Traditional methods for diagnosing abnormal valve clearance in marine diesel engines face challenges in noise suppression and feature extraction.

To address these issues, this study proposes a method combining adaptive maximum second-order cyclostationarity blind deconvolution (CYCBD) with an efficient wide path dense network (EWPDNet). First, the envelope harmonic product spectrum is used to estimate the cyclic frequency, while the Alpha evolution algorithm optimizes CYCBD filter length, enabling adaptive noise reduction and feature enhancement. Subsequently, EWPDNet integrates a parallel convolution path and an efficient channel attention module into DenseNet to improve the extraction of complex features.

Experimental results demonstrate that the proposed method achieves over 93% accuracy under various noise conditions, with an average accuracy of 95.88%.

The proposed approach provides a basis for the diagnosis of abnormal valve clearance faults in marine diesel engines.

The background noise in the engine room during actual ship navigation leads to the poor accuracy in fault diagnosis methods. To address this issue, this paper proposes a ship motor fault diagnosis method based on complementary ensemble empirical mode decomposition (EEMD) with adaptive noise (CEEMDAN) and a Bayesian residual efficient channel attention network (BRECAN).

First, the noisy motor fault signal is decomposed into multiple intrinsic mode components （IMFs） through adaptive noise CEEMDAN, the noise dominant signal and information dominant signal in the IMF are divided on the basis of detrended fluctuation analysis, and empirical wavelet transform （EWT） is used to de-noise the noise dominant signal. Next, the BRECAN network is constructed, based on the principle of Variational Bayesian (VI-Bayesian ) using the network parameters instead of the traditional network point estimation training method, the parameters are built to simulate the interference of synthetic noise on the model training, and the network is guided by the Residual Efficient Channel Attention (RECA) module to extract the fault difference features. Finally, the effectiveness of the method is verified via a motor fault simulation experimental platform.

The results show that the proposed method can achieve the accurate diagnosis of ship motor faults under strong noise conditions while still maintaining a diagnostic accuracy of over 90% under signal-to-noise ratio of −12 dB.

The results of this study can provide valuable references for the diagnosis of ship motor faults under strong noise conditions.