Analysis of Collected Data and Establishment of an Abnormal Data Detection Algorithm Using Principal Component Analysis and K-Nearest Neighbors for Predictive Maintenance of Ship Propulsion Engine

Abstract

Because ships are typically operated for more than 25 years after construction, they can be considered mobile factories that require economic maintenance before being scrapped. Therefore, for stable and efficient ship operation, continuous maintenance systems and processes are required. Ships cannot be operated when defects or failures occur in any of the numerous systems configured in them, and research is urgently needed to apply predictive maintenance to propulsion engines with high maintenance costs using machine learning. Therefore, this study analyzes the operation and control characteristics of the propulsion engine, acquires engine data from the alarm monitoring system of the ship in operation, and then preprocesses the data by constructing a data preprocessing algorithm that incorporates the engine control characteristics. In addition, principal component analysis and K-nearest neighbors were used to check whether preprocessing data were classified based on engine control characteristics, and an algorithm capable of detecting abnormal data was built and verified to lay the foundation for predictive maintenance of ship propulsion engines using machine learning.