Predominant Research Themes in Using Machine Learning in Structural Health Monitoring

Abstract

Structural health monitoring (SHM) using non-destructive machine learning (ML) based technologies has gained considerable interest in research and industrial communities. Integrating the conventional methods of SHM with novel ML techniques gives robust, sustainable, and promising solutions to SHM. This study presents text mining-based methodology to identify predominant research themes in using ML in SHM. Two analyses are performed on literature data of 375 research studies; (1) co-occurrence analysis of keywords applied on author specified keywords and (2) topic modeling using latent dirichlet allocation (LDA) approach applied on abstracts. The finding shows that the research studies predominantly focus on detecting and classifying structural damages, investigating sensing systems or sensors, and feature extraction and analysis. Moreover, convolutional neural networks and support vector machines are the two mainly used ML algorithms, and bridges, dams, and wind turbines are found as the top three investigated engineering structures. This work can be further extended to include a systematic review of past studies to have an in-depth understanding of using ML in SHM and to find potential contributions and research gaps in the studied area.