Effect of Database Generation on Damage Consequences' Assessment Based on Random Forests

Recently, the application of machine learning has been explored to assess the main damage consequences without employing flooding sensors. This method can be the base of a new generation of onboard decision support systems to help the master during the progressive flooding of the ship. In particular, the application of random forests has been found suitable to assess the final fate of the ship and the damaged compartments' set and estimate the time-to-flood. Random forests have to be trained using a database of precalculated progressive flooding simulations. In the present work, multiple options for database generation were tested and compared: three based on Monte Carlo (MC) sampling based on different probability distributions of the damage parameters and a parametric one. The methods were tested on a barge geometry to highlight the main effects on the damage consequences' assessment in order to ease the further development of flooding-sensor-agnostic decision support systems for flooding emergencies.

Automated summary

Research has explored using machine learning to assess damage consequences in ship flooding situations, finding that random forests are effective in evaluating the fate of ships and damaged compartments as well as estimating time-to-flood. Training random forests with precalculated flooding simulations, multiple database generation methods were tested to facilitate the development of sensor-agnostic decision support systems for flooding emergencies.