Скачать 99.68 Kb.
The example used to illustrate the new approach uses information taken from the composite wing box spreadsheet. An open standard semantic editor Protégé created by [Stanford University] was used to structure this information into related taxonomies. This holds the definitions of nodes representing information, and calculations to be performed. This information is saved using a generic structure based on keys that define all relationships, into a relational database. This enables storage of hierarchical data in a relational database and also allows for separation of information into tables according to category and use SQL (Structured Query Language) to query and structure the information as required. Vanguards' tree based decision support tool DecisionPro reads this information and represents it as colour-coded nodes. The decision support tool can perform calculations and so output results. Figure 4 shows how the decision support tool can represent a branch in the tree, visualise an equation and calculate a result. Red nodes represent processes, green nodes represent the part definition and magenta nodes represent resources.
Figure 4 - DecisionPro calculation
The current research uses a technique of interpreting information in order to create decision support programs automatically in response to user choices. This technique is then extended for use in the automatic creation of programs in other computer languages and systems. This can be achieved by automated translation of the DecisionPro information into other languages. The basis of this is that elaborators are nodes in the tree, which are automatically created and dynamically write objects. This allows the wing box definition to be translated to the decision support system for costing and then to other software such as web pages for further processing or visualisation. Taxonomies are created in Protégé for Parts, Materials, Consumables, Processes, Rates, and Tooling for a prototype costing system. New categories can be produced as required. Domain experts would edit the taxonomies; these experts can specify the relationships of classes and the equations to be used via a visual user interface in Protégé. These relationships are evaluated and translated to produce computer code. Figure 5 illustrates how code is produced from the semantic relationships.
Figure 5 - Translation Process Implementation
For the prototype to be extended and applied for external use each taxonomy would be filled with a structured tree representation of experts’ knowledge in the form of classes, values and equations. A costing tree can be automatically produced from these taxonomies. Equations created by the expert, together with choices made by the user of the decision support software, would determine how these taxonomies are linked for a particular costing. The costing tool user would then determine which costing equations are used, by answering questions on dialogue forms. These questions would be asked whenever multiple solutions were available. The benefit of this approach is that the user interface and calculations will be changed automatically to reflect any changes in the model. So if the problem to be modelled changes, only the information that defines the model needs updating, and not the user interface or calculation engine.
Implementation Example – Spar Hand Lay-Up Process
Figure 6 shows a simple diagram of a spar.
Figure 6 - Spar Diagram
Figure 7 shows the decision support systems tree view of the spar branch from the wing box cost model and contains information queried from the related taxonomies. The tree including all the default part definition information for the spar is produced automatically. The buttons in the tree enable choices to be made by the user about materials, consumables, rates and processes. Branches are created in response to these choices. The values in the branch nodes can then be changed as required. Figure 6 shows this.
Figure 7 - Spar branch automatically created from information source
The user will make choices so the decision support tree will be a subset of the information source tree.
DecisionPro visualises large trees by breaking them into individual pages, and indicating with a right arrow where there are further pages that can be viewed. Clicking the ‘Part Definition’ right arrow will display the corresponding information as illustrated in Figure 8. The ‘Derived Values’ branch contains parameters of the spar that are calculated from the spar dimensions.
Figure 8 - Part Definition Branch
Figure 8 is a DecisionPro reproduction of the part definition from the Protégé taxonomy. The DecisionPro software adds an extra functionality, which is to calculate, and store the results of equations captured in the Protégé taxonomy. The equation is represented as text in the Documentation field of the Periphery attribute of Derived Values as seen in Figure 9.
Figure 9 - Spar Periphery Calculation
Different types of information indicated by colour coding may be combined in a calculation. This is illustrated in Figure 10.
Figure 10 - Pre Preg Mass Calculation
Figure 11 shows a branch produced in response to the user’s choice of the ‘Hand Lay Up’ manufacturing process. This illustrates the link of process information to part and rate information. The process nodes are linked to the rate nodes ‘Tool Cleaning Rate’ and ‘Pre Preg Layup Rate’, and to the Part attribute ‘Area’ as these values are all required by process equations. Processes are represented in red, rates in orange, and part definition in green. A multiple classification structure is used in the decision support system, so a child can have multiple parents. The child is shown fully under its first parent but the child may appear again when it has other parents. An upward pointing arrow on the nodes’ right hand side designates that information is used in several locations in the tree, and its’ value can be seen in its’ first parent.
Figure 11 - Pre Preg Mass Calculation
When the user has finished making selections the cost break down of the spar will be displayed as a colour coded branch.
