Towards a Design Methodology for Self-optimizing Systems

Towards a Design Methodology for Self-Optimizing Systems

Author: Jurgen Gausemeier, Ursula Frank, Andreas Schmidt, and Daniel Steffen

Abstract: Self-optimizing systems will be able to react autonomously and flexibly to changing environments. They will learn and optimize their performance during their product life cycle. The key for the design of self-optimizing systems is to utilize reconfigurable system elements, communication structures and experienced knowledge. The concept of active principles of Self-Optimization is an important starting point.

5.2 Self-optimizing Systems

In terms of software engineering, this involves distributed systems of interacting agents:

“An agent is an autonomous, proactive, cooperative and extremely adaptive function module. The term “autonomous” implies an independent control system, which it proactively initiates actions.¬† Agents are regarded as function modules, which work in cooperation or competition with one another. “Adaptive” refers to a generic behavior at run time, which may also, for example, include learning capabilities. A function module is taken to be a heterogeneous subsystem with electronic, mechanical and IT-related components.”

Combining the paradigm of intelligent agents with mechatronic structures makes it possible to construct self-optimizing mechanical engineering systems.

“Self-optimization of a technical system refers to the endogenous modification of the target vector due to changing environmental conditions and the resulting target-compliant, autonomous adaptation of the structure, the behavior and the parameters of this system. Self-optimization, therefore, far exceeds known control and adaptation strategies. Self-optimization enables empowered systems with inherent “intelligence,” which are able to react autonomously and flexibly to changing environmental solutions”

The examination of self-optimizing systems is based on four aspects:

  1. the target system (e.g. a hierarchical system of targets or a target vector)
  2. the structure (i.e. topology of mechanical components, sensors and actuators),
  3. the behavior and
  4. the parameters.

The following principles determine Self-optimization:

  • Reconfiguring system elements
    • An adaptation to different environmental situations presupposes the presence of system elements which can be reconfigured or which can interact with other system elements in different combinations. In a chassis, for example, redundant actors (mechanical feather/spring, pneumatic spring, hydraulic cylinder) are used. They are used together in different ways (parallel/in series) to absorb different stimuli.
  • Communication
    • System elements behave like software agents. They pursue their targets according to the target system of the overall system. They achieve these targets by negotiations and co-operation with other system elements. For adjustment processes and negotiation principles, generic patterns are defined. Examples for communication relations are the chassis reconfiguration or an arrangement about the right of way between two vehicles.
  • Experienced knowledge
    • In order to ensure the optimal behavior in unknown operating situations or in situations that are not described in models, experienced knowledge embodied as cases is stored and used again in similar situations. It is shared with other systems, as well. So-called active principles of Self-optimization describe generic patterns of behavior, which can be used in many situations. Especially the use of active principles of Self-optimization creates greater opportunities and enables absolutely new functionalities.

 Active Principles of Self-optimization

Active principles of Self-optimization are meant to be a combination of a technical system and the influences on the technical system (the environment, the user, or other system elements) and adaptation components. The technical system consists of a structure model, in terms of the topology of mechanical components or the hierarchy of multi-agent systems, a behaviour model, such as differential equations or planning and learning systems, and the parameterization of the models. A target system prescribes the current goals which the technical system tries to achieve. In this way the active principle of Self-optimization allows for the endogenous
modification of the technical system according to changing influences, as well as for target-compliant, autonomous adaptation of parameters, behaviour and structure. Adaptation strategies and adaptation tactics define the kind and process of modifications for long-term and medium- to short-term adaptation to application scenarios. Adaptation costs represent the effort of adaptation in terms of energy consumption, time-delays, monetary payments and the like.



Leave a Reply