Shipyard Classification, Analysis, and Neural-Net software (Makai-SCAN)

Efficient and high-performance operations at the US Naval Shipyards remains a vital capability, providing the essential maintenance and construction of some of the US NavyÆs highest priority vessels and marine assets. Productivity and quality remain critical. A majority of shipyard operations remain labor intensive and are managed without the aid of automated tools, including infrastructure inspection and planning. Modernization of the Navy Shipyard Facilities is needed to maintain the NavyÆs competitive edge, reducing shipyard labor, resources, and production times, while simultaneously increasing quality. Specifically, there are limited technologies in place for real-time and automated monitoring of the shipyard infrastructure and construction progress. In the last several years there have been tremendous improvements in artificial intelligence, machine learning, and three-dimensional scanning that can be leveraged to support shipyard modernization, but have not yet been integrated together into a coherent and demonstrably viable product. A new integrated hardware-software solution that combines the latest in three-dimensional (3D) scanning systems (including stereo vision, LiDAR, and subsea multibeam sonar) with three-dimensional machine learning enabled data classification and segmentation would provide a step-change improvement in the ability to obtain real-time structural monitoring and quantification, reducing human labor, improving quality, and providing a near-real time shipyard digital twin representation. Makai Ocean Engineering (Makai) proposes to develop the Shipyard Classification, Analysis, and Neural-Net software (Makai-SCAN). The proposed tool will utilize three-dimensional point cloud data collected from aerial, surface, and subsea scanning sensors and process through a novel Machine Learning (ML) model that is trained to detect, classify, and segment the critical infrastructure. Each individual structural element or item will be clustered into its own unique point cloud enabling subsequent inspection and analysis at the component level, and the generation of CAD-level geometric and digital twin models.