Proceedings of CK2005, International Workshop on Computational Kinematics Cassino May 4-6, 2005 Paper 41-CK2005

Application of Computational Kinematics in the Digital Mechanism and Gear Library DMG-Lib Ulf Döring Faculty of Computer Science and Automation Computer Graphics Group [email protected], Torsten Brix and Michael Reeßing Faculty of Mechanical Engineering Engineering Design Group [email protected], [email protected] Technical University of Ilmenau PO Box 100565, 98684 Ilmenau, Germany

ABSTRACT – The DMG-Lib is an interdisciplinary project that was started to collect and preserve knowledge concerning mechanical devices. In contrast to common digital libraries an efficient problem oriented information retrieval will be supported. The realization of the long-term project is only possible by means of extensive employment of computational kinematics, which opens up multifaceted prospects – especially according to the efficiency of the workflow and the efficiency of the access to the knowledge space. Aside from a common survey of the DMG-Lib project the paper shows, how kinematic models are extracted, stored and applied. The described approaches are practically proved or work in progress as well as future work which shall be discussed on the workshop. KEYWORDS: Mechanism Design, Applied Computational Kinematics, Digital Library, Information Extraction, Information Retrieval

INTRODUCTION Mechanical devices will be important for the realization of motions in spite of an increasing application of electronically controlled direct drives. Modern technologies like nanotechnology or biomechatronics open up new areas of application for mechanisms. Therefore, the relevance of mechanism theory even increases. Current research and development activities are indicators for this trend. Examples are mechanisms with several drives, parallel and serial structures, compliant elements, mechanisms to control high precisions, large amount of mass inertias, complex spatial and coupled movements or mechanisms for miniature and micro assemblies [1]. The existing worldwide knowledge in the form of books, drawings, functional models etc. is mostly scattered, difficult to access and does not comply with today’s requirements concerning a rapid information retrieval. In addition to this, mechanism theory is taught less and less. Nevertheless the industry and research institutes require access to the whole mechanism and gear knowledge. Existing activities to provide such knowledge are promising (e.g. [2]) but by far insufficient. The most important problems are: 1. the amount of information originating form different resources and 2. the efficiency with which needed information can be retrieved. The development of a digital, internet-based library for mechanisms and gears is necessary. This library must contain large amounts of the worldwide knowledge about mechanisms and gears. Furthermore it must allow a user to access the knowledge in an efficient problem oriented way. In the next years the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) funds the digital mechanism and gear library (DMG-Lib) project which is of particular importance not only for product designers and researchers, but also for teachers and other user groups.

1

PROJECT OVERVIEW The main challenges are the size of the knowledge base as well as the necessity to combine a great variety of different types of resources (see Figure 1 which shows some examples), as there are: • hand- and textbooks, • articles, • photographs, • videos, • engineering drawings, • sketches, • functional models • calculation sheets of different software, • patents • etc. The aim is not only to integrate documents of these types. It should be possible for users to choose the type of documents they want to retrieve from the DMG-Lib. One example for this is the formal representation of a certain mechanism: • in the web-browser the user needs HTML code and images which describe important aspects of the mechanism, • for the import in his own software he may need a kind of XML description, • for the conference presentation he may need an AVI video file, • for interactive tests with some calculation software (e.g. Cinderella [5]) he may need a sequence of construction steps – if possible in the special format of the software • and so on. To allow such flexible handling of representation formats, converter software was / will be developed and integrated into the DMG-Lib portal.

Fig. 1: Different sources of knowledge which are used to build up the DMG-Lib

2

Project state

Identification, Rights, Procuring Digitalization

Registration Description of the digitalization process

digital raw data Conversion

converted data Enrichment

State meta data Common digital libraries

State meta data

enriched data

meta data

special software Search engine ... Analyses software Semantic net

Portal database

Raw data (sources)

Production process database

Project Coordination

Animation generator

Information management (Retrieval software)

Search results, Visualisation

special questions

Internet Portal Interactions

User

Fig. 2:

Scheme of the production workflow for the DMG-Lib on the left side as well as the content management software on the right side.

