This paper presents a novel data-driven Reference Governor with Model Predictive Control integrating local motion replanning and path following for collision avoidance. Employing a model-free Reference Governor, the proposed solution utilises system knowledge through Bayesian Optimisation to augment predetermined evasive trajectories, minimising path-following errors and simultaneously ensuring obstacle safety margins. A single-track vehicle model in combination with non-linear tyre models is used to capture the vehicle’s dynamics. The optimised control action is the vehicle steering angle, whilst the Reference Governor optimises parameters of a sigmoid reference signal to minimise the tracking error and guarantee safety with respect to obstacles in emergency manoeuvres. The proposed approach is evaluated on a single lane change using a high-fidelity simulation environment and its performance is compared to a baseline controller integrating path following and obstacle avoidance. The results show a 14% reduction of safety critical overshoot, maximising obstacle safety distance and a four times lower controller cycle time compared to the baseline. Furthermore, through a robustness analysis, it is demonstrated that the proposed approach is more robust towards model mismatches and perception-based errors, as seen by average 30% and 40% reductions in near-miss and collision rates.

This thesis explores the integration of hydrogen ( H2) into a residential hybrid energy hub and presents
results based on a case study of such an energy hub located at The Green Village, an open-field lab
environment at Delft university of Technology. The study focusses on analysing real operational data
from the energy hub to assess the performance, behaviour, and integration of the system. The energy
hub combines photovoltaic (PV) power generation, battery storage, and hydrogen-based components
including an AEM electrolyser, hydrogen storage, and a PEM fuel cell.

First, a detailed data analysis of the energy hub components is performed which lays a basis for the
model of the energy hub. Particular attention was paid to the ramp-up and ramp-down dynamics, the
power consumption and generation capabilities, and hydrogen consumption and generation of the
electrolyser and fuel cell, as these affect the overall efficiency and responsiveness of the system. In
addition, insights are given into the real capacity of the energy hub, and a comparison between the
intended operation of the energy hub and the real operation is given.
Following insights gained from the data analysis, a first step toward integrating machine learning into the
Energy Management System (EMS) was taken. One potential improvement identified was the use of a
machine learning algorithm that uses weather forecasts into EMS decision-making for the electrolyser.
This study explores both binary classification and regression models using outside temperature and PV
inverter power as inputs. Since inverter power correlates strongly with solar irradiance, these inputs
were considered sufficient for developing a preliminary machine learning model aimed at enabling
smarter control of the electrolyser.

A data-driven Simulink model of the energy hub was developed to simulate various operational scenarios,
sizes, and edge cases. These simulations revealed how the behaviour of the system changes under
different profiles of PV generation and residential demand, even when total energy consumption remains
constant.

The data results create a solid foundation for the development of the Simulink model to run experiments
with. The model has successfully demonstrated that the integration of hydrogen into the energy hub
facilitates the coverage of seasonal energy demands, adding flexibility, and flattening the energy
demands of the grid. However, it also showed that the system performance is highly sensitive to
operating conditions and system sizing. Recommendations for future energy hub upgrades are provided
through control optimisation and system sizing.

The speed and precision required in the precision motion industry is an ever-growing challenge. Linear control offers intuitive frequency-domain controller design methods that are based on a frequency response function (FRF) of the plant, which is obtained solely from measurement data. Unsurprisingly, linear controllers account for over 90% of the controllers currently used in the industry. However, as linear controllers are subject to inherent performance limitations such as Bode's gain-phase relationship and the waterbed effect, research on nonlinear control solutions to overcome these limitations is ubiquitous. Reset control first showed up in 1958 with the Clegg integrator (CI). According to a sinusoidal-input-describing-function analysis, the CI provides reduced phase lag compared to a linear integrator, suggesting Bode's gain-phase relationship can be overcome and consequently allowing for performance surpassing that of linear control systems. The drawbacks of the CI are the possible emergence of limit cycles and the excitation of high-frequent modes of the system originating from higher-order harmonics of the CI's input caused by discontinuities in the control signal. The generalised-first-order-reset-element-based integrator (GFbI) is the only reset element that can prevent the emergence of limit cycles and can reduce the generation of higher-order harmonics, whilst retaining the advantageous reduced phase lag the CI provides and allowing for closed-loop controller design based on a plant FRF while taking the effects of higher-order harmonics into account, matching design methods of linear controllers. Optimally tuning a reset controller is a complex and time-consuming task. Moreover, established tuning rules are still lacking. This work facilitates designing a reset controller containing a GFbI element by providing two contributions to utilise the potential of the GFbI element to improve upon linear control. The first is a comparative study on the effect of the controller element sequence, aimed at reducing the negative consequences of the higher-order harmonics generated by the reset element. The second is the proposal of an FRF-based optimisation algorithm utilising frequency-domain performance prediction methods to automatically tune a reset controller containing a GFbI element to adhere to the imposed constraints and maximise the benefit gained from the nonlinear element. Validation using a simulated and physical wire bonder showed the algorithm successfully tuned four different reset controllers using two different sequences. The performance of the tuned reset controllers was compared to that of equally-well-tuned linear controllers. In the first use case with the goal of suppressing a dominant vibration in the error signal, the median root-mean-square error was reduced by 16.245%. In the second use case, the goal was to improve the settling time, which was achieved by a median of 9.791%. The best results were achieved using a sequence in which the higher-order harmonics were avoided from passing through the lead filter in open loop, mitigating amplification thereof. The proposed tuning algorithm proved able to tune reset controllers containing a GFbI element such that the performance of linear control based on frequency-domain performance prediction metrics was surpassed, where two use cases confirmed the predicted performance increase through time-domain simulations and experiments on a physical setup.