Generative Adversarial Networks for Data Generation in Structural Health Monitoring

Furkan Luleci,F Necati Catbas,Onur Avci

doi:10.3389/fbuil.2022.816644

Abstract

Structural Health Monitoring (SHM) has been continuously benefiting from the advancements in the field of data science. Various types of Artificial Intelligence (AI) methods have been utilized to assess and evaluate civil structures. In AI, Machine Learning (ML) and Deep Learning (DL) algorithms require plenty of datasets to train; particularly, the more data DL models are trained with, the better output it yields. Yet, in SHM applications, collecting data from civil structures through sensors is expensive and obtaining useful data (damage associated data) is challenging. In this paper, one-dimensional (1-D) Wasserstein loss Deep Convolutional Generative Adversarial Networks using Gradient Penalty (1-D WDCGAN-GP) is utilized to generate damage-associated vibration datasets that are similar to the input. For the purpose of vibration-based damage diagnostics, a 1-D Deep Convolutional Neural Network (1-D DCNN) is built, trained, and tested on both real and generated datasets. The classification results from the 1-D DCNN on both datasets resulted in being very similar to each other. The presented work in this paper shows that, for the cases of insufficient data in DL or ML-based damage diagnostics, 1-D WDCGAN-GP can successfully generate data for the model to be trained on.

Highlights

During the operational life of civil structures, different types of damages can shorten the remaining useful life of the structures
Evaluation of the Generative Adversarial Networks (GANs) models can be categorized as qualitative and quantitative evaluation, where the former is based on visual evaluation, and the latter is based on numerical evaluation
This qualitative approach might not be an easy or efficient way for 1-D data as it suffers from some limitations, such as a limited number of generated output can be viewed by an observer in limited time or observed subjectively by different observers

Summary

Introduction

During the operational life of civil structures, different types of damages can shorten the remaining useful life of the structures. Arjovsky et al (2017) introduced a GAN that uses Wasserstein distance as a loss function (WGAN), which improves the training of GAN. In their network, instead of using a discriminator that estimates the probability of the generated images as being real or fake, they used a critic which scores the output’s realness or fakeness of a given image. Gulrajani et al (2017) proposed using a penalization of the gradient during the training of the critic due to using weight clipping on the critic, which enforces the Lipschitz constraint and lowers the learning capacity of the model They named the model Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP). The authors showed that the proposed method performed better than WGAN and provided more stable training

Objectives

Methods

Results

Conclusion