Turbomachinery blade behaviour analysis using a photogrammetric stereovision 3D based shape analysis approach

• Application of a 3D shape-based photogrammetric technique for condition monitoring. • Introduction of PCA-based geometric modes for boundary shape variation analysis. • Online turbomachinery blade analysis using a statistical approach. • Blade damage detection and classification through clustering of shape features. Vibrational data analysis is a widely popular tool employed for condition monitoring and asset integrity analysis of rotating machinery. Due to the intrusive nature of conventional contact transducers such as accelerometers, noncontact photogrammetric techniques such as 3D point tracking (3DPT) and digital image correlation (DIC) are considered viable alternatives for certain applications. These techniques have been successfully used to capture operational vibration data of turbomachines such as wind turbines. These techniques, however, typically require prior surface preparation by way of either attaching distinct traceable markers or spray painting of the specimen. Prior surface preparation not only necessitates costly machine shutdowns, but can also be very complex to implement. This paper presents a novel approach focusing on 3D shape variation analysis without the requirement for prior surface preparation. The analysis is performed on blade profile contours extracted from images of rotating blades. This novel approach incorporates statistical shape descriptor analysis on rotating blades to capture dynamics, identifying damage, and perform damage severity classification of the blades. The definition of principal component based geometric modes for a 3D shape is presented, and the relationship between these modes and the outline of a shape of interest successfully investigated. Shape Principal Component Descriptors (SPCDs) that can be analysed to characterize the dynamic properties of a machine are proposed. In conjunction with a Finite Element (FE) numerical model of a physical rotor, it is shown that these uniquely defined descriptors can be used to capture differences in the dynamic behaviour of rotor blades. A clustering technique is employed to classify individual damaged blades, and results from investigations to better understand descriptor sensitivity to vibration amplitude and shape orientation transformation are presented. Results from an experimental study of the physical rotor utilizing a stereoscopic camera setup are also presented. Results from principal component analysis (PCA) of the SPCDs further demonstrate the feasibility of this shape-based measurement technique for online condition monitoring of rotating turbomachines.