A transfer learning-based method for marine machinery diagnosis with small samples in noisy environments

The operating conditions of marine machinery are demanding, and their operational state significantly affects the safety of marine structures. Detecting faults is crucial for machinery health management and necessitates a highly precise diagnostic method. In this paper, we propose a fault diagnosis framework that employs transfer learning and dynamics simulation. A denoising convolutional autoencoder is used to reduce noise when monitoring vibration data in marine environments. To address the challenge of limited sample sizes in marine machinery fault data, a multibody dynamics simulation model is developed to acquire data under fault conditions. The fault features are extracted using a convolutional neural network model. Parameter transfer is applied to enhance the accuracy of fault diagnosis. The effectiveness and applicability of the framework are demonstrated through a case study of a bearing fault dataset.