TU Darmstadt / ULB / TUbiblio

Co-evolution of morphology and control in developing structures

Smalikho, Olga :
Co-evolution of morphology and control in developing structures.
[Online-Edition: http://tuprints.ulb.tu-darmstadt.de/6097]
Technische Universität Darmstadt , Darmstadt
[Dissertation], (2017)

Offizielle URL: http://tuprints.ulb.tu-darmstadt.de/6097

Kurzbeschreibung (Abstract)

The continuous need to increase the efficiency of technical systems requires the utilization of complex adaptive systems which operate in environments which are not completely predictable. Reasons are often random nature of the environment and the fact that not all phenomena which influence the performance of the system can be explained in full detail. As a consequence, the developer often gets confronted with the task to design an adaptive system with the lack of prior knowledge about the problem at hand. The design of adaptive systems, which react autonomously to changes in their environment, requires the coordinated generation of sensors, providing information about the environment, actuators which change the current state of the system and signal processing structures thereby generating suitable reactions to changed conditions. Within the scope of the thesis, the new system growth method has been introduced. It is based on the evolutionary optimization design technique, which can automatically produce the efficient systems with optimal initially non-defined configuration. The final solutions produced by the novel growth method have low dimensional perception, actuation and signal processing structures optimally adjusted to each other during combined evolutionary optimization process. The co-evolutionary system design approach has been realized by the concurrent development and gradual complexification of the sensory, actuation and corresponding signal processing systems during entire optimization. The evolution of flexible system configuration is performed with the standard evolutionary strategies by means of adaptable representation of variable length and therewith variable complexity of the system which it can represent in the further optimization progress. The co-evolution of morphology and control of complex adaptive systems has been successfully performed for the examples of a complex aerodynamic problem of a morphing wing and a virtual intelligent autonomously driving vehicle. The thesis demonstrates the applicability of the concurrent evolutionary design of the optimal morphological configuration, presented as sensory and actuation systems, and the corresponding optimal system controller. Meanwhile, it underlines the potentials of direct genotype – phenotype encodings for the design of complex engineering real-world applications. The thesis argues that often better, cheaper, more robust and adaptive systems can be developed if the entire system is the design target rather than its separate functional parts, like sensors, actuators or controller structure. The simulation results demonstrate that co-evolutionary methods are able to generate systems which can optimally adapt to the unpredicted environmental conditions while at the same time shedding light on the precise synchronization of all functional system parts during its co-developmental process.

Typ des Eintrags: Dissertation
Erschienen: 2017
Autor(en): Smalikho, Olga
Titel: Co-evolution of morphology and control in developing structures
Sprache: Englisch
Kurzbeschreibung (Abstract):

The continuous need to increase the efficiency of technical systems requires the utilization of complex adaptive systems which operate in environments which are not completely predictable. Reasons are often random nature of the environment and the fact that not all phenomena which influence the performance of the system can be explained in full detail. As a consequence, the developer often gets confronted with the task to design an adaptive system with the lack of prior knowledge about the problem at hand. The design of adaptive systems, which react autonomously to changes in their environment, requires the coordinated generation of sensors, providing information about the environment, actuators which change the current state of the system and signal processing structures thereby generating suitable reactions to changed conditions. Within the scope of the thesis, the new system growth method has been introduced. It is based on the evolutionary optimization design technique, which can automatically produce the efficient systems with optimal initially non-defined configuration. The final solutions produced by the novel growth method have low dimensional perception, actuation and signal processing structures optimally adjusted to each other during combined evolutionary optimization process. The co-evolutionary system design approach has been realized by the concurrent development and gradual complexification of the sensory, actuation and corresponding signal processing systems during entire optimization. The evolution of flexible system configuration is performed with the standard evolutionary strategies by means of adaptable representation of variable length and therewith variable complexity of the system which it can represent in the further optimization progress. The co-evolution of morphology and control of complex adaptive systems has been successfully performed for the examples of a complex aerodynamic problem of a morphing wing and a virtual intelligent autonomously driving vehicle. The thesis demonstrates the applicability of the concurrent evolutionary design of the optimal morphological configuration, presented as sensory and actuation systems, and the corresponding optimal system controller. Meanwhile, it underlines the potentials of direct genotype – phenotype encodings for the design of complex engineering real-world applications. The thesis argues that often better, cheaper, more robust and adaptive systems can be developed if the entire system is the design target rather than its separate functional parts, like sensors, actuators or controller structure. The simulation results demonstrate that co-evolutionary methods are able to generate systems which can optimally adapt to the unpredicted environmental conditions while at the same time shedding light on the precise synchronization of all functional system parts during its co-developmental process.

