17 results
Now showing1 - 10 of 17
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Data-driven, non-linear ship response prediction based on time series of irregular, long-crested sea states amidships(Elsevier, 2025-02-01); ; ; ; The accurate prediction of vessel responses in waves is crucial for decision-making and contribute to the operational safety and risk minimization. Short-term predictions can be carried out by estimating the vessel's motions and loads based on incident waves. Existing model-based approaches either require computationally intensive simulations that compromise real-time capability or use simplified models affecting the accuracy of the prediction. Therefore, this study explores the feasibility of using neural networks for mapping time signals of surface elevation data and a set of corresponding ship responses, i.e. the heave and pitch motions as well as the vertical bending moment. The approach followed here is built on the assumption that the wave profile amidships is known. A synthetic dataset was generated using a time-domain strip theory solver with considerations of non-linear effects on motions and loads due to large amplitude waves in a variety of irregular, long-crested sea state conditions. We propose two different neural network models, a multi-layer perceptron (MLP) and a fully convolutional neural network (FCNN), and compare their performances on measurement data obtained from model tests in a seakeeping basin. The evaluations also include the freak wave reproduction of the ‘new year wave’. The proposed networks are able to estimate the motions and bending moment accurately for a wide range of sea state conditions, surpassing current state-of-the-art models on the given data sets.Publicationtype: Journal Article78 97 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Data assimilation and parameter identification for water waves using the nonlinear Schrödinger equation and physics-informed neural networks(Multidisciplinary Digital Publishing Institute (MDPI), 2024-10-01); ; ; The measurement of deep water gravity wave elevations using in situ devices, such as wave gauges, typically yields spatially sparse data due to the deployment of a limited number of costly devices. This sparsity complicates the reconstruction of the spatio-temporal extent of surface elevation and presents an ill-posed data assimilation problem, which is challenging to solve with conventional numerical techniques. To address this issue, we propose the application of a physics-informed neural network (PINN) to reconstruct physically consistent wave fields between two elevation time series measured at distinct locations within a numerical wave tank. Our method ensures this physical consistency by integrating residuals of the hydrodynamic nonlinear Schrödinger equation (NLSE) into the PINN’s loss function. We first showcase a data assimilation task by employing constant NLSE coefficients predetermined from spectral wave properties. However, due to the relatively short duration of these measurements and their possible deviation from the narrow-band assumptions inherent in the NLSE, using constant coefficients occasionally leads to poor reconstructions. To enhance this reconstruction quality, we introduce the base variables of frequency and wavenumber, from which the NLSE coefficients are determined, as additional neural network parameters that are fine tuned during PINN training. Overall, the results demonstrate the potential for real-world applications of the PINN method and represent a step toward improving the initialization of deterministic wave prediction methods.Publicationtype: Journal Article31 38 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Data-driven generation of tailored wave sequences(American Society of Mechanical Engineers (ASME), 2024-06); ; ; ; This paper explores the applicability of machine learning techniques for the generation of tailored wave sequences. For this purpose, a fully convolutional neural network was implemented for relating the target wave sequence at the target location in time domain to the respective control signal of the wave board. The database was generated by means of extensive wave tank tests. The experimental campaign focused on the generation of very steep wave groups including wave breaking which cannot be covered by the simplified wave generation methods. The extensive experimental campaign was performed in a small wave tank with an fully automated approach including determination and control of the wave maker motion as well as data measurement. The training data set features wave groups of short duration based on JONSWAP spectra, where the parameters wave steepness, peak wave period and enhancement factor were systematically varied. At the end of the training process, the trained models are able to predict the wave maker control signal based on time series of the target wave defined for a specific target location in the wave tank. The accuracy of the trained models are evaluated by means of unseen validation data. In addition, the predictive accuracy of the trained models is compared with the classical linear transformation approach.Publicationtype: Conference Paper56 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Machine learning for phase-resolved reconstruction of nonlinear ocean wave surface elevations from sparse remote sensing data(Elsevier, 2023-11-15); ; ; ; Accurate short-term predictions of phase-resolved water wave conditions are crucial for decision-making in ocean engineering. However, the initialization of remote-sensing-based wave prediction models first requires a reconstruction of wave surfaces from sparse measurements like radar. Existing reconstruction methods either rely on computationally intensive optimization procedures or simplistic modelling assumptions that compromise the real-time capability or accuracy of the subsequent prediction