Machine-learning-based damage identification methods with features derived from moving principal component analysis

Abstract

This paper aims to propose machine-learning-based damage identification methods with features derived from moving principal component analysis (MPCA) to improve the damage identification performance for engineering structures. Previously, machine learning algorithms have usually used structural responses as inputs directly. These methods show low damage identification capabilities and are susceptible to noise. In this paper, the eigenvectors of structural responses derived from MPCA are employed as inputs instead. Several traditional machine learning algorithms are applied for verification. The results demonstrate that as compared to strains and frequencies, their eigenvectors as inputs for machine learning algorithms render better performances for damage identification.

Keywords:

• Machine learning algorithm
• principal component analysis
• eigenvector
• time window
• damage localization
• damage quantification

Additional information

Funding

This research was funded by the National Natural Science Foundation of China (Grant Nos. 11972162, 11602087, 11932007, 11772134, 11772131, and 11772132), China Scholarship Council (No. 201706155040) and the Science and Technology Program of Guangzhou, China (Grant No. 201903010046).