In order to realize the intelligent condition monitoring of marine engine room equipment, machine learning algorithms are introduced and a condition monitoring method based on manifold learning and an isolation forest is proposed.

As condition-monitoring data is multi-dimensional, the proposed method extracts useful features through manifold learning, thereby reducing the dimensions and complexity of the raw data. An isolation forest algrithm is introduced to utilize the normal condition data to train and construct multiple sub forest detectors, realizing the fault monitoring of the target equipment. To validate the proposed scheme, a two-stroke marine diesel engine was developed in Matlab/Simulink to simulate reliable normal and fault condition datasets.

Comparisons of the simulated datasets of the different fault monitoring schemes demonstrate that the proposed method has a highest fault detection rate of 98.5% and lowest false alarm rate of 3%.

The method proposed in this study improves the fault monitoring performance of marine equipment.