A data augmentation algorithm for surface inspection in point cloud data

Abstract

Due to the high standards in aircraft maintenance, high resolution sensors, such as white light interferometers, are needed. Those sensors scan surfaces in nanometer scale and generate point clouds. This data can be used to detect surface defects. Such anomalies should be identified during the inspection process to assess the current condition of the workpiece. Deep learning algorithms can be used to evaluate the data. However, in the domain of 3D data, the challenge of obtaining training data is amplified due to the time-consuming labeling process. Therefore, this work introduces an algorithm that combines surface features, like cracks, into surface data to generate new labeled training data. The resulting dataset is then used to train a deep learning algorithm to segment the cracks in the point cloud data. The results indicate that the augmented data enhances the training process.