Methodologies on the Mechatronics Domain Raivo Sell, Priit Leomar Department of Mechatronics, Tallinn Technical University, Tallinn, Estonia [email protected]
/ [email protected]
ABSTRACT Mechatronics requires the integration of a range of technologies, often with conflicting requirements. This problem is particularly important on the ground of the rapid development of the technologies. On the mechatronics field there is several product development methodologies used in nowadays. Mostly they are captured from classical domains such as mechanics or electronics. These methodologies are not very effective in mechatronics, which is cross-domain realm. Therefore we need new approaches for successful mechatronics system design. The paper analyses different approaches based on neural networks, multi-agent systems, genetic algorithms and hybrid systems, etc. from the point of mechatronics system design. All these technologies have many important properties for being used as aids at multi-domain system design, but like always, their success is specific aim dependent. There are no universal solutions, but combining different approaches is possible to find a good solution for most of mechatronical design problems.
1 Introduction Product development is the key factor for the market success of a product. Relevant product development methodology and technology are important pre-requirements for the successful result. This is even more important for the field of mechatronics, which is a complex domain, developed not long ago. Different potential methodologies and their application tools are arising very fast. On the background of the fast technology development, it is important to distinguish perspective tools, which are reasonable to use and cost effective in the mechatronics field. There are several techniques and methods for product development. Concerning very early design phase there are mostly heuristic approaches, although the decisions made in very beginning have a biggest affect for late design process. In this field new techniques and methods can help us to ensure that decisions have made are quality and optimized. Bringing together the demands of the customer and viewpoints of the engineer is very important and great attention has to be paid for that problem. The paper deals with new methodologies on mechatronical product development and presents some relevant analysis. The attention is also paid for tools that are using new technologies.
2 Techniques and Methods for Advanced Mechatronics Modeling Many non-traditional techniques and methods on engineering problem solving domain have been come to the fore recently. One of the reasons is definitely increase of the computing power. These opportunities allow us to solve the engineering tasks, which can’t be described with linear differential equations and are non-deterministic. The techniques for more advanced mechatronics system modeling which is taken into account are followings: • Neural Networks • Multi-Agent Systems • Genetic Algorithms • Fuzzy Planning • etc.
2.1 Artificial Neural Networks Artificial neural networks (ANN) are collections of mathematical models that emulate some of the observed properties of biological nervous systems and draw on the analogies of adaptive biological learning. The key element of the ANN paradigm is the novel structure of the information processing system. It is composed of a large number of highly interconnected processing elements that are analogous to neurons and are tied together with weighted connections that are analogous to synapses. Learning of ANN typically occurs by example through training, or exposure to a truthed set of input/output data where the training algorithm iteratively adjusts the connection weights (synapses). These connection weights store the knowledge necessary to solve specific problems .
Sum f f1(xi,wij)
Activation f f1(xi,wij)
Output f f1(xi,wij)
Wnj Xn Fig. 1. Neuron
O1 Wij Ai
Fig. 2. Neural Network Schema
Neural networks are designed to work with patterns - they can be classified as pattern classifiers or pattern associators. The networks can take a vector (series of numbers), and then classify the vector . Neural Networks are usually used for classification (i.e. market profiles, medical diagnosis, image recognition, etc), forecasting (i.e. future sales, production requirements, energy requirements, weather, etc.) and modeling (i.e. process control, systems control, dynamic systems, signal compression, robot control, etc.) The problems with neural networks are the lack of defining rules to help construct a network given a problem - there are many factors to take into consideration: the learning algorithm, architecture, number of neurons per layer, number of layers, data representation and much more .
2.2 Genetic Algorithms Genetic Algorithms (GA) are basically algorithms based on natural biological evolution. The architecture of systems that implement genetic algorithms is more able to adapt to a wide range of problems. A GA functions by generating a large set of possible solutions to a given problem. It then evaluates each of those solutions, and decides on a "fitness level" for each solution set. These solutions then breed new solutions. The parent solutions that were more "fit" are more likely to reproduce, while those that were less "fit" are more unlikely to do so. In essence, solutions are evolved over time. This way you evolve your search space scope to a point where you can find the solution . A GA needs to know only two things about a problem: 1. The set of possible solutions needs to be coded as a set of strings 2. For each string (solution) there must be a way of measuring how good it is compared to the other strings. The function that performs this task is the “fitness function”. The Fundamental Theorem of Genetic Algorithms:
where M(H, t) number of strings in population 't' with the schema 'H'. f(H) average fitness of the strings with the schema 'H'. F average fitness of the entire population. p1 probability of the schema being destroyed by crossover. p2 probability of the schema being destroyed by mutation. This equation describes exponential like growth of 'good' schema from one generation to the next. GAs are designed to search for, discover, emphasize and breed good solutions (or "building blocks") to a problem in a highly parallel fashion .
