Machine learning and data-driven fault detection for ship systems operations

Cheliotis, Michail and Lazakis, Iraklis and Theotokatos, Gerasimos (2020) Machine learning and data-driven fault detection for ship systems operations. Ocean Engineering. ISSN 0029-8018 (In Press)

[thumbnail of Cheliotis-etal-OE-2020-Machine-learning-and-data-driven-fault-detection]
Preview
 Text. Filename: Cheliotis_etal_OE_2020_Machine_learning_and_data_driven_fault_detection.pdf
Accepted Author Manuscript
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo
Download (838kB)| Preview

Abstract

Well maintained vessels exhibit high reliability, safety and energy efficiency. Even though machinery failures are inevitable, their occurrence can be foreseen when predictive maintenance schemes are implemented. Predictive maintenance may be optimally applied through condition, performance, and process monitoring. Most importantly, it can include the detection of developing faults, which affect the performance of ship systems and hinder energy-efficient operations of ships. Under this viewpoint, this paper proposes a new data-driven fault detection methodology in a novel application for shipboard systems, by exploring the "learning potential" of recorded voyage data. The proposed methodology, combines the benefits of Expected Behaviour (EB) models, by selecting the optimal regression model, with the Exponentially Weighted Moving Average (EWMA) for fault detection, in novel ship applications. It is seen that a multiple polynomial ridge regression model, with testing R2 score of nearly 0.96 and can accurately detect certain developing faults manifesting in both the Main Engine (ME) cylinder Exhaust Gas (EG) temperature and the ME scavenging air pressure. The early detection of developing faults can be used to supplement the daily monitoring of ship operations and enable the planning of pre-emptive rectifying actions by reducing sub-optimal machinery conditions.

CORE (COnnecting REpositories)