NEMO: a neural motion estimator for mooring line failure detection of offshore platforms.

• Ano de defesa: 2023
• Autor(a) principal: Sa\'ad, Amir Muhammed
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aprendizado computacional; Classificação; Classification; Estruturas offshore; Machine learning; Mooring lines breakage; Neural networks; O?shore platforms; Redes Neurais; Rompimentos de cabos de ancoragem
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/3/3141/tde-27042023-080040/

Floating oshore structures are maintained in the desired position by mooring lines attached to the seabed of the location. These systems are among the main components that guarantee not only the safety of the crew but also the various operations carried out on the platforms. In this thesis, the objective is to detect the rupture of the mooring lines of platforms with dierent levels of draft (load) based on the measurements of the platform motion provided by the Dierential Global Positioning System (DGPS) and Inertial Measurement Unit (IMU) sensors. For this, a Neural Motion Estimator (NeMo) system was developed. NeMo consists of two modules: a motion prediction module comprising of a feed forward neural network (Multilayer Perceptron MLP), which uses previous data from platform motions to predict future motion, and a multi-class classifier module, which uses the dierence between predicted motion and measured actual motion as inputs to indicate whether or not there has been a failure, for various groups of mooring lines. The system was trained and tested using simulated data from a time- domain platform motion simulator. Results of the implemented NeMo system showed it is able to detect the occurrence of failure in the mooring lines, with errors between the forecast and the measured movements when there was a line breakage. These errors are such that the developed multi-class classifier had a 99% accuracy prediction rate when classifying the platform motions.