Structural dynamic response synthesis: A transformer-based time-series GAN approach

This study introduces a novel Transformer-based Time-Series Generative Adversarial Network (TTS-GAN) for synthesizing structural dynamic responses, addressing the critical challenge of data scarcity in structural health monitoring (SHM). Unlike existing methods, TTS-GAN generates realistic multi-cha...

Full description

Saved in:
Bibliographic Details
Main Authors: Sayed Ali Khodaei, Maryam Bitaraf
Format: Article
Language:English
Published: Elsevier 2025-06-01
Series:Results in Engineering
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2590123025016196
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study introduces a novel Transformer-based Time-Series Generative Adversarial Network (TTS-GAN) for synthesizing structural dynamic responses, addressing the critical challenge of data scarcity in structural health monitoring (SHM). Unlike existing methods, TTS-GAN generates realistic multi-channel acceleration signals directly from random noise vectors, without requiring auxiliary time- or frequency-domain inputs. The model integrates a transformer architecture with an attention mechanism to capture complex temporal dependencies and employs a composite loss function that aligns generated outputs with real signals in both time and frequency domains. Validation on a three-story aluminum structure demonstrates that TTS-GAN accurately replicates key structural features, including amplitude distribution and natural frequencies. Comparative results confirm that TTS-GAN outperforms a baseline Time-Series GAN (TSGAN), particularly in frequency-domain fidelity. The proposed approach represents a novel and efficient data augmentation framework for SHM, enabling high-quality signal generation from limited measured data.
ISSN:2590-1230