Titel Deelnemers "Korte inhoud" "Partial decomposition of nonlinear Euler–Lagrange equations with a state transform" "Michiel De Roeck, Jasper Juchem, Guillaume Crevecoeur, Mia Loccufier" "The Euler-Lagrange (EL) formalism is extensively used to describe a wide range of systems. The choice of the generalised coordinates is not unique and influences the intricacy of the coupling terms between the equations of motion. A coordinate transformation can vastly reduce this complexity, yielding a (partially) decoupled system description. This work proposes a state transform of the original EL equation resulting in an identity inertia matrix. Since the centrifugal and Coriolis terms originate from the derivatives of the inertia (or mass) matrix, the matrix containing these terms is either reduced to a skew-symmetric one or in a limited number of instances reduced to zero. In contrast to prior work, that relied on solving a set of ordinary differential equations, the transformation matrix can be determined using an algebraic equation. As a result, the suggested methodology yields an easy-to-use and powerful tool for reducing and (partially) decoupling any equations of motion expressed in the EL formalism." "An optimal wind farm operation strategy for the provision of frequency containment reserve incorporating active wake control" "Nezmin Kayedpour, Jeroen De Kooning, Arash Ebneali Samani, Farjam Kayedpour, Lieven Vandevelde, Guillaume Crevecoeur" "This study proposes a novel operation strategy for wind farms' optimal Frequency Containment Reserve (FCR) provision that simultaneously distributes FCR and optimally controls wake formation. The power reserve allocation is dynamically decided at the wind farm supervisory control level, considering the intermittent wind power and direction, grid frequency stochasticity, and the aerodynamic complexity of the wake. A two-stage stochastic programming approach supports decision-making for an optimal contribution to day-ahead energy/FCR markets considering sub-hourly wind power and grid frequency uncertainty. Moreover, a novel method is used to reduce the computational complexity by employing a data-driven surrogate model of wake formation in the optimizer. This surrogate model consists of a neural network trained on the Gauss-Curl-Hybrid wake model in FLORIS. This deep learning approach allows fast estimation of the wake control parameters, i.e., the optimal yaw angles and axial induction factors. Then, a coevolutionary-based multi-objective particle swarm optimization searches for the optimal deloading of the WTs and maximizes the total power production and kinetic energy while minimizing wake. The performance of the proposed algorithm is evaluated on the C-Power wind farm layout in the North Sea. Simulation results demonstrate its effectiveness in improving the wind farm's overall performance for different operational conditions." "KeyCLD : learning constrained Lagrangian dynamics in keypoint coordinates from images" "Rembert Daems, Jeroen Taets, Francis wyffels, Guillaume Crevecoeur" "We present KeyCLD, a framework to learn Lagrangian dynamics from images. Learned keypoints represent semantic landmarks in images and can directly represent state dynamics. We show that interpreting this state as Cartesian coordinates, coupled with explicit holonomic constraints, allows expressing the dynamics with a constrained Lagrangian. KeyCLD is trained unsupervised end-to-end on sequences of images. Our method explicitly models the mass matrix, potential energy and the input matrix, thus allowing energy based control. We demonstrate learning of Lagrangian dynamics from images on the cl_bnmsqnk pendulum, cartpole and acrobot environments. KeyCLD can be learned on these systems, whether they are unactuated, underactuated or fully actuated. Trained models are able to produce long-term video predictions, showing that the dynamics are accurately learned. We compare with Lag-VAE, Lag-caVAE and HGN, and investigate the benefit of the Lagrangian prior and the constraint function. KeyCLD achieves the highest valid prediction time on all benchmarks. Additionally, a very straightforward energy shaping controller is successfully applied on the fully actuated systems." "Soft switching multiple model predictive control with overlapping cross-over time strategy in an industrial high speed pick and place application" "Babak Mehdizadeh Gavgani, Arash Farnam, Foeke Vanbecelaere, Jeroen De Kooning, Kurt Stockman, Guillaume Crevecoeur" "In industrial machinery, position-controlled systems executing repetitive tasks are of significant importance. In such mechanisms, the load-side inertia and torque vary with the angular position of the driven axis, thus imposing a significant nonlinearity in the system dynamics. Linear controllers have a relatively low computational time, but are unable to achieve the required performance over the whole operating region. Nonlinear controllers, such as nonlinear model predictive control (NLMPC), demand an extensive computational time that is difficult to bound. The current Soft Switching Multiple MPC (SSM-MPC) techniques rely solely on the gap metric or weighting techniques to guarantee soft switching between controllers. However, in nonlinear systems with fast dynamics, it may result in poor performance and/or high computational time. In this work, we have addressed these issues by introducing an overlapping cross-over time strategy in the switching unit of the SSM-MPC to ensure a smooth transition when switching in adjacent regions. The key idea behind the overlapping cross-over time strategy is introducing a delay i.e. cross-over time during switching. The crossover time helps to improve the completeness and non-redundancy of the model bank. The closed loop system stability is guaranteed using Lyapunov theory and a general switching stability proof is provided. The proposed SSM-MPC performance is benchmarked against the available SSM-MPC method and a single linear MPC on a high speed industrial pick and place machine through simulations and experiments. The results verify the superior performance of the presented SSM-MPC with 37.67% reduction in tracking error and 20.36% decrease in control signal oscillations at the highest operating speed of the machine compared to the available SSMMPC technique without cross-over time. Moreover, in contrast to the nonlinear MPC, it is proven that the computational load of the SSM-MPC is 77.5% lower, which