Semantic Segmentation of Point Cloud in Liquefied Natural Gas (LNG) plants

With Liquefied Natural Gas (LNG) facilities of large capacities, Australia is a major exporter of liquefied natural gas (LNG) in the world. Proper maintenance is essential to safe and reliable operation of existing LNG... [ view full abstract ]

With Liquefied Natural Gas (LNG) facilities of large capacities, Australia is a major exporter of liquefied natural gas (LNG) in the world. Proper maintenance is essential to safe and reliable operation of existing LNG facilities, and relies on accurate as-is model reflecting current conditions. For existing LNG plants without as-designed 3D model, there is an urgent need for efficient and reliable methods for creating as-is models. However, in practice, as-built modelling from point cloud requires intensive manual interactions, and thus is labor intensive and inefficient. To investigate automated methods of as-built modelling, this paper implements a semantic segmentation method to divide point cloud of LNG plants into semantic components. The implemented method considers class-level contexts to constrain segmentation process. Experiments are presented to evaluate the performance of the implemented approach.

Keywords: point cloud, semantic segmentation, LNG plants, as-built modelling