Abstract:Accurate structural response prediction forms a main driver for structural health monitoring and control applications. This often requires the proposed model to adequately capture the underlying dynamics of complex structural systems. In this work, we utilize a learnable Extended Kalman Filter (EKF), named the Neural Extended Kalman Filter (Neural EKF) throughout this paper, for learning the latent evolution dynamics of complex physical systems. The Neural EKF is a generalized version of the conventional EKF, where the modeling of process dynamics and sensory observations can be parameterized by neural networks, therefore learned by end-to-end training. The method is implemented under the variational inference framework with the EKF conducting inference from sensing measurements. Typically, conventional variational inference models are parameterized by neural networks independent of the latent dynamics models. This characteristic makes the inference and reconstruction accuracy weakly based on the dynamics models and renders the associated training inadequate. In this work, we show that the structure imposed by the Neural EKF is beneficial to the learning process. We demonstrate the efficacy of the framework on both simulated and real-world structural monitoring datasets, with the results indicating significant predictive capabilities of the proposed scheme.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to accurately predict the dynamic responses of complex structural systems in structural health monitoring and control applications. This usually requires that the proposed model can fully capture the basic dynamic characteristics of these complex structural systems. To this end, the author introduced a Neural Extended Kalman Filter (Neural EKF) based on neural networks for learning and predicting the underlying evolutionary dynamics of complex physical systems. ### Core Problems of the Paper 1. **Accurately Predicting Structural Responses**: In structural health monitoring and control applications, accurately predicting structural responses is a crucial requirement. Traditional models are often difficult to capture the dynamic characteristics of complex structural systems, resulting in insufficient prediction accuracy. 2. **Learning Underlying Dynamics**: How to use data - driven methods, especially deep - learning techniques, to learn and predict the underlying dynamic models of complex structural systems. 3. **Improving Traditional Methods**: Traditional variational inference models and Kalman filters have limitations when dealing with complex systems, such as relying on known dynamic models or relying too much on inference models, resulting in insufficient prediction capabilities. ### Solutions - **Neural EKF**: By introducing the state - transition model and the observation model parameterized by neural networks, the model can be trained end - to - end, thereby more accurately capturing the underlying dynamics of the system. - **Variational Inference Framework**: Combine the variational inference framework and use the extended Kalman filter for inference to ensure that the model fully utilizes the information of the dynamic model during the inference process. - **Replay Overshooting**: Adopt the Replay Overshooting method to optimize the objective function, ensure the accuracy of the transition model, and thus improve the prediction performance. ### Experimental Verification - **Simulated Data**: The effectiveness of the model was verified on simulated data. - **Actual Cases**: Through actual cases of earthquake monitoring and wind turbine monitoring, the powerful ability of the model in capturing the dynamics of complex physical systems was demonstrated. ### Main Contributions - **Innovative Model**: Proposed Neural EKF, which combines the advantages of deep learning and Kalman filtering, and is suitable for dynamic learning and prediction of complex structural systems. - **Combination of Theory and Practice**: It not only theoretically proves the effectiveness of the model, but also shows its powerful prediction ability in practical applications. - **Real - time Monitoring**: It can detect structural abnormalities and changes in real - time monitoring, and is suitable for structural health monitoring and decision - support. Through these methods and experiments, the paper effectively solves the key problems in the dynamics prediction of complex structural systems, providing new tools and methods for structural health monitoring and control.

Neural Extended Kalman Filters for Learning and Predicting Dynamics of Structural Systems

Structural Dynamic Response Reconstruction Based on Recurrent Neural Network–Aided Kalman Filter

Time Series Estimation Based on Deep Learning for Structural Dynamic Nonlinear Prediction

Time-Continuous Energy-Conservation Neural Network for Structural Dynamics Analysis

Neural filtering for Neural Network-based Models of Dynamic Systems

Real-Time Variational Method for Learning Neural Trajectory and its Dynamics

KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dynamics

DynaNet: Neural Kalman Dynamical Model for Motion Estimation and Prediction

Parameter estimation of structural dynamics with neural operators enabled surrogate modeling

An adaptive generalized extended Kalman filter for real-time identification of structural systems, state and input based on sparse measurement

LEARNEST: LEARNing Enhanced Model-based State ESTimation for Robots using Knowledge-based Neural Ordinary Differential Equations

NeuralSI: Structural Parameter Identification in Nonlinear Dynamical Systems

Extended Kalman Filter for Estimation of Parameters in Nonlinear State-Space Models of Biochemical Networks.

A New Nonlinear State Estimator Using the Fusion of Multiple Extended Kalman Filters

An Adaptive State Filtering Algorithm for Systems With Partially Known Dynamics

Neural modal ordinary differential equations: Integrating physics-based modeling with neural ordinary differential equations for modeling high-dimensional monitored structures

Ensemble Kalman Filters (EnKF) for State Estimation and Prediction of Two-time Scale Nonlinear Systems with Application to Gas Turbine Engines

Novel sparseness-inducing dual Kalman filter and its application to tracking time-varying spatially-sparse structural stiffness changes and inputs

Adaptive Variational Bayesian Extended Kalman Filtering for Nonlinear Systems

A hybrid model-data-driven framework for inverse load identification of interval structures based on physics-informed neural network and improved Kalman filter algorithm

State Space Model-Based Runge–Kutta Gated Recurrent Unit Networks for Structural Response Prediction