Deep learning applications in an industrial process plant: repository of segmented point clouds for pipework components

Abstract

The success of deep learning technology depends on the availability of adequate amounts of data for training deep neural networks. Many repositories of general two- (2D) and three-dimensional (3D) data are available, but relatively few repositories of 3D models from the engineering field exist. In industrial process plants, the 3D shapes of plants are captured accurately by creating point clouds through laser scanning. To develop 3D deep learning models that employ point clouds for process plants, it is necessary to first generate the point cloud data required to train deep neural networks for each constituent part (plant item) of a process plant. This study describes the results of an attempt to construct a segmented point cloud repository of the various pipework sections and fittings that constitute a process plant for use in deep learning applications.