Fault Feature Extraction and Diagnosis Method for Marine Diesel Engine Based on Time-delay Embedded manifold Learning

Aiming to address the issue of high-dimensional nonlinear characteristics in fault data due to the complex structure of marine diesel engines and the variability of their operating conditions, this paper proposes a fault feature extraction and diagnosis method for marine diesel engines based on time-delay embedding manifold learning. The method leverages time-delay embedding and manifold learning to effectively capture the intrinsic structure and nonlinear characteristics of the signal, extract high-dimensional nonlinear features, and perform low-dimensional mapping, thereby enabling the extraction of fault features. Based on that, by processing the features to generate 2D time-frequency figures, a deep learning network is employed for fault diagnosis. A marine diesel engine fault dataset was used for validation the effectiveness. The results showed that the diagnostic accuracy of the proposed method was 100%, which demonstrates the effectiveness of the proposed method. Additionally, by comparing two feature extraction methods, the superior performance of the proposed method in marine diesel engine fault diagnosis was further validated.