Fault Detection for Marine Diesel Engine Using Semi-supervised Principal Component Analysis

Abstract

The diesel engine is the core power equipment of ships, so keeping it in good working condition is vital for maintaining the overall efficiency of sea transportation. Generally speaking, the principal component analysis (PCA) based fault detection methods are highly depend on the unlabeled information of training data, that is means, the information contained in labeled data is largely ignored. However, the diesel engine working process is always accompanied with small amount of faulty samples, which may seriously affect the performance of the PCA model when the faulty samples are included for modeling. Besides, even though there are enough faulty samples for modelling, labeling these samples is also time-consuming and costly. In this paper, a semi-supervised PCA (SSPCA) is proposed and applied to diesel engine fault diagnosis instead of unsupervised learning, which incorporates both the labeled and unlabeled samples and gains enhanced fault diagnosis performance regarding marine diesel engine. The methodology presented is evaluated in the real-world diesel engine working process, the experimental results indicate good robustness to false alarms and certain practical significance towards the fault diagnosis of engineering object.