Skip to content
Log in

Similar papers

Open access Aug 2026

Physics-Informed Neural Network Framework for Input Load Estimation and Virtual Sensing of Offshore Wind Turbines

Recent advances in Physics-Informed Neural Networks (PINNs) have opened new possibilities for integrating structural dynamics and data-driven learning in Structural Health Monitoring (SHM). This work presents a physics-informed framework for input load estimation and virtual sensing of offshore wind turbine support structures, where the governing dynamics of the system are embedded directly into the learning process. Unlike purely data-driven models that require extensive labeled datasets, the proposed approach leverages known physical relationships among displacement, velocity, acceleration, and external loads to enhance interpretability and generalization. The method adopts an encoder–decoder neural architecture that maps measured accelerations and strains to a reduced-order modal space before decoding the corresponding dynamic responses and reconstructing the applied loads through embedded structural dynamics relationships. Physical consistency is enforced through the equations of motion and differential constraints between displacement, velocity, and acceleration, while automatic differentiation ensures temporal consistency without requiring explicit load data during training. This hybrid approach captures the temporal and spatial evolution of loads even with limited or noisy measurements. The framework is first validated on numerical simulations of an offshore wind turbine, accurately recovering unmeasured input loads and structural responses across diverse operating conditions. It is then demonstrated using experimental vibration data, confirming its robustness to sensor noise and sparse instrumentation. Results show that the proposed physics-informed strategy can recover complex loading patterns and provide virtual measurements that are otherwise inaccessible in practice. Overall, study advances the use of PINNs for inverse input load estimation problem in SHM, offering a computationally efficient and generalizable tool for condition monitoring and fatigue assessment of large-scale energy infrastructure.

Azin Mehrjoo, E. Tronci, Babak Moaveni · 0 citations
Open access Jun 2026

A Physics-Informed Deep Learning Method for Wind Turbine Impedance Modeling

Accurate impedance modeling of wind turbines (WTs) is essential for assessing the small-signal stability of power systems with high penetration of renewable energy. Existing approaches face a fundamental trade-off: physics-based “white-box” models require proprietary manufacturer parameters that are rarely disclosed, while purely data-driven “black-box” models often lack physical interpretability and exhibit poor generalization under unseen operating conditions. To address this gap, this paper proposes a gray-box framework—the Physics-Informed Hybrid Model (PIHM)—that integrates a simplified physical impedance branch with a Bidirectional Long Short-Term Memory (Bi-LSTM) network in a novel parallel architecture. The physical branch, systematically parameterized via a constrained phase-error minimization method, captures the dominant baseline dynamics and decouples the learning task, allowing the Bi-LSTM to focus exclusively on the complex nonlinear residual. The framework is validated on a high-fidelity simulation platform of a doubly fed induction generator (DFIG) wind farm. Quantitative results demonstrate that the PIHM achieves an average coefficient of determination (R2) of 0.989 and a mean squared error (MSE) of 1.24×10−4 on unseen test data, while producing smooth, physically consistent impedance profiles that generalize across four distinct wind speed conditions. These results establish the PIHM as a reliable, parameter-free tool for impedance-based stability analysis of modern wind power systems.

Libin Wen, Jinji Xi, Tannan Xiao et al. · 1 citation
#reinforcement learning Open access Oct 2026

Seismic Control of a Smart Base-Isolated Building with Nonlinear Behavior Using Deep Reinforcement Learning

DRL is highlighted as a promising data-driven strategy for robust and adaptive control of nonlinear structural systems under partial observability by addressing a critical limitation of passive systems and accelerates the decay of residual vibrations.

