Damage Classification Using Labelled and Unlabelled Measurements

Abstract

The use of a Gaussian mixture model for probabilistic and semi-supervised learning is introduced here for damage-classification. Semi-supervised learning allows for pattern recognition algorithms to learn from the available labelled and unlabelled data; this is particularly relevant to SHM, as labelling measurements from engineering systems is often impracticable and expensive. In this work, semi-supervised learning is shown to improve the classification performance for simulated SHM data by utilising the available unlabelled measurements alongside a set of labelled data.