process. We therefore address these issues by proposing a novel approach for phase-resolved wave surface reconstruction using neural networks based on the U-Net and Fourier neural operator (FNO) architectures. Our approach utilizes synthetic yet highly realistic training data on uniform one-dimensional grids, that is generated by the high-order spectral method for wave simulation and a geometric radar modelling approach. The investigation reveals that both models deliver accurate wave reconstruction results and show good generalization for different sea states when trained with spatio-temporal radar data containing multiple historic radar snapshots in each input. Notably, the FNO demonstrates superior performance in handling the data structure imposed by wave physics due to its global approach to learn the mapping between input and output in Fourier space.Publicationtype: Journal Article28 95 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Up-Net : a generic deep learning-based time stepper for parameterized spatio-temporal dynamics(Springer, 2023-03-24); ; ; ; In the age of big data availability, data-driven techniques have been proposed recently to compute the time evolution of spatio-temporal dynamics. Depending on the required a priori knowledge about the underlying processes, a spectrum of black-box end-to-end learning approaches, physics-informed neural networks, and data-informed discrepancy modeling approaches can be identified. In this work, we propose a purely data-driven approach that uses fully convolutional neural networks to learn spatio-temporal dynamics directly from parameterized datasets of linear spatio-temporal processes. The parameterization allows for data fusion of field quantities, domain shapes, and boundary conditions in the proposed Up-Net architecture. Multi-domain Up-Net models, therefore, can generalize to different scenes, initial conditions, domain shapes, and domain sizes without requiring re-training or physical priors. Numerical experiments conducted on a universal and two-dimensional wave equation and the transient heat equation for validation purposes show that the proposed Up-Net outperforms classical U-Net and conventional encoder–decoder architectures of the same complexity. Owing to the scene parameterization, the Up-Net models learn to predict refraction and reflections arising from domain inhomogeneities and boundaries. Generalization properties of the model outside the physical training parameter distributions and for unseen domain shapes are analyzed. The deep learning flow map models are employed for long-term predictions in a recursive time-stepping scheme, indicating the potential for data-driven forecasting tasks. This work is accompanied by an open-sourced code.Publicationtype: Journal ArticleScopus© Citations 4 55 152 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Towards neural network‐based numerical friction modelsFriction contacts can be found in almost all mechanical systems and are often of great technical importance. However, they are usually difficult to describe, and their behavior and influence on the whole system are hard to predict accurately. Modern product design and system operation strongly benefit from numerical simulation approaches today, but reliable friction models still represent a major challenge in this context. To tackle this problem, we employ neural network regression to capture the characteristics of frictional contacts and make them accessible for numerical methods in a minimal intrusive fashion. In particular, we test our approach using a Finite Element model of a 2D cantilever beam subject to stick-slip vibrations induced by a moving conveyor belt at its free end. As a reference solution, we perform a transient analysis based on a simple analytical friction model, where the kinetic friction force only depends on the normal load and the relative sliding velocity. We take the same friction model, add some artificial noise to mimic uncertainties coming with experimental measurements, and pick a limited set of data points to train a regression neural network. The machine learning friction model is then deployed in the Finite Element code to predict the kinetic friction force acting on the beam tip during the slip phases. The deflection curves obtained by the transient numerical analysis using the new neural network friction model agree well with the reference solution based on the underlying analytical model. The results indicate that data-driven approaches may also be capable of capturing more complex frictional contacts, including effects of temperature, humidity, and load history. The trained neural network friction models can then be employed in numerical simulations in a minimally intrusive manner. This approach opens up new possibilities to predict individual mechanical system behavior as accurately as possible.Publicationtype: Conference Paper43 240 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Machine learning-based state maps for complex dynamical systems: applications to friction-excited brake system vibrationsIn this work, a purely data-driven approach to mapping out the state of a dynamical system over a set of chosen parameters is presented and demonstrated along a case study using real-world experimental data from a friction brake system. Complex engineering systems often exhibit a rich bifurcation behavior with respect to one or several parameters, which is difficult to grasp using experimental approaches or numerical simulations. At the same time, the growing need for energy-efficient machines that can operate under varying or extreme environmental conditions also calls for a better understanding of these systems to avoid critical transitions. The proposed method combines machine learning techniques with synthetic data augmentation to create a complete state map for a dynamical system. First, a machine learning