allows implementation in a standard industrial controller." "Efficiency optimisation and converterless PV integration by applying a dynamic voltage on an LVDC backbone" "Hakim Azaioud, Arash Farnam, Jos Knockaert, Lieven Vandevelde, Jan Desmet" "A low-voltage DC (LVDC) backbone with a battery energy storage system (BESS) and distributed photovoltaics (PV) is proven to be a more efficient alternative compared to the traditional AC architecture. However, previous research has also proven the fact that the benefit strongly depends on the operating voltage level. In this study, a dynamic backbone voltage on architectures with and without distributed maximum power point trackers (MPPT) is investigated. The dynamic voltage is driven by a multi-objective optimisation algorithm that will minimise the cable and converter loss. Applying the dynamic voltage with MPPT leads to a massive decrease of the loss compared to a static voltage with MPPT. Although, if the optimisation objective is extended with the minimisation of the PV curtailment loss, the MPPT can be eliminated without causing a considerable curtailment loss while maintaining a high efficiency. The sensitivity analysis showed that even when the cable is longer, the orientations of the PV distributed systems are different or the PV system is exposed to dynamic shading, dynamic voltage without maximum power point is still competitive compared to the alternatives. However, heavy static shading conditions could cause a notable shift of the MPP making it very hard to track this operating point by the algorithm. The elimination of the MPPT leads to a simpler design and installation, lower investment costs and a reduction of the raw material use. The proposed strategies and architectures are hence favourable to be applied in LVDC backbones massively fed by roof-mounted or building-integrated PV systems. This study provides the general approach which could be extended with distributed electric vehicle chargers." "Improving energy flexibility in photovoltaic-battery systems through switching reinforcement learning control" "Siebe Paesschesoone, Nezmin Kayedpour, Carlo Manna, Guillaume Crevecoeur" "The electric power sector is undergoing radical changes to enhance sustainability. A main part of this transformation involves the integration of more renewable energy sources (RES) into the power grid, emphasizing the critical need for increased grid flexibility. However, current power grid controllers face challenges in achieving this flexibility due to the unpredictable nature of RES and load, compounded by the complex multi-physical nature of the entire energy system. To tackle these challenges and improve existing controllers, this paper introduces a novel control strategy employing Reinforcement Learning (RL). The proposed strategy is validated through its application to a Photovoltaic (PV)-battery system, where the RL agent learns through interactions with the environment, guided by a reward function based on energy costs. Recognizing the difficulty of maintaining optimality in dynamic environments with single-model RL, our approach advocates for a multi-model RL framework. The framework incorporates a clustering algorithm to identify the optimal number of models. Additionally, we introduce a switching RL strategy involving an offline phase for dataset partitioning and RL model training, followed by an online phase where a real-time switching decision unit selects the optimal RL model, dynamically adapting to current operating conditions. Applying our proposed approach to the PV-battery system demonstrates an 8.15% increase in cost efficiency compared to a conventional single-model RL framework." "A posteriori control densities : imitation learning from partial observations" "Tom Lefebvre, Guillaume Crevecoeur" "This paper treats a special case of the Imitation from Observations (IfO) problem. IfO is a generalisation of Imitation Learning from state-only demonstrations. Our treatment of IfO considers the case of feature-only demonstrations. This means that the full state is inaccessible for inference, and imitation must occur on the basis of a limited set of features. We refer to this setting as Imitation from Partial Observations (IfPO). This scenario has the advantage of allowing to address a wider variety of demonstrations, as well as solving the problem of heteromorphic student and teacher. We set out for policy learning methods that extract an executable state-feedback policy, directly from those features, which in the literature is known as Behavioural Cloning. In this theoretical work, we formalize the rational inference model of the student decision maker, devoted to imitation, as a controlled Hidden Markov Model. The IfPO problem is then reformulated as a Maximum Likelihood Estimation problem and treated using Expectation-Maximization. We name the resulting fixed point iterations A Posteriori Control Densities. We compare the presented approach to existing methods in the field and identify potential directions for further development, such as an extension to unknown transition and emission models." "Improved magnetic drug targeting with maximized magnetic forces and limited particle spreading" "Rikkert Van Durme, Guillaume Crevecoeur, Luc Dupré, Annelies Coene" "Differential flatness of slider-pusher systems for constrained time optimal collision free path planning" "Tom Lefebvre, Sander De Witte, Thomas Neve, Guillaume Crevecoeur" "In this work we show that the differential kinematics of slider-pusher systems are differentially flat assuming quasi-static behaviour and frictionless contact. Second we demonstrate that the state trajectories are invariant to time-differential transformations of the path parametrizing coordinate. For one this property allows to impose arbitrary velocity profiles on the slider without impacting the geometry of the state trajectory. This property implies that certain path planning problems may be decomposed approximately into a strictly geometric path planning and an auxiliary throughput speed optimization problem. Building on these insights we elaborate a numerical approach tailored to constrained time optimal collision free path planning and apply it to the slider-pusher system." "A supervisory learning control framework for autonomous & real-time task planning for an underactuated cooperative robotic task" "Sander De Witte, Tom Lefebvre, Thijs Van Hauwermeiren, Guillaume Crevecoeur"