A Study of Machine Learning Techniques for Detecting and Classifying Structural Damage

Abstract

Abstract-We report on work that is part of the development of an agent-based structural health monitoring system. The data used are acoustic emission signals, and we classify these signals according to source mechanisms, those associated with crack growth being particularly significant. The agents are proxies for communication-and computation-intensive techniques and respond to the situation at hand by determining an appropriate constellation of techniques. It is critical that the system have a repertoire of classifiers with different characteristics so that a combination appropriate for the situation at hand can generally be found. We use unsupervised learning for identifying the existence and location of damage but supervised learning for identifying the type and severity of damage. The supervised learning techniques investigated are support vector machines (SVM), naive Bayes classifiers, and feed-forward neural networks (FNN). The unsupervised learning techniques investigated are k-means (with k equal to 3, 4, 5, and 6) and self-organizing maps (SOM, with 3, 4, 5, and 6 output neurons). For each technique except SOM, we tested versions with and without principal component analysis (PCA) to reduce the dimensionality of the data. We found significant differences in the characteristics of these machine learning techniques, with trade-offs between accuracy and fast classification runtime that can be exploited by the agents in finding appropriate combinations of classification techniques. The approach followed here can be generalized for exploring the characteristics of machine-learning techniques for monitoring various kinds of structures.

Análise Comparativa da Aplicação de Técnicas de Clusterização no Monitoramento da Integridade Estrutural

“…Em Nick et al (2015) [10], no âmbito de detecção de dano em aeronaves, foram utilizadas técnicas de aprendizado não supervisionado: k-means e mapas autoorganizáveis. Técnicas de aprendizado supervisionado tabém foram aplicadas: máquinas de vetores de suporte, Naive Bayes e redes neurais.…”

State-Of-Research on Performance Indicators for Bridge Quality Control and Management

“…In addition, modern long-term SHM systems collect huge amounts of data that have to be adequately post-processed (Soyoz and Feng, 2009): the development of fast algorithms and new metrics represents therefore a relevant topic of research for advancing in this field. Among others, genetic algorithms and machine learning techniques (Nick et al, 2015;Liang et al, 2016) can be used to handle huge amount of data deriving from long-term SHMs, also taking into account effects of incomplete measurements (Marano et al, 2011) and operational/environmental variability over time (Figueiredo et al, 2011). Results of SHM are usually too linked to quantitative condition assessment: in this regard, interesting case study applications were presented by Catbas et al (2008), Frangopol et al (2008), , Liu et al (2010), considering the reliability index as quantitative measure to be linked with monitoring data.…”

A Damage Classification Approach for Structural Health Monitoring Using Machine Learning

“…The machine is trained to find the complex patterns and relationships between them and obtain generalized responses based on this training with right answers [35,36]. (ii) Unsupervised, where the machine is trained to find the similarities in the data and provide a clustering organization to indicate its proximity [37,38].…”