, selection cell, graphics cell, and logic cell in the developed computer-aided logical design system. Different objects encapsula

10、te the product form, design information, process, mail, and so on. Gorti et al. [ 11 ] also presented a knowledge representation model in their SHARED system and five objects to define design process, namely goal, plan,

11、specification, decision, and context.</p><p>　　The research presented in this paper provides a more natural way to integrate knowledge and design task into the MCAD system. A problem-solving method for knowl

12、edge primitives in design task is also discussed in the following sections.</p><p>　　3 Representation for design task model</p><p>　　Normally, the beginning of the development of a product model

13、 is an analysis of relevant products and related design tasks and designers. Normally, the beginning of the development of a product model is an analysis of relevant products and related design tasks and designers.</p

14、><p>　　Moreover, the knowledge driven methodology is given with the association between design objects and knowledge primitives. In the cognitive viewpoint, a general design process is considered to be composed

15、 of several design tasks with the participation of computer and human designers as illustrated in Fig. 1. To facilitate the design process, design tasks should be as automated as possible. In this paper it is assumed tha

16、t design objects and relevant task can be primarily identified.</p><p>　　Fig. 1. Automation of design process</p><p>　　Table 1. The description of design object in knowledge primitive</p>

17、<p>　　3.1 Design task</p><p>　　In the task analysis method, the task model has been viewed as structured analysis technique or one of implementation model which is about how-to.</p><p>　　H

18、ere, the traditional task analysis method is also used, but to make the design task moreautonomous and dynamic, different knowledge primitives and some necessary processing mechanism and are built into it. We present an

19、exelicit description schema for the design task:</p><p>　　T=(id, name, input, output, sub一task, knowledge primitive, goal, action)</p><p>　　.Id: unique identifier for the design task.</p>

20、<p>　　.Name: name of the task</p><p>　　.Input/output: requirements and identified dataflow such as the description of design object and design task (force, torque, velocity, mass, length, volume, etc.)

21、</p><p>　　.Sub-task: top-down analysis of problem solving results in hierarchical decomposition of the whole task into sub-tasks.</p><p>　　.Knowledge primitive: the domain knowledge and the know

22、ledge about strategic control, for example the knowledge conduct the next design task according to the previous one.</p><p>　　.Goal: goals are defined about what should be done within the task. They are usua

23、lly subject to certain constraints, and can be represented by knowledge primitive. Every design task has at least one knowledge primitive as its goal required to be achieved.</p><p>　　.Action: the design tas

24、k act accordingly in response to its resolution. Action is categorized into two types: default action and user-defined action. Default action drive the design</p><p>　　object in the MCAD system, and user act

25、ion needs human involvement, for instance, modification of validation condition.Table 1 illustrates the bending stress validation of the design task for gear design. In Bendstress_ Feedback, a feedback knowledge primitiv

26、e, of Table 1, is used to control the design task back to the previous one specified in it. Some of the variables are added by an iterative value before the current design task is changed.</p><p>　　This sche

27、ma of design task extends the traditional task analysis methodology in three ways:</p><p>　　1 .Knowledge primitives are used as constraints for problem-solving of design task, which makes the design task ex

28、ecutable. Different types of knowledge primitives work for the goals of a design task, this characteristic has a significant implication on the notion of task implementation.</p><p>　　2. Knowledge primitive

29、 can also be used as a guide of task implementation or an advisor for human involvement. There are not only the design objects, there are several successful design tasks and knowledge primitives in repository can be reus

30、ed or consulted for a new product development. Therefore, the efficiency of coordination between design tasks or cooperation between design task and human users could be greatly improved.</p><p>　　3. The ac

31、tion mechanism provides the capability for driving the MCAD system after the solution of the design task.</p><p>　　3.2 Knowledge primitive</p><p>　　Most of current knowledge-based design systems

32、 provide interactive means to integrate knowledge with specific high-level language. However, any language specializes in computation but not representing knowledge and reasoning. Furthermore, learning a special language