The workflow (see Figure 2), which takes the special needs of the different sources into account, can be described briefly as follows: • At first, relevant sources are identified, procured and (if necessary) digitized. • Then the documents are converted to allow a web-compliant representation. • The sources are evaluated by experts who decide: o How important is the resource? o How should it be integrated? o Which actions in the workflow must be performed to allow the integration? In contrast to other digital library projects the preceding digitalization steps are only a necessary preliminary for the steps of an extensive preparation. To allow a reasonable search for certain information and especially for solutions of a certain problem, the digitized data must be indexed and enriched. This is supported more and more by software - computer aided methods allow to analyze the text pages, images and image sequences. A set of according tools is under development. For some tasks standard software (e.g. Abbyy FineReader [6]) is integrated into the workflow. • Text documents are analyzed. The recognized full text and particularly its detected text structure support the term-based search. Here a semantic network [7] is used: o to allow cross-language retrieval [8], o to handle synonyms (e.g. to automatically translate terms used in older literature) and o to support intelligent search which is based on concepts like generalization and association. During the analysis of the text according entries in the semantic network must be created by the experts which control the analysis process of important documents. Furthermore they add cross references (who a document and to related locations in other documents) to selected documents. This and the detected structure of text documents eases browsing. • The analysis of the images and image sequences leads to a more abstract representation of the mechanisms and gears they show (an example is shown Figure 6). This representation can be transformed to other representations. We prefer a constraint-based one [3] that allows the simulation as well as the variation of the models. Nevertheless the abstract representation may be exported into other formats too.

3

• •

Thus the export of equations would allow the application of methods like homotopy or Gröbner bases. Furthermore formats of certain relevant software products could be exported directly. If a software manufacturer doesn’t allow to generate files in his format he could supply import filters for his software. Relevant properties, which are identified during the simulation, will be added to the search index. This is important when a user searches for a solution. Furthermore it is possible to animate selected images in the originally static pages of the digitized books. Feedback from the users to the project as well as communication among the users will be supported by the possibility to add comments and to post in DMG-Lib discussion forums.

The DMG-Lib project started in 2003 with partners form TU Ilmenau, RWTH Aachen and TU Dresden. Further partners could be enlisted, e.g. the German committee of IFToMM. International cooperation is striven for too. Since 2003 an interdisciplinary working group developed a workflow and a concept for the internet portal which meets the requirements of the different user groups. Questions concerning copyrights, digitalization costs, enhancement effort and content management are clarified. Extensive investigations to find appropriate sources go back to literature of the 16th century. Especially in the 19th and begin of the 20th century lots of relevant and free books where found. Often the content of those books (written in German, English, French or Russian) is still up to date. Patents were recognized as another important source for the DMG-Lib. Furthermore negotiations with living authors or their heirs took place. Finally the DMG-Lib has the potential to become a publication platform in the field of mechanism design. First digitalization results (books, videos of functional models) are analyzed and supplied in an implementation of the internet portal which currently serves as a platform for usability tests. It demonstrates how the deficits of previous digital libraries in the area of mechanisms and gears can be overcome (see Figure 3).

Fig. 3: Examples from a prototype of the DMG-Lib which is used for demonstration and usability tests.

EXTRACTION OF MODELS From the computational kinematics point of view the most important part of a model is the solution principle. It describes the general structure of the model (e.g. how many links, joints … are involved and how they are connected) as well as the according geometry (e.g. the position of points, the profile of a disk cam …). Before the models can be used, they have to be extracted from the very different sources and must be stored in a unified representation (see next section). This section shows some examples, how solution principles from planar mechanisms can be extracted from figures or photographs (originating from books, patents, databases, screenshots of some software … ), videos and software applications. To support the extended retrieval possibilities in the DMG-Lib portal (as described in a following section) each shown mechanism should be represented internally by a solution principle. That can

4

only be achieved by means of special software that allows an efficient extraction process. Such software is currently being developed in the project. The software combines image processing, pattern recognition and constraint solving techniques. Because of the ambiguities in the visualization of spherical or more general spatial mechanisms the extraction of the geometry of such mechanisms from images is difficult and may be very time consuming, in some cases it even may be impossible. Hence in the beginning of the DMG-Lib project there will be only exemplary extractions of the geometry of spherical and spatial mechanisms. The extraction of the structure would be possible but without the geometry and an according simulation of the solution principle it is hard to test the correctness of the extracted structure. Therefore the extraction focuses in the first time on planar mechanisms. Nevertheless the non-planar mechanisms shown in the images are classified by experts so that it is possible to retrieve the images in the portal. Furthermore the import/analysis of CAD data could be realized as a mediumterm task.