model is trained on experimental data, picking up hidden mechanisms and complex parametric relations of the underlying dynamical system. The model is then exploited to assess the state of the system for a set of synthetically generated data to obtain a state map over the complete space spanned by the chosen parameters. In addition, an extension of the concept to a probability state map is introduced. The results indicate that the proposed method can uncover hidden variables which drive dynamical transitions between different states of a system that were previously inaccessible.Publicationtype: Journal ArticleScopus© Citations 2 36 77 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Surface similarity parameter : a new machine learning loss metric for oscillatory spatio-temporal dataSupervised machine learning approaches require the formulation of a loss functional to be minimized in the training phase. Sequential data are ubiquitous across many fields of research, and are often treated with Euclidean distance-based loss functions that were designed for tabular data. For smooth oscillatory data, those conventional approaches lack the ability to penalize amplitude, frequency and phase prediction errors at the same time, and tend to be biased towards amplitude errors. We introduce the surface similarity parameter (SSP) as a novel loss function that is especially useful for training machine learning models on smooth oscillatory sequences. Our extensive experiments on chaotic spatio-temporal dynamical systems indicate that the SSP is beneficial for shaping gradients, thereby accelerating the training process, reducing the final prediction error, increasing weight initialization robustness, and implementing a stronger regularization effect compared to using classical loss functions. The results indicate the potential of the novel loss metric particularly for highly complex and chaotic data, such as data stemming from the nonlinear two-dimensional Kuramoto–Sivashinsky equation and the linear propagation of dispersive surface gravity waves in fluids.Publicationtype: Journal ArticleScopus© Citations 5 52 57 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Towards neural network-based numerical friction models(2022-08-18); ; Friction contacts can be found in almost all mechanical systems and are often of great technical importance. However, they are usually difficult to describe, and their behavior and influence on the whole system are hard to predict accurately. Modern product design and system operation strongly benefit from numerical simulation approaches today, but reliable friction models still represent a major challenge in this context. To tackle this problem, we employ neural network regression to capture the characteristics of frictional contacts and make them accessible for numerical methods in a minimal intrusive fashion. In particular, we test our approach using a Finite Element model of a 2D cantilever beam subject to stick-slip vibrations induced by a moving conveyor belt at its free end. As a reference solution, we perform a transient analysis based on a simple analytical friction model, where the kinetic friction force only depends on the normal load and the relative sliding velocity. We take the same friction model, add some artificial noise to mimic uncertainties coming with experimental measurements, and pick a limited set of data points to train a regression neural network. The machine learning friction model is then deployed in the Finite Element code to predict the kinetic friction force acting on the beam tip during the slip phases. The deflection curves obtained by the transient numerical analysis using the new neural network friction model agree well with the reference solution based on the underlying analytical model. The results indicate that data-driven approaches may also be capable of capturing more complex frictional contacts, including effects of temperature, humidity, and load history. The trained neural network friction models can then be employed in numerical simulations in a minimally intrusive manner. This approach opens up new possibilities to predict individual mechanical system behavior as accurately as possible.Publicationtype: Conference Presentation122 192 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Data-driven stability maps for friction induced vibrationsFriction-induced vibrations, apparent for example as squeal noise in car braking systems, have seen a significant amount of research effort in the past and remain an area of active research today. Due to the huge bifurcation parameter space and model parameter uncertainties, the instability mechanisms of these vibrations remain elusive. Particularly, bifurcation parameters and critical sensitivities vary from brake system to brake system. Numerical simulations cannot resolve all aspects for robust instability predictions today. Recent work has illustrated how data-driven approaches to friction-induced noise modeling can be an alternative approach to conventional physics-based modeling: A digital twin representing a specific brake system was shown to predict instabilities and mode coupling associated with squeal occurrence with high accuracy. This work builds upon previous data-driven approaches to obtain a digital twin from measurement data of a real-world disc brake system. After validation, this data-based model is exploited to study the effect of various loading conditions on the predicted squealing behavior of the system. By gradually altering the loading conditions, we generate stability maps that encode information about the stability regions of the respective brake system. These stability maps provide a data-driven perspective on the mechanisms underlying frictioninduced noise, fostering a deeper understanding of the bifurcation parameter space of the underlying system.Publicationtype: Conference Presentation147 195