33、, coding and maintenance is also a laborious process. The designer should concentrate his effort on the main function of the artifact with least programming.</p><p>　　Like any features and geometry primitive

34、, knowledge primitive can be added piece by piece in MCAD system. Every piece of knowledge is described in a familiar way for designer, i.e.,in engineering manner. Not only the engineering relationships among physical an

35、d geometrical properties of a design artifact are represented in knowledge primitive, the strategic knowledge to control the design tasks and design objects is also represented in the knowledge primitive.</p><

36、p>　　The knowledge primitive can be defined as follows:</p><p>　　K=(id,name,type, content, paras, status)</p><p>　　Id: Unique identifier for a knowledge primitive in repository,</p><

37、;p>　　Name: name of knowledge primitive,</p><p>　　Type: type of knowledge primitive, for example formula, if-then rule, check, table, diagram, standard serial, optimization, feedback etc. If-then rules are

38、 easy to understand and use. Though rule can replace some of the other knowledge primitive types,when involved variables increase, it will be too verbose for a designer to maintain. Table is a knowledge that is arranged

39、in columns and rows, and diagram is a graphic representation that shows the mathematical relationship between two or more set</p><p>　　Paras: the format information for parsing a knowledge primitive. For exa

40、mple, if Paras of a table is "material (value), treatment (value), HB (value)", the constraint parameters are "material" and "treatment", and constrained parameter is "HB". All of

41、them get one value, not a range or a formula, and if Paras of a rule is "C", then the rule can be parsed in programming language C.</p><p>　　Status: there are three status of knowledge primitive in

42、 design environment. They are normal status, suppressed status, and error status. A normal knowledge primitive takes effect in inference and computation process, whereas a suppressed one doesn't. There are several po

43、ssibilities for a knowledge primitive in error status, for example a syntax error, or condition of check is validated, or the exploration for a suitable result of feedback will not he reached etc.</p><p>　　S

44、ometimes a knowledge primitive has different effects in different design tasks. For example, in Fig. 2, a check can be an objective in a design task A for validation of known variables, and it can be used in another desi

45、gn task B as a condition (ordinary knowledge primitive) to solve unknown variables. Moreover it can be used as a condition for feedback in a third design task C of scheme selection.</p><p>　　It should be poi

46、nted out that, although design tasks are achieved progressively according to the underlying solving sequence, the representation of design task in knowledge primitives is generative and declarative in nature. Therefore t

47、he order of the knowledge primitive input is irrelevant to the order of knowledge primitive execution. It is easy to index, select, combine, and exclude. When problem solving is wrong, we can quickly localize the error k

48、nowledge primitives.</p><p>　　Fig. 2. Multiple uses of knowledge primitive</p><p>　　Fig.3. Design object in driven approach</p><p>　　Fig. 4. Mode of design task path</p><

49、p>　　3.3 Design objects</p><p>　　In mechanical product design, design object is considered as an identifiable entity which usually refers to functional assembly and part in product structure. We do not con

50、centrate on the functional and behavior of a specific design object in function model.Design object is characterized by its attribute and geometryoriented information. After the design task is solved, a default action is

51、 triggered for the interface of the CAD system, the geometric shape changes according to the solved paramete</p><p>　　In contrast to the task model, the nature of the design object is dynamically evolving. W

52、ith the addition of the knowledge primitive or each step of the feedback control, the design objects get the new results from the solution. After the stepwise refinement, more and more parameters are determined and valid

53、ated to conform to the manufacturing condition and standard.</p><p>　　3.4 Design task network</p><p>　　We can find most design processes of mechanical products have some common steps, such as st

54、ress validation, rigidity validation, bearing selection, optimization, geometric modeling etc. Despitethe difference content of mechanical parts, the organized modes always resemble. Task networks embody such organized m