b)

c)

d)

a)

Fig. 4: Process of the extraction of a solution principle from an image in a book: a) the book page with the drawing and b) the image after some image processing. c) To support fast positioning of the symbols for revolute pairs circles are proposed by a pattern matching algorithm. Then the symbols are added interactively. d) The extracted solution principle which can now be added to the database. Nowadays an extraction which is based on a single image must be done mostly manually. Figure 4a shows a drawing taken from [4]. In the extraction workflow (Figure 4b, 4c and 4d) the image is: • preprocessed (as needed: dewarped, color converted, resized …) and • analyzed to detect standard patterns like circles or lines. • Finally symbols are added interactively which describe the structure of the mechanism. The detection of standard patterns helps to find the correct geometry. Revolute pairs for instance can often be detected as circles (in photographs as well as in drawings), where the center of the circle is the position of the revolute pair in the solution principle.

5

If the extraction tool is connected with a simulation software, the behavior of the defined mechanism can be tested interactively to check if it behaves as expected. This is important to minimize the number of errors in the mechanism database. The extraction of solution principles from drawings can be combined with the extraction of the drawing style. This allows interesting and possibly didactically useful animations of drawings which can be integrated into the portal. For a proper visualization of single animation steps in the original style (see fig. 5) various information is needed: • the reconstructed background (parts of it are hidden in the original image), • the profiles (shapes/contours) of the drawn links, • parameters of the used lines: o color, o style (continuous or discontinuous according to a certain pattern), o width and o a function that describes how the line merges into the background, • layer information to handle overlaps and • a description how overlaps should be handled (ignore, change line style or hide). At the moment only software prototypes are available which demonstrate certain extraction and rendering functionality. Therefore the retrieval of such information is time consuming and should only be done in cases where an animation in the original style is really needed to understand a drawing.

(b)

(a)

Fig. 5.: A figure taken from [9]: (a) shows the mechanism in the original position and (b) shows the mechanism in a new position but rendered in the style extracted from the original image.

In contrast to an extraction based on a single image the video-based extraction has the potential to be performed automatically. That means if the quality of the video material is sufficient (in particular according to resolution, compression artifacts, highlights, motion blur) an expert will often only have to confirm the correctness of the solution principle which was found by the software. If the software fails, an arbitrary frame of the video sequence may be processed as single image as described above. Among different approaches especially one approach is especially promising to be robust against temporarily hidden elements and highlights. It is based on contour detection, contour tracking and constraint solving. Constraint solving is used in this approach to forecast the next positions of the parts (based on the current model and the detected position of driving elements) to enhance the contour tracking (it becomes faster and more robust). An other way to collect solution principles for the DMG-Lib is the conversion from other file formats. The use of XML-based formats is very popular and each serious software will provide at least an XML-export option. Therefore the chance to convert the XML-based description of solution principles defined in other software becomes better and better.

6

INTERNAL MODEL REPRESENTATION The DMG-Lib stores the extracted models as XML-files. An example is shown in Figure 6. At the moment the development of the format specification focuses on planar mechanisms because the extraction process of those models from images is currently supported. Later, support for spherical mechanisms and spatial mechanisms will be added successively. The provided example shows the information collected from the original source and first meta information which is needed to process the model in the DMG-Lib. When the model is exported, further information may be added, e.g. the classification according to a certain classification scheme, a preview in a certain image format or the description of its function.

