Structural damage detection using finite element model updating with evolutionary algorithms: a survey

Nizar Faisal Alkayem, Maosen Cao, Yufeng Zhang, Mahmoud Bayat, Zhongqing SuNeural Computing and Applications 2018 · DOI

TL;DR

This paper provides a comprehensive survey of structural damage detection techniques that leverage finite element (FE) model updating with evolutionary algorithms (EAs). It highlights the growing importance of EAs in overcoming limitations of traditional methods for identifying structural damage, offering a roadmap for future research in this critical area.

What Problem Does This Paper Address?

Structural damage, whether from design flaws, construction imperfections, or external factors like earthquakes, can significantly degrade a structure's performance and safety. Early detection is crucial for timely maintenance and ensuring structural reliability. While finite element (FE) models are widely used to simulate structural behavior, accurately updating these models to reflect real-world damage is a significant challenge. Traditional methods for FE model updating, such as sensitivity-based or statistical approaches, often struggle with computational expense, sensitivity to noise, and the inability to handle complex, nonlinear damage scenarios. This paper addresses the need for a comprehensive overview of how modern evolutionary algorithms (EAs) can be effectively employed to overcome these limitations in structural damage detection.

Methods

The paper categorizes FE model updating methods into direct and iterative/indirect approaches. It then delves into various techniques within these categories, including:

  • Direct Methods: Matrix-update, optimal matrix, error matrix, eigenstructure assignment. These are generally accurate but can be sensitive to noise and require precise measurements.
  • Iterative and Indirect Methods:
    • Sensitivity-based methods: Use measured responses to infer damage but are best suited for minor damage.
  • Statistical and Probability-based methods: Employ techniques like Bayesian frameworks and Monte Carlo simulations.
  • Response-Surface Method (RSM): A statistical approach for correlation.
  • Bayesian-Monte Carlo Method: Leverages Bayes' theorem for probabilistic analysis.
  • Computational Intelligence Techniques: This is where EAs shine. The paper focuses on how EAs, such as Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), are used. EAs are highlighted for their ability to handle complex, nonlinear, and multimodal optimization problems without requiring gradient calculations, making them ideal for damage detection where damage patterns can be distributed and local optima are common.
  • Single-objective EAs: These convert the multi-objective problem into a single one, often using a weighted sum of residuals.
  • Multi-objective EAs: These directly tackle multiple objective functions (residuals) simultaneously, aiming to find a set of optimal trade-off solutions (Pareto front).

The paper also details various "residuals" derived from dynamic characteristics (natural frequencies, mode shapes, frequency response functions, modal flexibility, modal strain energy) that are used to formulate the objective functions for damage detection.

Datasets

The paper does not explicitly mention specific datasets used for training or evaluation in a general sense. Instead, it refers to various studies that have applied these methods to different types of structures (beams, bridges, frames, trusses, 3D structures, etc.) and often used simulated data or experimental data from specific benchmark structures (e.g., the ASC-ASCE SHM benchmark 4-story building).

Key Insights and Ideas

  • EAs are well-suited for complex damage scenarios: Unlike traditional methods that struggle with nonlinearity and local optima, EAs excel at finding robust solutions even when damage is widespread or subtle.
  • The choice of residuals matters: The effectiveness of damage detection heavily relies on selecting appropriate dynamic characteristics (like modal strain energy) that are sensitive to damage.
  • Multi-objective EAs offer a more complete picture: While single-objective EAs can be effective, multi-objective EAs provide a set of Pareto optimal solutions, offering a better understanding of trade-offs and potentially more reliable damage identification.
  • Parameterization is crucial: How structural damage is represented (e.g., by adjusting stiffness, mass, or material properties) significantly impacts the updating process.
  • Noise and incomplete data are persistent challenges: Real-world structural health monitoring often involves noisy measurements and incomplete data, which can hinder accurate damage detection.

Key Results

The paper presents a case study comparing GA, PSO, and MOPSO on a 4-story building model with simulated damage and added noise. The results indicate:

  • All tested algorithms (GA, PSO, MOPSO) could detect damage even under noisy conditions.
  • PSO generally outperformed GA in terms of accuracy.
  • MOPSO achieved comparable accuracy to PSO while offering better consistency and reliability, demonstrating the advantages of multi-objective approaches.
  • The study highlights that MOPSO achieved better performance than GA and similar performance to PSO in terms of consistency and reliability.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • Focus on small/medium structures: Most studies have focused on simpler structures. More research is needed for large-scale, complex structures.
  • Combining diverse dynamic characteristics: Exploring novel combinations of dynamic characteristics for objective functions could improve damage detection.
  • Exploring underutilized EAs: Many powerful EAs exist but haven't been applied to this problem. Investigating their suitability and efficiency is recommended.
  • Comparative studies: More comprehensive comparative studies of various EAs are needed to identify the most efficient and reliable algorithms.
  • Real-world validation: Many studies rely on simulations. More experimental validation and studies on the impact of noisy and incomplete data in real-world scenarios are crucial.
  • Multi-attribute decision-making: While essential for multi-objective EAs, its application in FE model updating is still limited and warrants further research.

Summary updated

BibTeX
@article{alkayem2018structural,
  author = {Nizar Faisal Alkayem and Maosen Cao and Yufeng Zhang and Mahmoud Bayat and Zhongqing Su},
  title = {{Structural damage detection using finite element model updating with evolutionary algorithms: a survey}},
  journal = {Neural Computing and Applications},
  year = {2018},
  doi = {10.1007/s00521-017-3284-1},
}
Figure