Conceptual design by means of genetic algorithms.

Albers, Albert ; Enkler, Hans-Georg ; Frietsch, Markus 等

In conceptual design activities designers often resort to already existing components that are combined and arranged into a new system. Since there is a huge number of catalogues of different manufacturers, especially in multi domain systems, one big challenge is to find an optimal configuration. Furthermore designers have to deal with requirements and constraints that are changing during the development process. This article introduces a new approach to support designers in this task by means of a computer aided approach. General goal is a (semi-)automatic generation of compatible conceptual design proposals that meet the predefined requirements.

Key words: mechatronics, configuration, conceptual design, optimization

1. INTRODUCTION

An important approach for enterprises to be successful in the globalized market is to utilize computer aided software tools in product development. Starting from a low level of detail, most simulation tools cannot be used in early activities. Furthermore there is a huge number of catalogues of different manufacturers, especially in multi domain systems. Therefore a further challenge is to find an optimal configuration.

2. COMPUTER AIDED CONCEPTUAL DESIGN

New products are often based on a combination and arrangement of already existing components. The new system has to fulfill a predefined set of requirements. Due to the huge number of catalogues of different manufacturers, especially in multi domain systems with a huge number of interdependencies, it is an even bigger challenge to find an optimal configuration. It is necessary for designers to be at least familiar with the involved domains and their requirements and boundary conditions. Close communication is one of the main factors to avoid suboptimal solutions. Furthermore designers have to deal with requirements and constraints that are changing during the development process. A fast and automated evaluation of the current system design regarding these changes and--if necessary--the derivation of a new system design is desired. The impact of fuzzy requirements and boundary conditions on the 'optimal' system design should also be considered. Finally, a complete and systematic evaluation of possible component configurations even off the beaten track could possibly lead to innovative solutions. Manual exploration of all these aspects is often not possible due to time restrictions. Since many evaluations can be performed through structured procedures, e.g. calculation of the center of gravity, an automated computer aided approach seems appropriate.

Nowadays there is a huge number of product catalogues of a multitude of manufacturing firms. Almost every company has its own philosophy to create these catalogues in order to structure its product portfolio in the best way. This leads to the fact, that catalogues of even very similar products often differ significantly from supplier to supplier. Especially in the development process of mechatronic systems, the designers have to manage a lot of different catalogues to select the needed components from multiple domains like motors, controllers, brakes, sensors or gears. Another challenge is to keep 'up-to-date' with the newest products and innovations. One can find four different 'levels of assistance'. Using a level 0 catalogue-often big tomes with the complete product portfolio in one single book, annually updated, with text links to compatible components--will result in a very time consuming process: Level 1 catalogues provide comfortable search functionalities and hyperlinks to compatible components. Thus it is faster but not easier for the designer to look for compatible configurations. The bulk of manufacturers' web pages and also most of the product CDs offer these functionalities. Level 2 catalogues include tools like component filtering to avoid selection of incompatible configurations (http://shop. maxonmotor.com). The user does not have to look at the text-or hyperlinks to check for compatible components manually. This is done automatically by the software. By choosing one component, the number of possible configurations is reduced to assist the designer in finding the required combination. All preceding levels are limited to the products of only one particular manufacturer. This is in fact unsurprising because every company aims to distribute their own products. But from a designers point of view a manufacturer spanning solution would be much more auxiliary. A small step towards this goal is already realized in terms of CAD-models. In literature a level 3 approach based on a database of over one thousand motors and one thousand transmissions of different manufacturers was implemented. This database contains over ten thousand possible configurations. The designer has to enter the required torque, angular velocity and optional boundary condition and the selection process, based on an automated dimensioning, is executed automatically. This facilitates the selection-process and allows faster reactions in case of changing requirements or boundary conditions.

The remaining disadvantage of this framework is its lack of flexibility regarding additional boundary conditions like design space, EMC, resonance frequencies, dynamics and other multi-domain effects. Therefore additional tools, like CAx- or simulation tools, have to be integrated. Also fuzzy requirements or a weighting of different criteria against each other are not possible in the existing frameworks. Due to the situation described above, efforts in generating a greater support for the development of mechatronic systems have been made e.g. (Devanathan & Ramani, 2007), (Kim et al., 2003), (Yvars, 2008), (Aldanondo et al., 2003).

3. APPROACH

We propose a new level 4 catalogue approach that is divided into two main layers. The first layer is based on the 'conceptual verification' method developed at the IPEK and presented in (Brudniok, 2007). The second layer supports the designer in the following optimization, or rather, evaluation step. Therefore an interaction-process of this catalogue with CAD software is presented. The whole framework aims to assist the designer in time-consuming and simple tasks in order to gain more time for the creative part of his work, which cannot be transferred to a computer.

