Affiliation: University of Stuttgart
Title: From LPV Control Towards a General Synthesis Framework (slides)
Abstract: Linear parameter-varying controller synthesis is a well-developed technique which has found many applications also in an industrial context. However, many approaches are limited to the design of controllers which adapt themselves to on-line measurable time- varying parameters. Methods that rely on the representation of systems in feedback with the scheduling parameters are much less developed but considerably more flexible. They permit a seamless extension to the design of controllers that adapt themselves to changes in the system dynamics or even in the interconnection topology. In this tutorial presentation we highlight the core mechanisms behind such synthesis techniques and we provide illustrative examples for their surprising diversity.
Affiliation: Vrije Universiteit Brussel
Title: Frequency domain measurement and identification of linear time-varying systems (slides)
Abstract: This talk explores the advantages of the frequency domain for the identification of time-varying systems. First, we will discuss how valuable and sensible information can be extracted from these systems by using well-designed periodic excitation signals and, surprisingly, by interpreting the measurement results in the frequency domain. The spectral response allows for a visual detection and classification of time variation, nonlinear distortion and noise. Also, a non-parametric estimate of the time-varying transfer function is extracted. Then, the estimation of parametric models is discussed. Discrete- and continuous-time linear time-varying systems are identified by cleverly switching between the time and the frequency domain. A consistent estimator is defined to take input and output noise into account. Finally, opportunities are discussed to use kernel based regression to alleviate the difficult task of model structure selection.
These concepts and techniques will be illustrated on simulation and measurement examples.
Affiliation: European Space Agency
Title: Robust & LPV Control In European Space Systems Status, Advances & Potentials (slides)
Abstract: Despite that Robust and Linear Parameter Varying (LVP) systems theory and design tools have made a substantial technological progress in the last decades, their industrial practitioners, within the field of space applications, remains limited to a select group originating from the robust control community (Why?). Only when forced by the challenging programmatic and mission needs, transitions were made to resort to the formal design framework provided by the robust control community. We shall present some of the success stories of practical high performance ESA missions (LISA Pathfinder, Rosetta, Mars Express, Bepi-Colombo, Ariane V) being realized thanks to robust control. We shall give some insight how we are working towards formalising the use robust control machinery to maps specific scientific needs towards the realisation of high precision pointing systems. The connection and transition from robust control to LPV systems is justified by the complexity of the future ESA high performance missions. These need to remain at reduced development cost. We shall illustrate some of the envisaged ESA missions that do not only ask for robustness, but also for in real-time optimized adaptation in the view autonomy. We shall see how the transition from Robust to LPV control is planned as potential solution in the realization of more efficient re-usable launch systems, the next generation of planetary high precision entry descent and landing systems. The same holds for the future ESA science and operation missions for in orbit autonomous rendezvous, robotic assembly, refueling and mated flight operations, covering autonomy, high precision pointing, agility, adaptation needs. Despite the seemingly slow industrialisation of LPV control systems in space applications, we shall illustrate several motivating examples revealing the our experience and understanding in the potentials of LPV modelling, analysis and design techniques. We conclude with the exploring some future avenues in the field.
Affiliation: Technische Universität Hamburg
Title: Fast Nonlinear Predictive Control via quasi-LPV Models (slides)
Abstract: Due to their considerable practical importance, fast nonlinear predictive control schemes have been receiving considerable attention over the last two decades. This talk will present a recently developed approach to nonlinear MPC that is based on a quasi-LPV model of the plant. The nonlinear optimization problem is solved by repeatedly optimizing the input sequence for an LTV system (typically this amounts to solving one or two QP per sampling period). Stability can be guaranteed via terminal constraints. As for the construction of suitable quasi-LPV models, we consider two cases. When a first-principles model of the nonlinear plant is available, we show how an LPV model can be constructed following a velocity-based approach. Alternatively, we present a data-driven approach to constructing an LPV model online that is based on a truncated Koopman operator representation. The real-time capability of the proposed methods is illustrated with experimental results.
Affiliation: Delft University of Technology
Title: System identification of human joint dynamics – beyond LTI (slides)
Abstract: Humans interact with their environment and constantly correct for disturbances while executing complex movements. The central nervous system continuously regulates joint dynamics to optimize this interaction. System identification is an emerging tool in human movement control allowing to assess system behavior, like joint dynamics, in a quantitative way. Time-invariant system identification techniques have demonstrated wide-ranged adaptation between postural tasks. In order to assess how humans change joint dynamics and understand the origin of this adaptation, time-varying system identification techniques are required in combination with experimental protocols that provoke time-varying behaviour. Recently several novel time-varying identification algorithms have been developed with specific applications in mind, resulting in each having their own sets of assumptions and limitations. Such algorithms are typically evaluated in academic simulations that abide by these restrictions. Few studies have actually quantified time-varying joint impedance during movement or the transition between conditions. Major reasons are the experimental challenge of estimating time-varying joint impedance during movement as well as the lack of standardized time-varying identification techniques for these applications. Ideally the time-varying algorithm is able to track fast changes, at least as fast as humans can adapt, can be applied on a single trial of data to investigate trial-by-trial changes in human control actions, and requires little to no prior information about the time variance or model structure. Time-varying system identification techniques are highly valuable to investigate for example the impedance of the knee joint during walking. Such data can be used to investigate pathological gait, to evaluate the effect of therapy and to tune active biomimetic knee prostheses.