Bayesian Graph Neural Networks for Strain-Based Crack Localization

Mylonas, Charilaos; Tsialiamanis, George; Worden, Keith; Chatzi, Eleni

doi:10.1007/978-3-030-76004-5_30

Cited by 4 publications

(7 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, [Mylonas et al, 2022] used a Bayesian GNN to infer the position and shape of an unknown crack via patterns of dynamic strain field measurements at discrete locations. Lastly, [Lino et al, 2021] developed a multiscale GNN that efficiently diffuses information across different scales making it ideal for tackling strongly non local problems such as advection and incompressible fluid dynamics.…”

Section: Brief Review Of Earlier Work On Geometric Deep Learningmentioning

confidence: 99%

“…By using residual blocks the NN itself can choose its depth by skipping the training of a few layers using skip connections. This is currently the most common way to train deep NNs [He et al, 2015;Kim et al, 2016;Zagoruyko and Komodakis, 2017;Lim et al, 2017] and it also applies to GNNs [Sanchez-Gonzalez et al, 2020;Pfaff et al, 2021;Mylonas et al, 2022]. The structure of the MLPs of the Processor is the same as the structure of the encoder MLPs.…”

Section: Modelmentioning

confidence: 99%

“…As already discussed in [3.6], in the first layer of the GNN we choose to encode the edge and node features independently into the latent dimension as suggested by [Sanchez-Gonzalez et al, 2020;Pfaff et al, 2021;Mylonas et al, 2022]. We want to test if this choice results in an improved GNN and thus we perform an experiment between 2 GNNs with the same parameters where one uses a GN block to encode the inputs into the latent dimension and the other encodes them independently.…”

Section: D3 Independent Decodermentioning

confidence: 99%

See 2 more Smart Citations

A Graph-based probabilistic geometric deep learning framework with online physics-based corrections to predict the criticality of defects in porous materials

Krokos¹,

Bordas²,

Kerfriden³

2022

Preprint

View full text Add to dashboard Cite

Stress prediction in porous materials and structures is challenging due to the high computational cost associated with direct numerical simulations. Convolutional Neural Network (CNN) based architectures have recently been proposed as surrogates to approximate and extrapolate the solution of such multiscale simulations. These methodologies are usually limited to 2D problems due to the high computational cost of 3D voxel based CNNs. We propose a novel geometric learning approach based on a Graph Neural Network (GNN) that efficiently deals with three-dimensional problems by performing convolutions over 2D surfaces only. Following our previous developments using pixel-based CNN, we train the GNN to automatically add local fine-scale stress corrections to an inexpensively computed coarse stress prediction in the porous structure of interest. Our method is Bayesian and generates densities of stress fields, from which credible intervals may be extracted. As a second scientific contribution, we propose to improve the extrapolation ability of our network by deploying a strategy of online physics-based corrections. Specifically, we condition the posterior predictions of our probabilistic predictions to satisfy partial equilibrium at the microscale, at the inference stage. This is done using an Ensemble Kalman algorithm, to ensure tractability of the Bayesian conditioning operation. We show that this innovative methodology allows us to alleviate the effect of undesirable biases observed in the outputs of the uncorrected GNN, and improves the accuracy of the predictions in general.

show abstract

Section: Brief Review Of Earlier Work On Geometric Deep Learningmentioning

confidence: 99%

Section: Modelmentioning

confidence: 99%

Section: D3 Independent Decodermentioning

confidence: 99%

See 1 more Smart Citation

A Graph-based probabilistic geometric deep learning framework with online physics-based corrections to predict the criticality of defects in porous materials

Krokos¹,

Bordas²,

Kerfriden³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…There have been some damage localization studies that use strain-based sensing. 28 Such studies, however, have focused on larger cracks. It would be advantageous to explore sensors that perform "global" monitoring and provide insights on early failures, thereby eliminating the constraint of locating the high-stress concentration upfront.…”

Section: Introductionmentioning

confidence: 99%

A random forest‐based fatigue damage classification framework using global loading and displacement information

Lee

Dharmadhikari

Basak

2023

Fatigue Fract Eng Mat Struct

View full text Add to dashboard Cite

The article presents a random forest‐based damage classification methodology built upon the force and displacement data obtained via fatigue testing of Al7075‐T6 specimens. Four features, namely, average displacement (D), material stiffness (ST), energy dissipation rate (EDR), and cumulative energy dissipation (CED), are defined from the collected data. These features are classified into healthy and cracked states based on the information obtained from a high‐resolution confocal microscope and a digital microscope. Random forest‐based classifiers are, thereafter, trained and tested using the individual as well as different combinations of the four features. The results reveal that, when used individually, CED shows the best accuracy of 80% in detecting a fatigue crack. When two or more pairs of features are used, the accuracy increases significantly to beyond 95%. CED is observed to be the most important feature for developing efficient classifiers for fatigue crack detection. The results demonstrate that the random forest classifier is a viable option to accurately detect fatigue damage based on the four features using only global information.

show abstract

“…In seminal works, the adoption of graph convolutional networks, in which the problem of feature extraction and classification is mapped in the graph domain rather than resorting to the standard time/frequency representation, has shown the benefit of learning data patterns in a more flexible and self–adaptive way. For example, in [ 6 , 7 ], graph models were applied for crack detection and localization in the framework of vibration diagnostic, showing outstanding performances. Capsule neural networks have also demonstrated great potential to tackle SHM issues.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning Meets Compressed Sensing in Vibration-Based Monitoring

Zonzini

Carbone

Romano

et al. 2022

Sensors

View full text Add to dashboard Cite

Artificial Intelligence applied to Structural Health Monitoring (SHM) has provided considerable advantages in the accuracy and quality of the estimated structural integrity. Nevertheless, several challenges still need to be tackled in the SHM field, which extended the monitoring process beyond the mere data analytics and structural assessment task. Besides, one of the open problems in the field relates to the communication layer of the sensor networks since the continuous collection of long time series from multiple sensing units rapidly consumes the available memory resources, and requires complicated protocol to avoid network congestion. In this scenario, the present work presents a comprehensive framework for vibration-based diagnostics, in which data compression techniques are firstly introduced as a means to shrink the dimension of the data to be managed through the system. Then, neural network models solving binary classification problems were implemented for the sake of damage detection, also encompassing the influence of environmental factors in the evaluation of the structural status. Moreover, the potential degradation induced by the usage of low cost sensors on the adopted framework was evaluated: Additional analyses were performed in which experimental data were corrupted with the noise characterizing MEMS sensors. The proposed solutions were tested with experimental data from the Z24 bridge use case, proving that the amalgam of data compression, optimized (i.e., low complexity) machine learning architectures and environmental information allows to attain high classification scores, i.e., accuracy and precision greater than 96% and 95%, respectively.

show abstract

Bayesian Graph Neural Networks for Strain-Based Crack Localization

Cited by 4 publications

References 5 publications

A Graph-based probabilistic geometric deep learning framework with online physics-based corrections to predict the criticality of defects in porous materials

A Graph-based probabilistic geometric deep learning framework with online physics-based corrections to predict the criticality of defects in porous materials

A random forest‐based fatigue damage classification framework using global loading and displacement information

Machine Learning Meets Compressed Sensing in Vibration-Based Monitoring

Contact Info

Product

Resources

About