3.1 Layer 1: Iterative selection process

When designing complex new systems, the design task is commonly segmented into smaller subtasks resp. subsystems. For complex mechatronic systems a segmentation based on the functional structure like proposed in (Brudniok, 2007) is recommended. The objective of the first layer is the generation of compatible design proposals for these subsystems. The therefore required library has to contain all component specific information like engine speed, torque, efficiency factors, CAD data but also metadata like type of motor, compatibility parameters, level of preference for a specific component, etc. To allow easy exchange and update of this library we propose an XML file format. Based on this component library the selection process is performed using various requirements and boundary conditions. In contrast to the already existing level 3 solution, the user has easy access to every property of the components. Therefore a multitude of various criteria can be tested and taken into account when generating design proposals. The main advantage is the possibility to do so with complete compatible solutions. Firstly the total number of possible configurations has to be reduced by eliminating the component combinations that do not lead to feasible configurations. Secondly 'don't-like' components were eliminated to realize company specific preferences. Thirdly components with parameters outside a specific range were eliminated as well. Then the generation of compatible configurations is performed. This is done by creating all possible combinations of e.g. motors and gears following the product hierarchy. Every potential solution consists of 'component primitives', that are combined using one library for each primitive. CAD-models can be used either for the selection process or for the later step. Additionally the designer does not have to look for each single model, if a configuration is chosen at the end of the process. Within the selection process, the models can be used for a multitude of analysis e.g. design space or centre of gravity. For correct assembly of single components in the CAD environment a new approach is used. Each component has to be provided with working surfaces for connecting them to each other and to a possibly existing environment. Additionally a new concept is used to determine the validity of design proposals. This concept extends the conventional interfaces to a more general approach for mechatronic systems. The basic idea is that both physical elements e.g. connectors but also non-hardware elements like bus-protocols can be seen as interfaces. If combining two components into one system, both working surfaces have to be present in order to generate a valid design proposal. To realize the integration of fuzzy or indistinct requirements, thresholds and ranges are used. The evaluation is accomplished iteratively for each single criterion to steadily reduce the number of possible solutions. The evaluation methods vary strongly in their computing time. Hence the optimal sequence regarding the computing time of those tests will be evaluated automatically in the future. The output of this processes are several different design proposals that fulfil the requirements. By creating a target function containing the fulfilment level of each criterion, a customized weighting of the different requirements can be realized. The final design proposals are further optimized by the optimization layer. Initially it is a relatively time intensive process to collect all relevant datasets and to feed them into the database. Firstly this has to be done only once resp. for new data only and the longer the catalogue is used, the bigger is the benefit of this framework. Secondly the long term goal is to establish a standard catalogue system. Each company could provide their catalogues to facilitate it for the user to keep its own library up to date. This concept offers also big advantages to manufacturers: a fast distribution of a new product, resp. the knowledge that a new product is available, without having to wait for the new printed catalogue.

3.2 Layer 2: Optimization of the component configuration

After having generated a set of compatible component configurations it is necessary to arrange the components spatially, i.e. to define position and orientation. During this process several restrictions such as design space, electromagnetic compatibility, etc. have to be taken into account. In order to provide an automated process, we propose an integrated approach using a combination of CAD, CAx and genetic algorithms. There exist several free and commercial software implementations-such as (http://www.cs.sandia.gov/ DAKOTA). In order to integrate CAD, CAx and DAKOTA, it is necessary to develop an interface to link DAKOTA and CAD models of the components chosen in the iterative selection process described above and-if necessary-to further CAx analysis tools. During the process, parameterized CAD models related to the respective components are loaded, assembled and located using a set of parameters. Subsequently, the configuration is analyzed by CAD with respect to available space, collisions, etc. In general, system configurations can be evaluated in many respects. Parameters for locating the components are generated by means of DAKOTA. Data between CAD and DAKOTA are exchanged using small ASCII files including a new set of parameters for CAD or evaluation results for DAKOTA. Due to its high flexibility, the user may integrate additional analyses such as FEA.

4. CONCLUSION

This article introduced a novel method to support designers in conceptual design phase by means of a computer aided approach. General goal of this framework is a (semi-)automatic generation of compatible conceptual design proposals that meet the predefined requirements. Currently, the presented method is being implemented in a software tool. The method will be evaluated during the development process of the humanoid robot ARMAR IV and V (http://www.sfb588.uni-karlsruhe.de). This will include extending the current elementary component library but also the componts of one class (motors, sensors, etc.) by adding new component classes e.g. couplings or brakes.

5. ACKNOWLEDGEMENTS

We are grateful for the support of the DFG (Deutsche Forschungsgemeinschaft-German Researach Foundation) within Collaborative Research Center (CRC) 499 'Micro Molding' and CRC 588 'Humanoid Robots'.

6. REFERENCES

Brudniok, S. (2007). Methodische Entwicklung hochintegrierter mechatronischer Systeme am Beispiel eines humanoiden Roboters (Methodical development of highly integrated mechatronic systems by the example of a humanoid robot), IPEK, University of Karlsruhe (TH), Karlsruhe

Devanathan, S. & Ramani, K. (2007). Combining constraint satisfaction and non-linear optimization to enable configuration driven design, Proceedings of ICED, Paris, France, Aug 2007, The Design Society, Glasgow

Kim, J.; Will, P.; Ling, S. R.; Neches, B. Knowledge-rich catalog services for engineering design, Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 17, pp. 349-366, 2003

Yvars, P.-A. (2008). Using constraint satisfaction for designing mechanical systems, Interactive Design and Manufacturing, 2, 3, (August 2008), 161-167, 1955-2513

Aldanondo, M.; Hadj-Hamou, K. & Lamothe, J. Product (2003). Generic modelling for configuration: Requirement analysis and modelling elements, Intelligent agent-based operations management, Hermes Penton Science, ISBN 19039-9643-0 (2003), London