Cluster analysis based on image feature extraction for automated OMA

Abstract

This study introduces a new method for automated operational modal analysis (OMA) that uses image feature extraction on stabilisation diagrams to cluster data in parametric models. The implementation of automated OMA, a modal analysis that does not require as much human engagement as traditional methods, is a difficult challenge. Without requiring user input, the stabilisation diagram and clustering tools separate real poles from spurious (noise) poles. However, the maximum within-cluster distance between representations of the same physical mode from different system orders is required by existing clustering algorithms, and additional adaptive approaches must be used to optimise the selection of cluster validation criteria, as a consequence of a significant computational work. The proposed image clustering procedure is based on an input stabilisation diagram image that was constructed and displayed independently at a pre-defined interval frequency, and standardised image features in MATLAB were utilised to extract image features from each generated stabilisation diagram image. The image feature extraction was then used to create an image clustering diagram with a predetermined fixed threshold for classifying physical modes. Even for closely spaced modes, image clustering has been shown to give reliable output results that can recognise actual modes in stabilisation diagrams using image feature extraction, without the need for any calibration, user-defined parameter at start-up, or additional adaptive approach for cluster validation criteria.