An innovative machine learning system for real time condition monitoring of ship machinery

In this paper an innovative on-line condition monitoring system is introduced. It consists of an object-oriented database, a machine learning algorithm and a model to predict machinery failure. The in-house built Object-Oriented Condition Monitoring Database (CMD), avoids the challenges of relational-object mismatch issues which are common when using relational databases in complex applications involving machine learning techniques. The database intelligently stores data from various sensors and then feeds the data into a pipeline to the diagnostic and prognostic system, offering a constant evaluation of the ship machinery at high speed and accuracy. The suggested Condition Based Maintenance (CBM) framework is based on detecting the change of the condition of the machinery in real time by utilizing the Local Outlier Factor (LOF) algorithm for novelty detection. Two case studies are presented that include real-life data from sensors onboard a tanker ship and prediction of failure of the cylinders of two Diesel Generators (DGs)