A visualisation suite of tools, named Cost Map, has been created by [Bru et al.]. The Cost Map is aimed at rendering graphically any numerical data (e.g. cost, time, and uncertainty of manufacturing processes). Due to its graphical qualities, the package exploits human cognitivity to ensure efficient information delivery. The need for such tools was made apparent when models such as the wing box spreadsheet started delivering volumes of information, which could not be easily absorbed and analysed in their textual form. The usefulness of the package was ensured by making it compatible with a wide range of standard formats, thus allowing it to accept input from numerous tools. The suite consists mainly of a standalone Visual Basic application and a DecisionPro model, which can output in various formats including SVG for Web visualisation, as well as CSV and XML for data exchange with external packages. The Cost Map emphasises that cost information generation should not be seen as the end of the costing activity. Indeed, cost information is generated to be analysed and take decisions.
Figure 12 shows an example from the DecisionPro version of the Cost Map. Here, the red shades represent over cost, making it easier to visually identify areas of the product where cost reduction decisions need to take priority. Statistical distributions can also be described and reflected in the tree as part of the DecisionPro package facilities.
Figure 12 - Cost Map DecisionPro Version
Figure 13 shows an example from the Visual Basic component of the Cost Map. The colour, size, and shape of the nodes represent aspects of cost, time, and uncertainty. This information can be exported to SVG and a range of other formats, some interactive for exploring the data without the need for the package, and some non-interactive formats for inclusion in documents.
Figure 13 - Cost Map Visual Basic Version with Speech Recognition
The Visual Basic application is not only interactive via keyboard and mouse events, but also via multimedia means. Whenever information can be displayed, it can be read aloud (such is the case in Figure 13 for instance). Moreover, the package can hear spoken commands. These facilities improve the package accessibility for people having difficulty reading from the screen and/or interacting with the standard input devices (i.e. keyboard and mouse).
Prior hypertextual data mining facilities, consisting of both manual and query based filters, allow the user to ignore or concentrate on particular areas of the data structure (most likely, but not limited to, a tree). Other graphical representations are available for both filtered data (in the form of graphs) and the whole dataset (in the form of matrices). Therefore, cost analysis can be carried out globally or locally; in the former case, all information is represented relatively to the whole dataset.
Creation of web pages for a previous aerospace customer led to the view that this could be a popular method of delivery for information and models. Web Output is important as an alternative to spreadsheets because of the problems with spreadsheets explained earlier, and because they provide a multi-user standards based freely accessible method for conveying information and models. Current research into providing web-based models is partly based on previous work completed for providing parametric models online. Figure 14 shows an example of a parametric model.
Figure 14 - Parametric Cost Estimation
Parametric Cost Estimation samples can be found at http://www.cems.uwe.ac.uk/~cbru/.
Further research involved studying the methods used and success of others that had used this approach. This research is outlined in [Ciancarini et al.]; [Huang and Mak]; [Kim et al.]; [Morris et al.]; [Nidamarthi et al.]; [Reed et al.]; [Zhang et al.]. [Li] outlines how a Web-based solution can be applied to distributed process planning. The above research reinforced the view that this is a sensible research approach.
It is possible to output text files so decision trees can be translated into other programming, meta-programming or structured languages. This enables provision of a visual web interface to the models without having to be locked in to proprietary solutions, and ensures the maximum amount of access for users. Figure 15 shows the SVG version of the cost map output in this way.
Figure 15 - SVG Cost Map
Cost map samples can be found at http://www.cems.uwe.ac.uk/~cbru/.
The production of part diagrams using SVG can be automated in a similar manner to that used for the automated production of DecisionPro costing models. Figure 16 shows an example of such an interactive visualisation of a Spar. This interactive diagram was output from the part definition described in the part taxonomy.
Figure 16 - Interactive Spar Diagram (SVG)
The way this interacts with the user can be seen by viewing these examples on the Interactive SVG links page at http://www.cems.uwe.ac.uk/~phale/InteractiveSVGExamples.htm. These provide examples of wing box components. The Internet Explorer examples require an SVG plug in, which can be downloaded for free. The Mozilla Firefox examples are produced using a native XML implementation of SVG and so do not need a plug in. The Firefox examples are rendered in a build of the browser which integrates an SVG renderer, so SVG contents appear as an integral part of the document instead of an embedded object. Both sets of examples are produced automatically using an intermediate tree that draws the component outlines dynamically from the design parameter values. Once defined, a component or a feature could be held in a library and re-used. SVG developers should introduce this automated construction and storage of SVG diagrams, and so make this available to those who do not have access to CAD software. This could allow 3D modelling and collaboration over the web as envisaged by [Park and Fishwick].
Figure 17 shows the DecisionPro tree translated into XML and displayed as a web tree using a stylesheet. The menu uses a stylesheet created by Emmanuele De Andreis [De Andreis].
Figure 17 - XML web translation from Decision tree
This stylesheet is used for the site map on the web site at http://www.cems.uwe.ac.uk/~phale/SiteMap.html.
The XML can also be displayed on the web using a Flash program created by [Rhodes et al.]. This creates a tree with a three dimensional look and a use of colour, shading, and movement of the nodes that makes it an intuitive user interface that is easy to navigate. When a node is chosen, this is moved to the centre of the display and all the other nodes are moved or rotated to position themselves in relation to it. This interface is illustrated in Figure 18.
