A fault diagnosis method of marine diesel engine valve clearance based on adaptive CYCBD and EWPDNet

Objectives Traditional fault diagnosis methods for abnormal valve clearance in marine diesel engines struggle with noise suppression and feature extraction. Methods To address this, we propose a method combining adaptive maximum Second order cyclostationarity blind deconvolution (CYCBD) and an efficient wide path dense network (EWPDNet). First, envelope harmonic product spectrum estimates the cyclic frequency, while the Alpha evolutionary algorithm optimizes CYCBD filter length, enabling adaptive noise reduction and feature enhancement. Then, EWPDNet integrates a parallel convolution path and an efficient channel attention module into DenseNet to improve complex feature extraction. Results Experimental results show that the proposed method achieves over 93% accuracy under various noise conditions, with an average diagnosis rate of 95.88%. Conclusions The findings provide a basis for diagnosing abnormal valve clearance faults in marine diesel engines.