Optimization Method of Wavelet Neural Network for Suspension Bridge Damage Identification

Deqing Guan, Jie Li, Jun Chen2016 · DOI

Abstract

In this paper, an optimization method of wavelet neural network for structure damage identification is established. The suspension bridge is used as the research object. Firstly, wavelet coefficients modulus maxima are obtained by wavelet transform to determine the location of structural damage. Then, the connection weights and thresholds of the neural network are optimized by the particle swarm optimization, the optimized neural network is constructed to determine the degree of damage of the bridge. The validity of the method is verified by numerical simulation of multi span suspension bridge.

Introduction

Suspension bridges are vital infrastructure components that face constant degradation from environmental factors and heavy traffic. For engineers, the primary challenge in structural health monitoring (SHM) is not just detecting damage, but precisely localizing it and quantifying its severity. Traditional methods, such as standard Back-Propagation (BP) neural networks, often suffer from slow convergence or become trapped in local optima due to random weight initialization. This study proposes a hybrid approach that leverages Wavelet Transforms (WT) for damage localization and Particle Swarm Optimization (PSO)-enhanced Wavelet Neural Networks (WNN) for accurate damage quantification.

Methodology: A Two-Stage Pipeline

The researchers developed a two-stage identification process to address the complexities of bridge degradation:

  1. Damage Localization (Wavelet Transform): Structural damage causes a "singularity"—a sudden change in stiffness. By applying the Wavelet Transform to structural strain data, the system identifies "modulus maxima" at the exact point of discontinuity, effectively pinpointing the damage location.
  2. Damage Quantification (PSO-WNN): Once localized, the severity of the damage is determined using a Wavelet Neural Network. To optimize the network's performance, Particle Swarm Optimization (PSO) is employed to find the global optimal initial weights and thresholds, avoiding the pitfalls of random initialization.
Flowchart of the PSO-WNN algorithm integration.

Numerical Simulation and Dataset

The proposed method was validated using a Finite Element Model (FEM) of a multi-span suspension bridge developed in ANSYS. The bridge model consists of a 120m middle span and 60m side spans, divided into 354 units. Damage was simulated by reducing the stiffness of "Unit 20" by increments ranging from 6% to 24% for training purposes. The model's accuracy was then tested against three specific damage levels: 5%, 7%, and 9%.

Results and Discussion

The performance of the optimized PSO-WNN was compared against a standard WNN model. The results demonstrated that the PSO-optimized version significantly reduced prediction errors. For the test damage levels of 5%, 7%, and 9%, the PSO-WNN achieved relative errors of 2.2%, 3.34%, and 1.47%, respectively. In contrast, the standard WNN produced errors as high as 8.54%.

Performance comparison between network output and actual damage degree.

The high accuracy at the 5% damage level is particularly noteworthy, as early detection is critical for maintaining structural integrity and public safety.

Conclusion

This research demonstrates that combining Wavelet Transforms with PSO-optimized neural networks provides a superior method for bridge damage identification compared to traditional neural network approaches. While the current study relies on numerical simulations, it establishes a strong foundation for future applications in real-world structural health monitoring, where noise and multi-point damage will be key areas of focus.

Summary updated

Abstract

In this paper, an optimization method of wavelet neural network for structure damage identification is established. The suspension bridge is used as the research object. Firstly, wavelet coefficients modulus maxima are obtained by wavelet transform to determine the location of structural damage. Then, the connection weights and thresholds of the neural network are optimized by the particle swarm optimization, the optimized neural network is constructed to determine the degree of damage of the bridge. The validity of the method is verified by numerical simulation of multi span suspension bridge.
BibTeX
@misc{guan2016optimization,
  author = {Deqing Guan and Jie Li and Jun Chen},
  title = {{Optimization Method of Wavelet Neural Network for Suspension Bridge Damage Identification}},
  year = {2016},
  doi = {10.2991/aiie-16.2016.45},
}
Figure