XML Header Location in the DMG-Lib database and associated information retrieved from the original source [4]

Mechanism description with coordinate systems, components and relations between components

Fig. 6: XML-file for a mechanism found in [4]

7

USE CASES There is a great variety of use cases where computational kinematics can be applied. Some were mentioned in a previous section (extraction of models). From the vantage point of the present some use cases are only general possibilities which we want to discuss at the workshop. Some use cases may open new research fields. At first basic possibilities are discussed. There are: • Analysis of the structure This is an important “by-product” and does not belong to computational kinematics directly. But it can be combined with computational kinematics algorithms very successfully. For the analysis of the structure the elements and connections of a solution principle are interpreted as nodes and links of a graph (for example see Figure 7). Graph or subgraph algorithms can be applied to recognize known models and submodels respectively. The structure can be analyzed independently from certain parameters like the length of links (or more commonly the relative position of kinematic pairs carried by the link) or the position of points. • Analysis of the movement This is the main use case in the DMG-Lib. It can be applied when parameters are known or at least assumed (e.g. during the model extraction process as described above). Especially movement simulation can lead to important information about the paths of points, their velocities and accelerations. • Analysis of the function Based on the results of movement analyses, more abstract information can be found which describes functional aspects like transfer functions or special features like the presence of instantaneous dwells. • Variation of parameters Parameter variation in connection with movement analysis and an appropriate search strategy leads to a powerful optimization approach for solution principles.

Fixed Revp

Link

Revp

Revp

Link

Link

Fixed Revp

Revp

Link

Link

Revp

Link

Revp

Fixed Revp

Fig. 7: Structure of the graph which represents the solution principle shown in Figure 4d. Three types of nodes are used. Revp is the abbreviation of revolute pair.

These basic possibilities can be applied or combined respectively in more concrete use cases which will be discussed now. • Generation of new images or image sequences To improve the understandability of drawings or even photographs the generation of an animation or video is very useful. The generation is based on the position of points or certain directions acquired from motion analysis. This information about the geometry is combined with information about the style which shall be used to render the new image in the sequence. Figure 8 shows images which are rendered in different styles. Furthermore it is possible to enrich the images with additional information obtained from the analyses (e.g. arrows which visualize kinematic parameters). The generation of images can be performed offline when an expert decides that e.g. from an didactical point of view it is important to supply an image or animation in a certain style and with some additional information. The results would be added permanently to the content of the DMG-Lib. Users who want to understand a certain mechanism in more detail or who have to prepare a presentation may desire an image/animation generator as an online service. Beside the style and additional information the output format (e.g. animated GIF, SVG or X3D) may be chosen by the user.

8

a)

b)

c)

d)

Fig. 8: A mechanism from [4] shown in different styles: a) a possible symbolic style, b) symbolic style with visualization of layers, c) 3D-style with first embodiment and d) the original style used in [4]. •

Searching certain functionality An engineer will often search for a certain functionality to solve a current mechanical design problem in the company or research institute. Engineers with good education and/or experience in mechanism design may be able or may have the wish to select a certain mechanism or a set of mechanisms directly (e.g. by name) which are optimized using a certain software later. As described in the introduction section the number of those users will probably decrease in the future. Therefore extended search possibilities are needed, e.g.: o Stroke-based The user may make a stroke to describe his idea of the path of a point or a transfer function. This stroke can be compared with strokes obtained from analyses stored in the database. The database will become quite large. Note that movement and transfer behavior of a mechanism depend on the parameter values. Therefore for each mechanism a combinatorial variety of parameter sets has to be analyzed. That leads to a lot of data which must be stored and processed by matching algorithms. Figure 9 shows a stroke which represents a desired path and three mechanisms which can generate similar curves. o using keywords to describe certain properties During the analysis of a mechanism (in connection with different parameter sets) special properties may be detected, e.g. the number of instantaneous dwells, ranges with linear movement. The use of fuzzy logic may allow to map certain properties like numbers or ratios to keywords. For example the number of certain elements could be mapped to terms like “less complex” or “very complex”. Nevertheless to describe the user’s intent, numbers like the “degree of complexity” could be used directly too. A further example for such a number is the ratio which describes the relative size of the working area of a special point (according to the area needed by the whole mechanism). This may also be detected during the analyses.

9

Fig. 9: Stroke of a desired path (upper left) and four mechanisms which may be used to generate such a path. The two lower mechanisms are found according to the roberts-chebyshev theorem. The user may select the mechanism which fits best to his needs, for instance according to allowed elements (maybe he does not want a prismatic pair) or the position of the connections with the frame (relative to the path).

