Modelling Fluid Power Systems is commonly done using first principle models which can be computationally demanding or time-consuming to fit. This paper introduces and explores the use of three Neural Network Structures in Fluid Power Systems modelling, referred to as Pressure Derivative Neural Network Structure, Internal Energy Neural Network Structure and Continuity Based Neural Network Structure. The structures are Lagrangian Neural Network like structures, which can parametrise arbitrary Lagrangians using Neural Networks. Lagrangian Neural Networks have predominantly been applied in mechanical systems, with no prior applications in Fluid Power Systems. This approach is applied to a valve-controlled cylinder drive as a case study. Datasets are both generated from a simulation and a lab setup. Primarily, a training scheme comparing states integrated ahead in time, using NN-predicted derivatives, are investigated. The results show promising performance training on simulation data, also including noise. However, training on the lab data proved to inhibit the Neural Network Structures from learning the dynamics. An investigation into the differences between simulation and lab data revealed that the discrepancies affected the Internal Energy Neural Network Structure and Continuity Based Neural Network Structure differently. Both structures were most affected by inaccurate valve position and velocity data.
With the reduction in grid inertia due to penetration of non-rotating generation sources in modern power grids, flexible operation of hydropower plants has become necessary. Variable Speed Hydropower Plants (VSHP) have emerged as an alternative to this. For quick frequency support in the grid post-disturbance, droop of the generation source plays major role as it determines the ramping of generation. However, there lies a gap on studies involving variable droop implemented in VSHPs, and its capability of frequency support. Also, the variable droop during operation demands very fast guide vanes operation, which could lead to violation on hydraulic pressures and flows limits. This paper implements a Non-linear Model Predictive Control (NLMPC) algorithm for operating a VSHP in varying droop scenario, offering a dynamic droop enabling the VSHP to pick up more load when needed. The NLMPC imposes hydraulic limits as constraints on the optimization problem whose main objective is to provide fast power response to balance the grid frequency. The controller also simultaneously ensures stable operation of hydraulic system of VSHP which is decoupled from the grid. The simulation is run on IEEE 9 bus power system injected by a VSHP and is compared with base case of same test system for different magnitudes of step load increment and contingency. The results yield a maximum of 3.5% reduction in frequency nadir and significant reduction in Rate of Change of Frequency (RoCoF) in the grid. This reduction in system frequency nadir and RoCoF can be claimed as significant contribution in grid support. In addition to this, capability of NLMPC to control the guide vanes leading to hydraulic control simultaneously when the VSHP is slowing down to provide optimal frequency support to the grid highlights its suitability in hydropower control applications.
A moving average smoothing method for extraction of cycles in time series data is described, with focus on obliquity cycles in fossil data. Since obliquity cycles affect sea temperature, they have left traces in oxygen isotope 518O(t) values over time, as found in deep-sea drilling projects. Such values are commonly used as proxy for temperature, which is an important driver for evolutionary change. The proposed method is more generally intended for cases where the environmental driver of phenotypic evolution has been found to include obliquity cycles, either by power spectrum analysis or simply by inspection of raw or smoothed time series. The method gives improved mean trait predictions and better understanding when applied on stickleback fish fossil data from around 10 million years ago. The possibility of extracting obliquity cycles information will depend on the dynamics of the time series, and the method is thus not universally applicable. It may, however, be possible to adapt the size of the moving window to problems under study, or possibly to obtain improved predictions by inclusion of a sinusoidal component in the mean trait prediction modeling.
Determination of the pose of workpieces is important for robotic applications in manufacturing, including handling, assembly, machining and welding. Established methods based on 3D sensors may fail for workpieces with highly reflective materials. In this paper, we take advantage of recent development in machine learning to determine the pose of reflective workpieces without the use of depth data. Our proposed method is based on deep iterative matching of image data of the workpiece with a computer-aided design model. Starting with an initial estimate of the workpiece pose, the method iteratively aligns the computer-aided design model projections with an image of the actual workpiece, adjusting the pose until computer-aided design model matches the image of the workpiece. The deep learning-based approach optimizes this alignment by updating the pose estimate at each iteration, achieving high precision even for geometrically complex or reflective surfaces. This refinement process enhances accuracy in robotic applications where precise workpiece positioning is critical, such as in automated welding and assembly tasks. We use photorealistic rendering to create two datasets for pretraining the network, which reduces both training time and the need for real labeled data. After the network is trained on synthetic data, it is fine-tuned and tested on real images of reflective aluminium workpieces. We show that the proposed deep iterative matching method outperforms established methods based on iterative closest point with two 3D scanners due to large errors in the scans caused by reflections.
This paper presents a data-driven framework utilising neural networks to approximate the Koopman operator for a discrete-time representation of an electrically actuated two-link mechanical system, enabling the application of linear control techniques. A validated dynamic model is used to generate training data. The Physics Informed Fully Embedded Koopman (PIFEK) framework embeds the original state space directly into the Koopman space, leading to sensible tuning of the controllers. The tuning was done using bandwidth analysis of the Koopman dynamics, yielding estimates of the true dynamics. This resulted in satisfactory control performance, including accurate tracking of a trajectory. Comparisons with a conventional PI controller show that while performance is similar, the data-driven PIFEK approach requires significantly less system-specific knowledge, underscoring its potential for efficient control design.