55、odes.</p><p>　　All the design tasks constitute a design task network (DTN).</p><p>　　A DTN helps the designer to manage the design process in a higher level. DTN is based upon the identification

56、 of design task and task planning. With the process of task decomposition, the corresponding DTN can be obtained. With appropriate decomposition, some fully automated design tasks can be obtained, as shown in Fig. 4. A

57、rectangle node is simply the graphic representation of design task. An arrow shows the execution path conducting the design task. There are four fundamental execution path</p><p>　　.Sequential path: One task

58、 precedes another. Information exchange is required in task network. As shown in Fig. 4a after the preceding tentative design task finished, the next stress design task then is activated. If the former task failed, the l

59、ater task cannot be executed. If DTN is the total in precedence, the design process will halt when any one of the task fails.</p><p>　　.Parallel path:Design tasks in two different paths can be executed at th

60、e same time without any information exchange between them. Figure 4b shows that bearing selection and key selection can be executed simultaneously.</p><p>　　.Optional/alternative path: Optional/alternative p

61、ath makes the task model more flexible. Apparently in Fig. 4c if a designer decides to begin with the bending stress design and another designer decides to begin with the contact stress design, the result will be differe

62、nt.</p><p>　　.Iterative or explorative path: The design process involves frequent revisions of previous results, restructuring of the design objects or exploration of tentative solution. For example,a design

63、 task with a validation goal always has a feedback condition. If one of parameters is invalidated according to the condition, this task may modify it by adding an iterative value. With the new value, the current task fee

64、dbacks to the most foregoing one in which the parameter is first used.</p><p>　　4 Implementation methodology</p><p>　　There are three fundamental principles in the strategy of reasoning and comp

65、utation for the DTN. The first is to take a design task as the unit of reasoning and computation. Second design task should be solved when the output of its subtask has been determined. Thereby task analysis is topdown p

66、rocess, but the reasoning and computation process is bottom-up process. Third, the process of reasoning and computation is divided into preprocessing and post-processing.</p><p>　　The work in preprocessing i

67、s to get solutions of formulas,equation sets, rules, tables, diagrams etc. Then during postprocessing, the result should be validated by check, rounded by standard serial, and improved by optimization or feedback accordi

68、ng to the algorithm defined in these knowledge primitives.In fact, preprocessing is repeatedly employed by post-processing until an optimized or suitable result is found.</p><p>　　4.1 Knowledge constraint<

69、;/p><p>　　After the knowledge primitives are analyzed, a solution path of variants is required for the specialized solvers which will be discussed subsequently. The solution path also provides the designer th

70、e visual analysis between design variants. Constraint propagation techniques provide a powerful mechanism for accomplishing this job. In other words, constraint can be thought of as a guiding mechanism for the hybrid eng

71、ine to find a feasible solution path. For simplicity, only the solution path for</p><p>　　Currently, some advanced analytical tools [9] for the processing constraint have been developed. Design sheet [12] is

72、 one of the most advanced constraint management systems. In this system, a graph-theoretic decomposition process is used to identify subsets of equations that need to be solved simultaneously.</p><p>　　We ex

73、tended this method to get the solution path of related variants in knowledge primitives. The concept of knowledge constraint is brought forward which carries out the restrictive function between design variants. First kn

74、owledge primitives are decomposed further into knowledge constraints, denoted as k. The knowledge constraint represents the constrained relationship which is generated by the knowledge primitive. For a rule, table or d

75、iagram, the association of a constrained variant and th</p><p>　　For example, in Fig. 5 there are two constrained variants, i.e.,g and n, then two knowledge constraints k1 and k2 are generated. k1 represents

76、 the constrained relationship between K1 andvariant g. k2 represents the constrained relationship between K1 and variant n 。Although there is no explicit constrained variant in Kq, but the number of its knowledge constra

77、ints can be determined by the number of equations. The decomposition result of Knowledge primitives K1，K2, K3, Kq should be as follows:</p><p>　　After the knowledge primitives are decomposed, a bipartite gra