2.3 Multi-Agent Systems A multi-agent system (MAS) is a loosely coupled network of problemsolver entities that work together to find answers to problems that are beyond the individual capabilities or knowledge of each entity . More recently, the term multi-agent system has been given a more general meaning, and it is now used for all types of systems composed of multiple autonomous components showing the following characteristics : • each agent has incomplete capabilities to solve a problem
• • •
there is no global system control data is decentralized computation is asynchronous
3 Complex Mechatronics System Modeling with Artificial Intelligence Modeling complex mechatronics system like most of the real-life products are, there is not reasonable to concentrate only one particular method described previous chapter. The mechatronics problem needs more complex approach and one of the possibilities is to combine these techniques and methods. The ongoing research is studying the best way to combine new techniques. The research result so far is the specification of the methodology for effective early design. The methodology exploits UML language to define a mechatronical problem and the constraints. The problem definition is intuitive – the functionality and the behavior of the future system are described. Genetic algorithms are used to find optimal solution for described problem. As mentioned above, genetic algorithms have two main requirements - possible solutions needs to be coded strings and there has to be way to measure the solution. The combinations of the components are possible solutions which have to be tested. These different combinations can be represented by different strings. „Fitness function“, which is needed for measuring each solution, is evaluated from the system description composed by the engineer. To evaluate the best fitness function the neural network schema is used.
Problem Definition UML
DB of Components & principles
Adjustment GA Modifications
Specification Block Diagram
Fig. 3. Methodology diagram
The GA evaluates the set of building blocks and a value for fitness is assigned to each solution (set of building blocks) depending on how close it is to solving the problem. Those solutions with a higher fitness value are more likely to reproduce offspring. If the new generation contains a solution that produces an output that is close enough or equal to the desired answer then the problem has been solved. If this is not the case, then the new generation will go through the same process as their parents did. This will continue until a solution is reached . The solution is a unique combination of pre-defined components and/or completely new components created concurrently consisting of combined sets of predefined properties as carriers of certain component sub-functions. The result is represented, as a block diagram of the mechatronic system compatible in structure with bond graphs. The overall result of the process is a specification of the system – optimized block diagram that meets the initial requirements, specified at
the very beginning by UML. Onward design process is not supported by this methodology at the moment. This can be done by well-known software like 20-sim, Dymola, etc.
4 Conclusions A shot overview of techniques and methods for advanced mechatronics modeling is described and some guidelines are proposed. Currently the subject of mechatronic system design has taken a new dimension since many state-of-art technologies have become widely available and researchers have turned their attention to emerging technologies that allow engineers to understand and model multidisciplinary systems. Important point is, to design the system, the customer actually needs. Here the new systems based on artificial intelligent methods can play a significant role.
5 References 1. (1997) Artificial Neural Network. Battelle Memorial Institute 2. Matthews J. (2000) Introduction to Neural Networks. In: Generation5 3. Durfee, E.H., Lesser, V.R. and Corkill, D.D. (1989) Trends in Cooperative Distributed Problem Solving. In: IEEE Transactions on Knowledge and Data Engineering, KDE-1(1), pages 63-83. 4. Jennings, N.R., Sycara, K. and Wooldridge, M. (1998) A Roadmap of Agent Research and Development. In: Autonomous Agents and Multi-Agent Systems Journal, N.R. Jennings, K. Sycara and M. Georgeff (Eds.), Kluwer Academic Publishers, Boston, Volume 1, Issue 1, pages 7-38. 5. Tschichold-Gürman, N., Siegwart, R., Bühler, P. (2000) Smart Mechatronics. Unterlangen zur Vorlesung 30-608, ETH Zürich 6. Matthews J. (2000) An Introduction to Genetic Algorithm and Genetic Programming. In: Generation5 7. Goldberg, E. (1989) Genetic Algorithms in Search, Optimization and Machine Learning. David Addison-Wesley 8. R. Sinha, V.C. Liang, C.J.J. Paredis, and P.K. Khosla, (2001) Modeling and Simulation Methods for Design of Engineering Systems, Journal of Computing and Information Science in Engineering. Vol. 1, pp. 84-91
Acknowledgment The work is carried out by the support of Estonian Science Foundation Grant No 5177 and Estonian Ministry of Education & Science Grant No 0142506s03