京都大学 / 0048 / 新制・課程博士 / 博士(人間・環境学) / 甲第19793号 / 人博第764号 / 新制||人||184(附属図書館) / 27||人博||764(吉田南総合図書館) / 32829 / 京都大学大学院人間・環境学研究科共生人間学専攻 / (主査)教授 神﨑 素樹, 教授 森谷 敏夫, 教授 石原 昭彦 / 学位規則第4条第1項該当 / Doctor of Human and Environmental Studies / Kyoto University / DFAM

This dissertation develops a general framework for designing stabilizing feedback controllers and observers for dynamics with state/time dependent gains on the control signals and measured outputs. These gains have potential singularity periods but satisfy a technically non-trivial condition referred to as persistence of excitation. A persistence filter design constitutes the primary theoretical innovation of this work around which the controller and observer development is centered. Application areas of singular gain systems considered in this study include robotics, biomechanics, intelligent structures and spacecrafts. Several representative problems involving singular, time-dependent gains are addressed. The specific contributions of this dissertation are outlined as follows: (i) a stabilizing feedback for linear, single-input systems with time-varying, singular control scaling is designed that allows arbitrary exponential convergence rate for the closed-loop dynamics. An adaptive control generalization of this result allows asymptotic convergence in presence of unknown plant parameters. An extension to a special, single-input nonlinear system in the controller canonical form is also proposed. It is proven that this control design results in bounded tracking error signals for a trajectory tracking objective; (ii) observer design for linear, single-output systems with time-varying, singular measurement gains is considered. A persistence filter similar in structure to the control counterpart aids an observer design that guarantees exponential state reconstruction with arbitrary convergence rates; (iii) the observer and controller designs are combined to obtain an exponentially stabilizing output feedback controller for linear, single-input, single-output dynamics with singular gains on both the control and measurements. A novel separation property is established as a consequence. The construction motivates applications to stabilization with reversible transducers which can switch between sensor and actuator modes. The results are verified on two illustrative applications, vibration control using piezoelectric devices and inverted pendulum stabilization with a DC motor. The linear result is further generalized to include state dependent gains; (iv) application of the persistence filter theory to spacecraft attitude stabilization using intermittent actuation is explored. The intermittence is characterized by a time-varying, periodically singular control gain. A nonlinear persistence filter allows construction of an exponentially stabilizing controller and simulations verify convergence with intermittent actuation where conventional proportional-derivative control fails; (v) a stabilization result for a special multi-input, linear system with time-varying matrix control gains is presented. The matrix gain is assumed to be diagonal but allows fewer controls than states subject to a controllability assumption in absence of the singular gain matrix. The single-input adaptive control results are shown to extend to the multi-input case. An application to angular velocity stabilization of an underactuated rigid spacecraft is considered. / text

La commande intermittente a pour but principal d'augmenter le rendement de l'ensemble machine-convertisseur par une réduction des pertes. Dans le but de maintenir le couple moyen de la machine requis par la charge, cette commande augmente le couple de référence et éteint quelques phases pendant chaque période électrique ou chaque période mécanique. Par ce fait, elle réduit les pertes du convertisseur et les pertes ferromagnétiques de la machine. Elle dispose également d'une liberté de sélection des phases activées qui influence des fréquences générées du courant, de la force radiale et du couple. Par conséquent, cette stratégie de la commande intermittente, qui est appelée le glissement, peut éviter l'excitation de fréquences naturelles du stator, ou bien celles de la chaine cinématique. Elle a pour but de réduire les effets négatifs de la commande intermittente vis-à-vis de comportements vibratoire du stator et pulsatoire de la chaine cinématique qui sont respectivement à l'origine du bruit acoustique et des à-coups mécaniques. En fin, la commande intermittente est validée sur des cycles de conduite et peut économiser jusqu’à 5,17 % de l’énergie électrique consommée. / The intermittent control aims to increase the efficiency of the machine-converter system by reducing the losses. In the purpose to maintain the average torque of the machine required by the load, this control increases the reference torque and turns off some phases during each electrical period or each mechanical period. By this fact, it reduces the converter losses and the ferromagnetic losses of the machine. It also has a freedom in the selection of the activated phases which influences the generated frequencies of the current, the radial force and the torque. Therefore, this strategy of the intermittent control, which is called sliding strategy, can avoid the excitation of the natural frequencies of the stator, or those of the drivetrain. It aims to reduce the negative effects of the intermittent control vis-à-vis the vibration of the stator and the pulsation of the drivetrain which are respectively at the origins of acoustic noise and mechanical jerks. Finally, the intermittent control is validated on driving cycles and can save up to 5.17 % of the consumed electrical energy.

<p>Our ability to control unstable objects highlights the sophistication of voluntary motor behaviour. In this thesis, we used an inverted pendulum (i.e., stick) balancing paradigm to investigate the task, learning and context-dependent attributes of unstable object control. We hypothesized that learning would mediate the functional integration of posture and upper limb dynamics and expected changes in the task demand and context to be reflected in the control of posture and the upper limb. We found that training increased the average length of balancing trials and applied this result to further investigate the circumstantial properties of unstable object control.</p> <p>We investigated the temporal structure of posture and upper limb dynamics using statistical and nonlinear time series analysis. We demonstrated that subjects used an intermittent strategy to control the inverted pendulum (Chapters 3 and 5) and found that motor learning modulated the statistical and spatiotemporal attributes of posture (Chapter 5) and upper limb displacements (Chapters 2, 3 and 5). We confirmed the balance control strategy was intermittent by showing that posture and upper limb time series are composed of two independent timescale components: a fast component linked to small stochastic displacements and a slow component related to feedback control (Chapters 3, 4 and 5). The interplay between timescale components was affected by the balancing context (Chapter 3) and task demand (Chapter 4).</p> <p>Chapter 5 investigated the acquisition of individual and coupled posture-upper limb control mechanisms. We found that motor learning involved two independent adaptation processes. The first process modified the timescale composition of posture and upper limb displacements and was followed by incremental changes in the occurrence and duration of correlated posture-upper limb trajectories. In Chapter 6, we investigated learning-mediated changes in multijoint coordination and control. Motor learning led to the flexible, error-compensating recruitment of individual joints and we showed that the preferential constraint of destabilizing joint angle variance was the putative mechanism underlying performance.</p> <p>This thesis performed a detailed examination of unstable object control mechanisms. The undertaken studies have provided knowledge about the acquisition and adaptation of control mechanisms at multiple levels of the motor system. Our data provide convergent evidence that the control mechanisms governing complex human balancing tasks are intermittent and modulated by the task and context.</p> / Doctor of Philosophy (PhD)