Ort: Darmstadt
Fachbereich(e)/-gebiet(e): 18 Fachbereich Elektrotechnik und Informationstechnik > Institut für Automatisierungstechnik und Mechatronik > Regelungsmethoden und Robotik
Hinterlegungsdatum: 26 Mär 2017 19:55
Offizielle URL: http://tuprints.ulb.tu-darmstadt.de/6097
URN: urn:nbn:de:tuda-tuprints-60978
Gutachter / Prüfer: Adamy, Prof. Jürgen ; Sendhoff, Prof. Bernhard
Datum der Begutachtung bzw. der mündlichen Prüfung / Verteidigung / mdl. Prüfung: 19 Januar 2017
Alternatives oder übersetztes Abstract:
AbstractSprache
Stets steigende Anforderungen an neuartige technische Lösungen hängen oft mit der Komplexität deren Aufgaben zusammen. In vielen Bereichen der modernen Technik kommt es zu unvorhersehbaren und erschwerten Umweltbedingungen. Hierbei steht der Entwicklungsingenieur vor der Herausforderung ein adaptives und invariantes Gesammtsystem zu entwickeln, welches auch dann funktionsfähig ist, wenn die Umgebungsbedingungen stark von den Standardwerten abweichen. Die Schwierigkeiten dabei sind häufig sowohl die unbekannten Verhältnisse der Umgebung als auch deren Einfluss auf das Systemvehalten. Für solche Anwendungsgebiete wird die Entwicklung von extrem robusten technischen Applikationen benötigt. Die gehören zu der Klasse der adaptiven Systeme und verfügen über spezielle mechanische und sensorische Vorrichtungen um die Veränderungen der Umgebungsbedingungen wahrnehmen zu können und dementsprechend den Zustand des Systems durch die vorhandenen Aktuatoren optimal anzupassen. Die optimale Konfiguration der Morphologie und die Regelung des adaptiven Systems ist meistens unbekannt und wird anhand des bereits vorhandenen Vorwissen über das Systemverhalten manuel gewählt. Im Rahmen der vorliegeneden Dissertation wurde ein neues Konzept entwickelt zur automatischen Generierung der optimalen Konfiguration der Sensorik, Aktuatorik und Regelung des Systems, basierend auf den Methoden der evolutionären Algorithmen. Die Entwicklung des Gesammtsystems, bestehend aus den sensorischen und aktuatorischen Komponenten sowie dem Regler, findet hierbei parallel mithilfe von einem kombinierten inkrementalen evolutionären Algorithmus statt. Die Optimierung fängt mit einer möglichst einfachen Systemlösung, idealerweiser mit einem einzigen Sensor und Aktuator und einer sehr vereinfachten Reglerstruktur an. Im Laufe des weiteren Optimierungsverlaufs, basierend auf einem im Rahmen der Dissertation entwickeltes Wachstummodels, nimmt das System schrittweise an Komplexität zu mit Hilfe einer graduellen Erweiterung der Morphologie und Signalverarbeitungsstruktur. Der neu vorgestellte co-evolutionäre Algorithmus wurde an den Beispielen eines simulierten adaptiven Tragflügelprofils und dem vereinfachten Model eines autonoum fahrendes Fahrzeug erfolgreich appliziert. Die Simulationsergebnisse der beiden Beispielanwendungen zeigen, dass die co-evolutionären inkrementalen Methoden den Entwicklungsprozes der realen, komplexen, adaptiven Anwendungen wirksam vereinfachen und automatisieren können. Die unterschiedlichen Komponenten der dabei entstehenden Lösungen sind, ähnlich zu biologischen Systemen, evolutionär optimal aufeinander abgestimmt und effektiv hinsichtlich der Hardware- und Softwareressourcen.Deutsch
Export:

Optionen (nur für Redakteure)

Eintrag anzeigen Eintrag anzeigen