Timely and effective maintenance is imperative to minimize operational disruptions and ensure the reliability of marine vessels. However, given the low early warning rates and poor adaptability under complex conditions of previous data-driven fault prediction methods, this paper presents a hybrid deep learning model based on multi-layer attention mechanisms for predicting faults in a marine diesel engine. Specifically, this hybrid model first introduces a Convolutional Neural Network (CNN) and self-attention to extract local features from multi-feature input sequences. Then, we utilize Long Short-Term Memory (LSTM) and multi-head attention to capture global correlations across time steps. Finally, the hybrid deep learning model is integrated with the Exponential Weighted Moving Average (EWMA) to monitor the operational status and predict potential faults in the marine diesel engine. We conducted extensive evaluations using real datasets under three operating conditions. The experimental results indicate that the proposed method outperforms the current state-of-the-art methods. Moreover, ablation studies and visualizations highlight the importance of fusing multi-layer attention, and the results under various operating conditions and application scenarios demonstrate that this method possesses predictive accuracy and broad applicability. Hence, this approach can provide decision support for condition monitoring and predictive maintenance of marine mechanical systems.

The safe operation and intelligent optimization of ships heavily rely on the predictive maintenance of marine engine performance. However, existing methods for predicting engine performance face limitations due to the various characteristics of ship engines, which depend on the sailing state and environmental conditions. To address this, a novel deep learning framework, the parallel channel-temporal attention-based temporal convolutional network (PCTA-TCN), is proposed to predict marine engine performance across varying load ranges. The PCTA-TCN framework introduces channel attention and temporal attention mechanisms in two independent branches, adaptively adjusting the focus on important features and temporal dependencies in the measured data. Parallel TCN backbones are used to extract high-level feature representations from different sequences. Finally, the model generates predictions by performing a nonlinear mapping on the extracted features at multiple levels. Experimental validation using sensor-acquired data demonstrates that PCTA-TCN consistently achieves an R² value exceeding 0.956 across all datasets. Compared to the baseline TCN model, the proposed method improves prediction accuracy by 13.95%, 49.81%, 15.64%, and 19.73% under four different load conditions. Furthermore, PCTA-TCN outperforms other state-of-the-art methods, particularly under complex and variable operating conditions.

Condition monitoring and fault diagnosis of marine machinery are essential for ensuring the safety and stability of offshore operations. Traditional approaches, which typically rely solely on either physical mechanisms or large volumes of data, often struggle to achieve both high accuracy and adaptability to component degradation. To address this issue, this paper proposes a novel digital twin framework based on deep multimodal information fusion (DMIF-DT) that integrates the complementary strengths of a physics-based model (PBM) and a data-driven model (DDM). An attention-enhanced multi-scale convolutional network is developed to extract features from sensor measurements and calibrated simulation data. A low-rank fusion strategy is then employed to map the two modalities into a unified high-dimensional space for effective integration between the PBM and the DDM. Furthermore, a performance-degradation adaptive correction module, based on sensor measurements and probabilistic distribution features, is used to map health factors and feed them back to the PBM for component-level correction, thereby supporting periodic model updates. Experimental results show that the proposed DMIF-DT framework significantly reduces parameter prediction errors and achieves an average fault-diagnosis accuracy of 98.02% under diverse operating conditions. By uniquely integrating a performance-degradation adaptive correction module with attention-enhanced multimodal fusion, the framework bridges physics-based and data-driven insights, providing a new approach for the intelligent health management of marine machinery.