Figure 18 - Flash interface for navigating exported XML tree
Figure 19 shows the view resulting from clicking on the Spar Part Definition. This shows the parents, children, siblings, and contents of that node. It also allows navigation to any of the related nodes.
Figure 19 - Flash viewing of Spar Part Definition node
This example is on the web site at http://www.cems.uwe.ac.uk/~phale/Flash/FlashHCI.htm.
Researchers at the University of Queensland Australia have developed a hyperbolic browser to display RDF files, this is explained in [Eklund et al.].
Figure 20 shows the DecisionPro tree translated into Java and visualised.
Figure 20 - Translation from decision tree into Java
A translation can also be performed into the Java based Cost Estimator System [Koonce et al.] [Wujek et al.]. Figure 21 shows this.
Figure 21 -Translation from decision tree to Cost Estimator
The use of open standards for representing information makes it possible to enable searches that understand the semantics of the information and so can track all of the relationships between items.
Figure 22 shows the interface for making a search. In this example the user wants to retrieve all the information related to a spar.
Figure 22 - Semantic Search interface
The result is shown as a series of trees for each item that contains the word spar. Each keyword match is the root of a tree. Each tree shows the item found and all its' children and attributes. Figure 23 shows an image of the top part of the results, this is part of the branch for the first item returned.
Figure 23 - Results from semantic search
The information is held in a taxonomy so it is not HTML that is being searched but the taxonomy itself. Because the information is held in a structured way, it is much more likely that searchers will find what they are looking for because the search can follow the relationships represented in the taxonomy. One of the key aims of semantic web research and Web 2.0 is to make this kind of search possible over the web as a whole. The semantic web is a longer-term vision for managing information over the web. Web 2.0 is the shorter-term practical implementation of techniques, which can ease current information search and management problems. XML files can be queried using XQuery a W3C working draft [World Wide Web Consortium]. RDF files can be queried using SPARQL a W3C specification, for which a demonstration is available at http://xmlarmyknife.org/api/rdf/sparql/query and a tutorial based on this has been developed by [Dodds]. A web interface has been developed for Protégé [WebProtege]. An example of the use of this is shown in Figure 24, where a search is made for information on the cure cycle for composites manufacturing.
Figure 24 - WebProtege Searching - Cure Cycle
[Lee et al.] present a distributed visual reasoning system for intelligent information retrieval on the Web. The user can design a query by linking active icons, and then inputting the required parameters. The user can then see the structure of the query and obtain results from the information database.
So far the taxonomies include the traditional object oriented relationships such as child, parent, sibling, attribute, and instance. There are other types of relationship that would need to be modelled in order to maximise the capabilities of software that would use the taxonomies. Examples of these are shown in Figure 25.
Basic key relationships used within the object oriented programming domain between objects have been described. These key relationships depict families of objects that may share attributes and methods through inheritance. They also describe aggregations of objects that make (usually) some geometric sense. Semantic descriptions with more relationship types than the ones described so far allow a more expressive depiction of a problem domain, and can aid some forms of search within a model.
One of the main advantages of a semantic net description, in terms of automated model generation, is that labelling relationships between objects allows the depiction of a number of aspects of a domain in one model, and with a consistent syntax. This allows representation of, say, a product structure and its manufacturing processes together. A single node then is the only representation of that node within the model, with all its relationships depicted as arcs emanating/terminating at the node.
An example is shown in Figure 25, where a rear spar is focussed upon.
Figure 25 -Extended Semantic information
More expressive semantic descriptions are possible within the XML world through the use of one of the standard OWL dialects. Protégé has an OWL plug-in available that provides this functionality, together with links to reasoning tools for maintaining and analysing the logical constructs.
It is also important not to stay limited on one ontology development environment but instead explore how ontologies can be developed using a range of development tools and translated between each where necessary. For this reason Jenna and other open source tools developed by Hewlett-Packard Semantic Web Researchers [Hewlett-Packard], and KAON from the University of Karlsruhe [Volz et al.] are being investigated.
|Keywords Distributed decision, constraint propagation, negotiation, decision support. Introduction||Development of a Product model to support engineering change management|
|Chapter 9 e-business Decision Support Learning Objectives||Decision Support and Expert Systems (mis 538)|
|A spatial decision support system for bank location: a case study||Representing and using legal knowledge in integrated decision support systems: DataLex WorkStations|
|Decision support and disease management: a Logic Engineering approach John Fox and Richard Thomson||Support for Graphical Modelling in Bayesian Network Knowledge Engineering: a visual Tool for Domain Experts|
|Product news plan b sales double, Glamour poll shows strong support for Rx-to-otc switch, wcc welcomes new staff member||[Abel 1992] Abel, D. J., Yap, S. K., Ackland, R. G., Cameron, M. A., Smith, D. F. et Walker, F. G., Environmental decision support system project: An|