Graphical Computing Solution for Industrial Plant Engineering

Abstract

When preparing an engineering operation on an industrial plant, reliable and updated models of the plant must be available for correct decisions and planning. However, especially in the case of offshore oil and gas installations, it can hazardous and expensive to send an engineering party to assess and update the model of the plant. To reduce the cost and risk of modelling the plant, there are methods for quickly generating a 3D representation, such as LiDAR and stereoscopic reconstruction. However, these methods generate large files with no inherit cohesion. To address this, we propose to find a solution to efficiently transform point clouds from stereoscopic reconstruction into small mesh files that can be streamed or shared across teams. With that in mind, different techniques for treating point clouds and generating meshes were tested independently to measure their performance and effectiveness on an artifact-rich data set, such as the ones this work is aimed for. Afterwards, the techniques were combined into pipelines and compared with each other in terms of efficiency, file size output, and quality. With all results in place, the best solution from the ones tested was identified and validated with large real-world data sets.