Takehiko Asai · 0 citations
Open access Aug 2026

A Unified Physics-Constrained Deep Reinforcement Learning Framework for Parameter Identification of Nonlinear Hysteretic Models

Reliable nonlinear structural analysis requires hysteretic parameters that reproduce cyclic stiffness, strength, pinching, degradation, and energy dissipation. Conventional calibration is often tailored to one constitutive model and unit system, while repeated population searches become costly as dimensionality and parameter coupling increase. This study develops a unified physics-constrained deep reinforcement learning framework for OpenSees Steel02 and DowelType identification. Target responses and candidate parameters are expressed in dimensionless coordinates; bounded latent variables are decoded into admissible model parameters and mapped back to source units after calibration. A twin-delayed deep deterministic policy gradient (TD3) agent performs continuous search, with differential evolution providing local refinement when required. Validation used synthetic targets, random initial vectors, public steel records, and ten experimental hysteresis records from the authors’ research group; particle swarm optimization and a genetic algorithm served as benchmarks. The framework satisfied an NRMSE threshold of 0.02 in all 384 held-out Steel02 evaluations and achieved a mean NRMSE of 0.0166 on synthetic DowelType targets. On the ten experimental DowelType records, TD3+DE reached a mean NRMSE of 0.0654 and 30% success under the relaxed 0.05 threshold, giving accuracy comparable with tuned PSO at the same online OpenSees-call budget while retaining a reusable learned initialization step. One normalized workflow can rapidly obtain response-equivalent fits for distinct hysteretic laws and return solver-ready parameters in physical units.

Hanlin Dong, Chunhua Liu, Mingji Fang et al. · 0 citations
Preprint Jul 2026

SpectONet: A Physics-Guided Spectral Deep Operator Network for Euler-Bernoulli Beam Dynamics

This paper proposes a novel physics-guided spectral deep operator network, termed SpectONet, for solving Euler-Bernoulli beam (EBB) vibration problems. The proposed framework integrates the operator-learning capability of DeepONet with physics-informed constraints and Chebyshev-Gauss-Lobatto (CGL) sensor placement. Unlike conventional DeepONet frameworks, which commonly employ uniformly distributed sensors, SpectONet uses nonuniform spectral sensor locations with a higher concentration of points near the domain boundaries. This sampling strategy improves the finite-dimensional representation of boundary-sensitive structural responses while requiring only a limited number of branch-network inputs. The governing beam equation, together with the associated initial and boundary conditions, incorporated into the training objective to promote physically consistent and generalizable predictions. Numerical experiments on three synthetic EBB vibration problems and a real-world bridge vibration dataset demonstrate the effectiveness of the proposed framework. Comparisons with strong baselines such as, Vanilla DeepONet, PI-DeepONet, PINN, and CNN-UNet show that SpectONet consistently achieves lower prediction errors across all considered evaluation metrics. In particular, SpectONet achieves at least \(64\%\) improvement over the considered baseline models across the three synthetic problems and at least \(37\%\) for the real-world problems. These results demonstrate that SpectONet provides an accurate, computationally efficient, and physically consistent operator-learning framework for structural vibration analysis.

Shivani Saini, R. Vats, A. Sahoo · 0 citations
Aug 2026

Efficient pitching moment prediction for canard-controlled missiles via transfer learning-based deep learning

Aerodynamic design of aerospace vehicles often necessitates extensive Computational Fluid Dynamics (CFD) simulations, which are computationally expensive. To address this, we propose an efficient multilayer neural network framework to predict the pitching moment coefficient of canard-controlled missiles. A low-fidelity database of over 28,000 points was rapidly generated using Missile DATCOM and used to train an initial neural network with four hidden layers. The core of our methodology is an architectural transfer learning approach, where this pre-trained model initializes a high-fidelity network, significantly reducing the need for costly CFD data (using only 120 samples). The Levenberg-Marquardt algorithm’s hyperparameters were fine-tuned to optimize performance. The final model achieved a Root Mean Square Error (RMSE) of 0.0055 on a random test dataset. The model’s stability and generalization capability were further confirmed through 5-fold cross-validation, which demonstrated robust performance. This highly accurate and validated model provides a powerful surrogate tool for rapid design optimization based on trim angle of attack and geometric parameters.

M. Shojaeefard, Masoud Nobakhti · 0 citations