Is it worth changing pattern recognition methods for structural health monitoring?

12th International Conference on Damage Assessment of Structures

Bull, L.A.^1 ; Worden, K.^1 ; Cross, E.J.^1 ; Dervilis, N.^1

Department of Mechanical of Engineering, University of Sheffield, Sheffield, United Kingdom^1

关键词: Active Learning;    Artificial intelligence tools;    Classification algorithm;    Classification technique;    Gaussian Processes;    Pattern recognition algorithms;    Pattern recognition method;    Performance metrics;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/842/1/012006/pdf
DOI  :  10.1088/1742-6596/842/1/012006

来源: IOP

【 摘 要 】

The key element of this work is to demonstrate alternative strategies for using pattern recognition algorithms whilst investigating structural health monitoring. This paper looks to determine if it makes any difference in choosing from a range of established classification techniques: from decision trees and support vector machines, to Gaussian processes. Classification algorithms are tested on adjustable synthetic data to establish performance metrics, then all techniques are applied to real SHM data. To aid the selection of training data, an informative chain of artificial intelligence tools is used to explore an active learning interaction between meaningful clusters of data.