Mechanics-Informed Autoencoder Enables Automated Detection and Localization of Unforeseen Structural Damage [NatureComms] [arXiv]

@article{li2024mechanics,
  title={Mechanics-Informed Autoencoder Enables Automated Detection and Localization of Unforeseen Structural Damage},
  author={Li, Xuyang and Bolandi, Hamed and Masmoudi, Mahdi and Salem, Talal and Lajnef, Nizar and Boddeti, Vishnu Naresh},
  journal={arXiv preprint arXiv:2402.15492},
  year={2024}
}

Overview

We propose Mechanics-Informed Damage Assessment of Structures (MIDAS), a near-real-time SHM framework for automated damage detection and localization. The key contribution of MIDAS is the seamless integration of inexpensive sensors, data pre-processing in the form of compression, and a customized autoencoder called Mechanics-Informed Autoencoder (MIAE).

Check out the blog post at here.

We train MIAE on a representative dataset of normal service structural operation response and efficiently reduce the raw sensor data stream for network training. The properties of damage are unknown/undefined during training. Once trained, the reconstruction errors between the network's input and output are used to construct a model reference representing the intact structural pattern. Data collected from the damaged structure are tested in the trained model. In the top right section of the figure, the observable shift in reconstruction errors effectively highlights the detection of anomalies indicative of damage. The incorporated mechanics knowledge notably enhances the distribution shift, leading to significant improvements in damage detection rate and the potential for early damage detection. In the bottom right section, we performed sensor-wise error comparisons based on norm value and displayed it in a contour. This process assists in early localization of anomalies representing the onset of damage.

Damage detection

MIAE outperforms standard autoencoder in detecting different damage sizes across all metrics and can detect damages earlier.

MIAE also performs better than other machine learning methods, represented by 3 typical crack sizes.

Damage localization

MIAE can localize much smaller damages, which is represented by achieving a higher damage localization rate as damage propagates.

Number of sensors as a variable

Even with limited sensors, MIAE demonstrates higher detection and localization performance than other baselines.

Run the model

git clone https://github.com/human-analysis/midas-shm
cd midas-shm
python main.py -e crack

There are 3 scenarios conducted: FEM simulations ('simu', 'simu_noise', and 'simu_temp') and three experimental works ('crack', 'bc', 'beam_column') representing the crack and boundary condition variations on gusset plate and damage on a beam column structure. The trained model is provided in saved_models. By executing the code for different scenarios, the article's corresponding results (damage detection and localization figures) are generated in the figs folder. A Source_data file is provided to reproduce the figures in the article?

Requirements

A requirements.txt file has been provided.