Data-driven structural damage detection approach for Digital twin-Structural health monitoring

As part of the project activity funded by the Newton Fund Institutional Links through the U.K. Department of Business, Energy, and Industrial Strategy and managed by the British Council under Grant 429715093, Research paper ” Data-Driven Structural Health Monitoring using feature Fusion and Hybrid Deep Learning ” was accepted for publication in Q1 journal “IEEE Transactions on Automation Science and Engineering” in October 2020.

Toward a Digital Twin model automatically monitoring the operational state of large-scale structures in a near-real-time fashion, it requires a Structural Damage Detection (SDD) approach, both time-efficient, and data-efficient, while still maintaining accurate detection results. Conventional methods either resort to time-consuming sophisticated preprocessing techniques such as experimental modal analysis or data-eager image-based deep learning duplicating multiple times raw measured data. Therefore, we engineer an efficient yet reliable data-driven approach only require simple and practical signal preprocessing operations in coupled with a hybrid deep learning architecture CNN-LSTM using 1D time-series to perform SDD task. The applicability and effectiveness of the proposed approach are supported by three case studies with increasing complexities from experimental laboratory data to 3D synthetically numerical data and to real data of bridge Z24 in Switzerland. Obtained results show that the proposed approach achieves comparable performance with the 2DCNN counterpart while having time complexity reduced by more than 50% and no supplement storage required for images. From the proposed method, various studies are conducted to provide insights into the effect of different parameters on structural damage detection performance: i) the method maintains a good performance when there is data contamination of up to 10% random noise; ii) a reduction in length (10%) of input time-series can lead to a significant decrease in detection accuracy (20%), and iii) increasing the number of sensors improves the damage detection accuracy effectively.

Feature-fusion hybrid deep learning approach for structural damage detection
Case study 1: Laboratory Data from Los Alamos National Laboratory, USA. 
Case Study 2: My Thuan Stayed-Bridge, Vietnam
Case study 3: Progressive Damage Tests of Z24 Bridge, Switzerland.
Parametric study on the impact of the number of sensors on the detection accuracy

Leave a Reply