Exploration of Damage Identification Method for a Large-Span Timber Lattice Shell Structure in Taiyuan Botanical Garden based on Structural Health Monitoring

Guoqing Wang, Chenjia Xu, Shujia Zhang, Zichun Zhou, Liang Zhang, Bin Qiu, Jia Wan, Honggang Lei2023 · DOI

TL;DR

This paper introduces a novel damage identification method for large-span timber lattice shell structures using time series analysis and Principal Component Analysis (PCA). By analyzing vibration data, the method effectively identifies damage location and severity, offering a robust solution for structural health monitoring of complex structures like the Taiyuan Botanical Garden dome.

What Problem Does This Paper Address?

Large-span spatial lattice structures, like the dome in the Taiyuan Botanical Garden, present significant challenges for structural health monitoring (SHM). These structures often exhibit incomplete modal information, high modal density, and a large number of degrees of freedom, making traditional damage detection methods prone to misjudgment. Furthermore, obtaining accurate input and output signals for modal analysis in practical engineering settings is often difficult. This paper aims to overcome these limitations by proposing a damage identification method that relies on time series analysis of vibration response data, avoiding the need for explicit modal parameter identification or artificial excitation.

Methods

The core of the proposed method involves several key steps:

  • Time Series Modeling (ARMA): The study utilizes the Autoregressive Moving Average (ARMA) model, a powerful tool for analyzing time series data. The ARMA model captures the statistical relationships within vibration response data, allowing for the identification of system characteristics without direct knowledge of input forces.
  • Order Determination: The Akaike Information Criterion (AIC) is employed to select the optimal order (p, q) for the ARMA model, balancing model complexity with its ability to fit the data.
  • Parameter Estimation: The Long Autoregressive (LAR) method is used to estimate the ARMA model parameters. This method transforms nonlinear regression problems into linear ones, simplifying calculations and improving efficiency.
  • Dimensionality Reduction (PCA): To handle the potentially high dimensionality of ARMA model parameters and to filter out noise, Principal Component Analysis (PCA) is applied. PCA extracts the most informative features from the AR coefficients, reducing the data size while retaining crucial information related to structural damage.
  • Damage Sensitive Feature (DSF) Construction: A Damage Sensitive Feature (DSF) is constructed to quantify the structural health.
  • Mahalanobis Distance (MD): The Mahalanobis distance is used to measure the difference between the feature parameters of a structure in an unknown state (potentially damaged) and its healthy state. MD is chosen for its ability to account for correlations between variables and eliminate scale influences.
  • DSF Formula: The DSF is calculated as the ratio of the Mahalanobis distance of the unknown-state sample to the Mahalanobis distance of the healthy-state sample, relative to a training set. A DSF value close to 1 indicates a healthy state, while values greater than 1 suggest damage, with higher values indicating more severe damage.

Datasets

The study utilizes two main sources of data:

  • Finite Element (FE) Model Data: A detailed 3D FE model of the Taiyuan Botanical Garden dome was created using Midas Gen. This model was used to simulate various damage conditions (DCs) by reducing the stiffness of local timber beam and steel cable elements. Five distinct damage scenarios were simulated, involving damage to single or multiple components at different levels (40% and 80%). This simulated data served to validate the proposed damage identification method.
  • Structural Health Monitoring (SHM) Data: Real-world vibration data was collected from nine vibration monitoring points (MPs) installed on the Taiyuan Botanical Garden dome. This data was collected over two years, with initial data (October 2020) serving as the healthy-state baseline and data from two years later (October 2022) representing the unknown-state.

Key Insights and Ideas

  • Time Series for Damage Detection: The paper demonstrates that time series models, specifically ARMA, can effectively capture structural damage signatures without requiring explicit modal analysis or input excitation, which are often difficult to obtain in practice.
  • PCA for Feature Engineering: Applying PCA to the ARMA model parameters is a crucial step that significantly reduces dimensionality and enhances the robustness of the damage identification process by filtering out noise and highlighting the most damage-sensitive features.
  • DSF Sensitivity and Robustness: The constructed DSF, based on Mahalanobis distance, proved to be sensitive to both single and multiple component damages, and showed a positive correlation with the degree of damage. It also exhibited robustness to small damages and environmental noise.
  • Spatial Correlation of Damage: The study observed that damage at a specific location can lead to an increase in the DSF of surrounding monitoring points, with the magnitude of the increase inversely proportional to the distance. This spatial correlation can aid in localizing damage even if monitoring points are not directly above the damaged area.
  • Practical Applicability: The method was successfully validated on both simulated FE data and real SHM data from a large-scale, complex structure, highlighting its practical value for engineering applications.

Key Results

  • FE Model Validation: The proposed method successfully identified simulated damage scenarios in the FE model, accurately locating damage and reflecting its severity. For instance, a 40% timber beam damage at MP M resulted in a DSF of 2.1799, significantly higher than other MPs.
  • SHM Data Analysis: Analysis of SHM data after two years of service indicated that the Taiyuan Botanical Garden dome was in a relatively healthy state, with most DSFs close to 1.0. However, MP S2 showed a slightly higher DSF (1.097), suggesting a potential localized issue that warrants attention.
  • Efficiency of PCA: The study found that the first three AR coefficients, after PCA, were sufficient to achieve high accuracy in damage identification, demonstrating the efficiency of the dimensionality reduction technique.

Limitations and Future Work

  • Number and Distribution of MPs: The authors acknowledge that the optimal number and distribution of measurement points (MPs) require further investigation.
  • FE Model Refinement: Future work could involve more refined correction of the FE model to better match real-world behavior.
  • Long-Term Monitoring: While the study analyzed data over two years, continuous long-term monitoring and analysis would provide deeper insights into the structure's degradation patterns.
  • Environmental Factors: The paper mentions the potential impact of temperature gradients due to glass covering, which could influence sampling data. Further research could explore methods to explicitly account for or mitigate such environmental effects.

Summary updated

BibTeX
@misc{wang2023exploration,
  author = {Guoqing Wang and Chenjia Xu and Shujia Zhang and Zichun Zhou and Liang Zhang and Bin Qiu and Jia Wan and Honggang Lei},
  title = {{Exploration of Damage Identification Method for a Large-Span Timber Lattice Shell Structure in Taiyuan Botanical Garden based on Structural Health Monitoring}},
  year = {2023},
  doi = {10.3390/s23156710},
}
Figure