Geometrical Identification of Defect in Parts Using Imaging and Photogrammetry Using Intelligent Methods

This paper presents a damage identification method for components using intelligent techniques and tools. In this method, the damaged area on the point cloud of the affected component is first determined using photogrammetry, followed by clustering. The damage volume is calculated by fitting a plane to the damage boundary, enclosing the volume between the internal damaged surface and the fitted plane. The point cloud clustering is performed using the K-means method, while plane fitting and point cloud alignment are achieved using features from the segmented image of the component, including the damage center and the upper edge line of the model. Image segmentation is carried out using Mask R-CNN to isolate different objects. After completing the above steps, the toolpath for deposition is generated within the resulting volume and fed into a simulated robotic arm mechanism to guide the laser deposition system. This method is particularly useful for small-scale damage, especially in cases where an undamaged reference sample of the component is unavailable.