Marine engines combustion diagnostics employing fourier series and ANN

Safe operations of marine engines is ensured by appropriate maintenance techniques requiring accurate assessment of engine's health status. The use of machine learning methods can considerably enhance the combustion diagnostics and hence facilitate the cost-effective and timely maintenance of marine engines.
This study aims at assessing the potential of Fourier series coefficients (FC) obtained from in cylinder pressure signal and developing an artificial neural network (ANN) model that can support the engine diagnostics of marine engines. A ferry ship with two propulsion engines of the four-stroke type was employed as the reference system in this study. Digital twin of the thermodynamic zero dimensional type, which was calibrated by using the engines shop test measurements, is employed to generate the required data-sets in the whole engine envelop, whilst considering the most typical engine anomalies, including degradation and faults.
The results demonstrate that first 20 harmonics contains required information to estimate fault severity within 0.016 RMSE range.