Background & Goal: Traditional human joint impedance identification uses time-invariant tech- niques. As joint impedance varies during movement, time- varying (TV) identification techniques can better capture it. Previously, ensemble-based techniques have been used, which require multiple realisations of the same behaviour. Single-realisation-based techniques are more suitable for use in human joint impedance identification, as identical behaviour is hard to guarantee in humans. This study assessed if Bayesian regularisation can improve the TV identification of human joint impedance compared to a recursive least squares sliding window technique. Method & Results: The methods were tested in both a simulation and an experimental study. Four different TV conditions (0.5, 1, and 4 Hz sines, 0.5 Hz block wave) were simulated using an inertia-damper-stiffness model with TV damping and stiffness, with a TV voluntary torque added. In the experimental study three participants were tested in two conditions, with wrist torques following either a 0.5 or 1 Hz sine. Reduced multisine angular perturbation input signals were used. Both methods were able to isolate the voluntary torque component in the 0.5 and 1 Hz sine conditions, but only the Bayesian method could follow the sharp transitions and fast varying behaviour of the block wave and 4 Hz sine. The stiffness estimates of the Bayesian method varied less than the sliding window estimates. Both methods identified voluntary torque more accurately than stiffness, reflecting its larger contribution to the output. Variance- accounted-for (VAF) scores higher than 85% were found; VAF only dropped below 85% when the voluntary torque was not identified correctly. Discussion & Conclusion: The Bayesian method outper- forms the sliding window, finding less variable parameter estimates with improved temporal accuracy, indicating that the Bayesian method is a viable approach for single- realisation time-varying joint impedance identification. It is still limited as it can introduce a bias in the stiffness estimates and because the quality of the parameter esti- mates depends heavily on the input signal used. Future work should address the input dependence and extend the method to more detailed model structures before it is ready for clinical application. ... Read more

Ongoing technological advancements demand the need for increasingly fast and precise motion systems, in which piezoelectric nanopositioners play a key role. However, due to their inherent dynamics, achieving open loop bandwidths close to the first resonance has proven to be challenging. An additional limitation arises from linear control, which is constrained by the Bode's phase-gain relationship and the waterbed effect. These limitations make achieving good tracking and disturbance rejection performance while maintaining stability and robustness at high bandwidth challenging.
Given the difficulty of achieving bandwidths close to the first resonance, this thesis investigates the even more challenging problem of extending the open loop bandwidth beyond the first resonance.
To address the challenge, a linear controller with barely positive stability margins is designed to achieve a bandwidth beyond the first resonance. A reset controller is then added on top of the linear controller to achieve stability and robustness without affecting the gain behavior. Unlike previous applications, here the nonlinearity introduced by the reset controller is extremely high, making the handling of higher-order harmonics particularly challenging. To address this, a novel shaping filter is introduced in order to reduce the magnitude of higher-order harmonics, along with pre- and post-filters which smooth the control action signal and keep it within the input saturation constraints.
The controller is fine tuned and validated in simulation using both frequency and time domain tools, and later implemented in an experimental test setup. Simulations show that achieving stable and robust post-resonance control is feasible, however mismatches between simulations and experiments appear during experiments. It is found that these are caused by an excessive amount of nonlinearities in the reset controller, making DF and HOSIDF tools from simulations unreliable. Nonetheless, experiments demonstrate that post-resonance control is feasible, and achievable only with reset control, outperforming the linear controllers currently used in industry. ... Read more

As wind turbines grow in size to meet global clean energy goals, managing structural loads becomes increasingly important for improving longevity and cost-effectiveness. Individual Pitch Control (IPC), which adjusts each blade’s angle to counteract aerodynamic forces, is a key method for reducing these loads.

The industry-standard Multi-Blade Coordinate (MBC) method demodulates cyclic load signals from each blade into static signals in a non-rotating frame, then controls the blades in tandem based on these static signals. While widely adopted, MBC’s inherently coupled control may limit performance as turbines increase in size. Larger rotors face greater spatial and temporal wind speed variations and blade flexibility, which could make locally acting strategies more attractive.

An alternative approach, Single Blade Control (SBC), regulates each blade independently in a rotating reference frame. By addressing disturbances locally, SBC could potentially reduce pitch actuation effort and extend component life while maintaining similar load reduction performance.

This research compares MBC and SBC through theoretical analysis and simulations, focusing on the trade-off between load reduction and actuation effort, as well as pitch miscalibration effect. The simulation results show that MBC-IPC offers more effective load reduction relative to pitch actuation effort in the majority of the considered operating conditions. However, SBC-IPC remains robust in the presence of pitch miscalibration and tends to achieve better performance in such scenarios.
... Read more