Application of unsupervised support vector machine for condition assessment of concrete structures

Abstract

This paper presents research work that aims at developing a robust method for condition assessment of real-life concrete structures for the detection of small cracks at an early stage of development. A method is presented that utilises an unsupervised one-class support vector machine (SVM).Measured acceleration data from the current state of a structure are used as input parameter. The first singular value of the measured response data is utilized for training and testing of new data sets. Two damage identification approaches are demonstrated, one implementing the SVM for each measurement sensor separately, and another one implementing the SVM for all sensors combined. The use of one-class SVM is well suited for the condition assessment in structural health monitoring since they can detect the advancement of cracks by assigning progressively negative decision values. The presented method is based on unsupervised and non-model-based approaches, and hence there is no need for any representative numerical/finite element model of the structure to be created. To demonstrate the feasibility of the method in the detection and assessment of gradually evolving deterioration, it is tested on a replicate structure of a concrete jack arch which is a main structural component on the Sydney Harbor Bridge – one of Australia’s iconic structures. The test structure is a concrete cantilever beam with an arch section which is located on the eastern side of the bridge underneath the bus lane. A cut is introduced to the structure using a saw and its length is progressively increased in four stages while the depth is kept constant; these four damage cases correspond to less than 0.5% reduction in the first three vibrational modes of the structure which is considered a very small damage. It is demonstrated that the presented method can reliably detect the progression of the crack in the structure and thus can enable the real-time monitoring of infrastructures.