After the retrieval of mechanisms which fit the needs of the user to some extent, a parameter optimization can be applied. This may be done externally by the software of the user. For this possibly export/import interfaces will be needed (to avoid that the user has to rebuild a mechanism interactively in his own software). In general an online optimization of the parameters could be desired too. But this may cause performance problems on the server and an according service would probably be very restricted. •

Direct search for certain structures Based on the database of graphs which describe the structure of stored mechanisms such structures are searchable too. The structure searched for may be entered interactively using a tool for solution principles (similar to Figure 4d), a graph tool (similar to Figure 7) or some textual interface. In this way, users could be able to check if some structures are already described. Furthermore a person, who has to check how innovative a patent is, could find according occurrences in a variety of sources. The most important property of this search process is, that it is independent on names which may be very different according to languages, history and even schools.

•

Comparison of structures within DMG-Lib A very important use case of the search for certain structures occurs in the DMG-Lib itself. When a certain structure is found again (may be as graph or subgraph), the according sources can be linked by a cross reference. Furthermore the identifiers can be linked in the semantic net. This makes the name-based search more powerful.

10

•

Automatic generation of names for mechanisms Based on the structure, the results of the kinematic simulation and a set of rules, the name of a certain mechanism could be generated automatically. This could lead to more objectivity and transparency in the naming process. International cooperation with involvement of the IFToMM (esp. Permanent Commission on Standardization of Terminology) is striven for. Strictly speaking, the implementation in the DMG-Lib will not store certain names directly with the mechanisms. It is advantageous to generate a link to a node in the semantic net. This node is connected with the current default name of the detected type of mechanism. Alternatively older identifiers or names in other languages can be found without the problem of homomorphism (which may occur when names are stored directly). Furthermore it will be possible to change the default name of all occurrences of a certain mechanism at once according to latest naming conventions. One final remark to the term semantic net used above. This semantic net is a local network which stores information to concepts used in the DMG-Lib. It is not directly part of the Semantic Web as it is propagated by the World Wide Web Consortium [10]. Nevertheless the DMG-Lib becomes part of the Semantic Web because it will provide metadata concerning the stored documents, e.g. via an OAIinterface [11].

CONCLUSION In general a lot of the use cases discussed above could be realized without the help of computational kinematics methods. But what we need in the DMG-Lib project is a very large database which has a lot of qualified cross references. This can not be achieved using human resources only. Therefore DMG-Lib specific software is developed to achieve an effective workflow. As shown in this paper, computational kinematics can be applied successfully in this software. It makes the production process for the content more effective and improves the quality of the content. Furthermore the retrieval process can be fitted better to the needs of the users.

REFERENCES 1. Huang T. 2004. Proceedings of the 11th World Congress on Theory of Machines and Mechanisms. Tianjin. China Machinery Press. 2. Web resource of the Kinematic Models for Design Digital Library, 2005 http://kmoddl.library.cornell.edu 3. Brix, T., Döring, U., Reeßing, M. 2005. Constraint-based computational kinematics. Proceedings of the International Workshop on Computational Kinematics. Cassino, Italy. Paper 40-CK2005. 4. Grübler, M. 1917. Getriebelehre – Eine Theorie des Zwanglaufs und der ebenen Mechanismen. SpringerVerlag. 5. Web resource of the interactive geometry software Cinderella, 2005. http://www.cinderella.de 6. Web resource of the FineReader OCR software, 2005. http://www.abbyy.com/finereader 7. Lehmann, F. and Rodin, E. Y. 1992. Semantic Networks in Artificial Intelligence. Elsevier Science Ltd. 8. Grefenstette, G. 1998. Cross-Language Information Retrieval. Norwell, MA. Kluwer Academic Publishers. 9. Hülsenberg H. A. 1877. Beitrag zur Theorie des Universalzirkels von Peaucellier, mit besonderer Berücksichtigung seiner Anwendung als vollkommene Geradführung. Berlin, Düsseldorf. Zeitschrift des Vereines Deutscher Ingenieure. 10. Web resource which describes the Semantic Web, 2005. http://www.w3.org/2001/sw/ 11. Web resource providing information to the Open Archives Initiative, 2005. http://www.openarchives.org

11