Physics-Informed Neural Network–Based TMD Parameter Identification and Response Prediction

Document Type

Article

Publication Date

1-1-2025

Abstract

Tuned mass dampers (TMDs) are crucial for mitigating excessive structural vibrations. Accurate acquisition of TMD parameters and responses from limited data is vital for assessing TMD performance and structural safety. Conventional physics-based methods require ideal environmental conditions, while pure data-driven approaches face limitations in generalization and interpretability. To address these issues, this study proposes a physics-informed neural network (PINN) that synergizes physical principles with data-driven techniques for TMD parameter identification and response prediction. The governing equations of TMD motion are embedded into a multilayer perceptron (MLP) architecture as physical constraints. Task-specific loss functions are designed for distinct tasks, and a tailored adaptive moment estimation (Adam) optimizer is utilized. To examine the performance of the proposed PINN-based method, it is applied to a single-degree-of-freedom (SDOF) system with a TMD. The results show that the proposed method can accurately identify the TMD parameters and predict the TMD responses. A comprehensive analysis is further conducted to evaluate the influence of key factors including observation noise, the number of training data points, sampling frequency, model hyperparameters, and physical equation errors. Additionally, the PINN-based method is compared with the data-driven method to validate the effectiveness of the proposed method.

Publication Source (Journal or Book title)

Structural Control and Health Monitoring

This document is currently not available here.

Share

COinS