Remaining useful life (RUL) prediction is a cornerstone of Prognostic and Health Management (PHM) for power machinery, playing a crucial role in ensuring the reliability and safety of these critical systems. In recent years, deep learning techniques have shown great promise in RUL prediction, providing more reliable and accurate outcomes. However, existing models often struggle with comprehensive feature extraction, especially in capturing the complex behavior of power machinery, where non-linear degradation patterns arise under varying operational conditions. To tackle this limitation, we propose a multi-feature fusion model leveraging a dual-attention mechanism. Initially, convolutional neural networks (CNNs) and channel attention mechanisms are employed to preliminarily extract spatial features. Subsequently, a layer combining a Gate Recurrent Unit (GRU) and self-attention mechanisms is used to further extract and integrate temporal features. Finally, RUL values are predicted via regression. The effectiveness of the proposed method was validated on C-MAPSS datasets, and its superior performance in RUL prediction was demonstrated through comparative analysis with other methods.

Informed machine learning (IML), which strengthens machine learning (ML) models by incorporating external knowledge, can get around issues like prediction outputs that do not follow natural laws and models, hitting optimization limits. It is therefore of significant importance to investigate how domain knowledge of equipment degradation or failure can be incorporated into machine learning models to achieve more accurate and more interpretable predictions of the remaining useful life (RUL) of equipment. Based on the informed machine learning process, the model proposed in this paper is divided into the following three steps: (1) determine the sources of the two types of knowledge based on the device domain knowledge, (2) express the two forms of knowledge formally in Piecewise and Weibull, respectively, and (3) select different ways of integrating them into the machine learning pipeline based on the results of the formal expression of the two types of knowledge in the previous step. The experimental results show that the model has a simpler and more general structure than existing machine learning models and that it has higher accuracy and more stable performance in most datasets, particularly those with complex operational conditions, which demonstrates the effectiveness of the method in this paper on the C-MAPSS dataset and assists scholars in properly using domain knowledge to deal with the problem of insufficient training data.

Under heavy load conditions, bearings are subjected to non-uniform and frequently changing loads, which leads to randomness in the spatial distribution of bearing degradation characteristics. Aiming at the problem that the traditional degradation index cannot accurately reflect the degradation state of heavy-duty bearings in the whole life cycle, a new degradation evaluation method based on multi-domain features is proposed in this paper, which aims to capture the early degradation point of heavy-duty bearings and characterize their degradation trend. Firstly, the energy entropy feature is obtained by improving the wavelet packet decomposition, and the original multi-domain feature set is constructed by combining the time domain and frequency domain features. Then, the optimal feature matrix is formed by using the comprehensive evaluation index. Finally, integrating probability and distance information, a comprehensive degradation index was constructed to evaluate the degradation, determine the initial degradation time, and quantitatively analyze the bearing degradation state. The validity of the proposed method is verified in two datasets. The proposed method can accurately identify the early degradation of bearings and track the state of bearing degradation, so as to realize the degradation assessment.

To improve the accuracy of bearing load measurement in propulsion shafting, a bearing load measurement method based on redundant strain and jacking data fusion is proposed. Firstly, to address the issue of low measurement accuracy in strain gauge methods (SGM), a redundant strain gauge method (RSGM) is introduced to enhance the precision of strain-based measurements. Secondly, synchronize the measurement of strain data during the jacking phase and fuse the measurement data from the jacking phase based on the Bayesian estimation method to enhance the measurement accuracy of the jack-up method (JM). Lastly, the pre-decision results of the two types of measurement data are integrated at the decision level to obtain the final measured load. Simulation experiments have verified the correctness and applicability of the proposed measurement method. The research results indicate that, compared to the SGM, the RSGM offers higher measurement precision; the fusion measurement method can achieve more accurate results than the JM and RSGM under different bearing support conditions and jack-up scenarios.