78、ph is constructed with knowledge constraints and referenced variants as two kinds of nodes. Edges in the graph connect knowledge constraint nodes to variant nodes, and indicate that the variant is referenced in the knowl

79、edge constraint. The bipartite graph is shown in Fig. 6a.</p><p>　　The next step is to assign directions to many of the edges in the graph. There are some rules for directing the graph, as follows:</p>

80、<p>　　.If the knowledge constraint is decomposed from rule, table and graph, and variant is constraint, edge is directed from constraint variant. If the knowledge constraint is decompose from rule, table and graph

81、, and variant is constrained, the edge is directed away from the constraint variant.</p><p>　　.If the knowledge constraint is to represent a equation, there is exactly one edge directed away from the equatio

82、n.</p><p>　　.If there is an edge directed into a variant, all other edges are directed away from that variant.</p><p>　　Fig. 5. Example of knowledge primitives</p><p>　　Fig. 6. Know

83、ledge constraint network</p><p>　　4.2 Hybrid engine</p><p>　　For the diversity of design knowledge, a single solver cannot handle all different forms of knowledge primitives. Therefore,for a spe

84、cific form of knowledge primitive a corresponding solver is developed based on a numerical or symbolic algorithm.Both table and diagram are highly formalized and a specialized solver can be parsed and solved with great e

85、fficiency. Whereas empirical rule and decision-making rule would be processed more efficiently in the forward or backward inference method.Multi-so</p><p>　　.Solver for rule, check and feedback</p>&l

86、t;p>　　.Solver for table and diagram,</p><p>　　.Solver for equation set,</p><p>　　.Solver for optimization and evaluation, and so on.</p><p>　　These different knowledge solvers ar

87、e coordinated by the hybrid inference engine, as shown in Fig.7.</p><p>　　When the knowledge primitives of a specified design task are submitted, the hybrid engine dispenses them to the corresponding solvers

88、. The solvers parse the knowledge primitive into a constraint network which is composed of knowledge primitives and referred variants. Then the variants will be sequenced. According to the sequenced solution path, the va

89、riants are solved by different solvers in pre-processing and then improved in postprocessing. Finally the actions trigger the design objects to ref</p><p>　　4.3 System architecture</p><p>　　The

90、knowledge-based design system was developed based on the models introduced earlier and its architecture is shown in Fig. 8. The system architecture comprises several components in a MCAD environment, such as hybrid engin

91、e, design object management, task analysis, knowledge management, domain knowledge base, design task database, and design object library.</p><p>　　Intesolid2.0 is the parametric feature based geometric model

92、ing MCAD. It provides a userfriendly platform which enables the designer to add a knowledge primitive or to import a design task conveniently. It is possible to drive the design objects directly with the action mechanism

93、 after the knowledge primitive is solved. By incorporating design tasks and knowledge primitives, it can dramatically improve the designer's productivity and guarantee the correctness of results.</p><p>

94、　　The knowledge management module allows the designer to do some analysis, such as how-to/what-if analysis, tradeoff studies,and so on. Design object management module manages the relationship between input/output of the

95、 design task and the parameters between the design objects. Design task analysis tool provide the substantial capabilities for task planning and task definition.</p><p>　　In some cases, the common design obj

96、ect and design task can be retrieved from the design task database and reused in the current case. The design task database has been developed to maintain the consistency and to support the selection of design task accor

97、ding to specification. Indexed by the name and the goals, design task can be retrieved and reused in another design context. Relational database management system implements the association design task and its schema.<

98、;/p><p>　　Fig. 7. Function of the hybrid engine</p><p>　　Fig. 8. Proposed system architecture</p><p><b>　　5 Example</b></p><p>　　This example used for illustra

99、tion is the design of a two-stage transmission gearbox. Transmission gearbox is usually designed to transmit power and is essentially used as a speed reducer or for stepping up the speed and is manufactured as a separate