Hui Cao

To address the challenges of knowledge acquisition and low information utilization efficiency in marine diesel engine fault diagnosis, this study develops a knowledge graph-based question answering (QA) system integrating intent recognition and named entity recognition (NER). First, a domain-specific knowledge graph is constructed to provide a structured representation of multi-source fault knowledge. For semantic understanding, RoBERTa- Whole Word Masking (WWM) is employed for intent recognition, while an RW-Bidirectional Long Short-Term Memory (BiLSTM)-SD-conditional random field (CRF) (RoBERTa-WWM-BiLSTM-scaled dot-product attention-CRF)-based NER architecture is adapted to domain-specific fault descriptions. Based on the recognized intents and entities, predefined Cypher query templates are instantiated and executed on a Neo4j graph database to retrieve relevant diagnostic knowledge. Experimental results show that the adapted NER architecture achieves an F 1-score of 92.19%, while the RoBERTa-WWM-based intent recognition model achieves an overall F 1-score of 88.85%. The end-to-end QA system achieves an accuracy of 90.40% (452/500) on an independent evaluation set of 500 fault diagnosis questions. The results indicate that the proposed approach can effectively organize dispersed marine diesel engine fault knowledge and support practical fault diagnosis QA.

In real-world industrial environments, frequent condition changes, mechanical degradation, and incomplete fault labeling often lead to data distribution shifts and label space asymmetry between the source and target domains. Moreover, compounded by the emergence of previously unseen fault types, severely undermine the generalization capability of conventional intelligent diagnostic methods. Although existing open-set domain adaptation methods attempt to address unknown classes, most still rely on the assumption of full label space consistency, limiting their applicability under complex and uncertain industrial conditions. To overcome these limitations, this paper proposes a novel model for generalized open-set cross-working condition fault diagnosis, named Multi-Strategy Progressive Adaptation Network (MSPAN). The model allows for partial class overlap between source and target domains, each containing private classes, thereby reducing reliance on strict label alignment, complete class coverage, and prior knowledge of the target domain. This significantly enhances the model’s adaptability and flexibility under realistic operating conditions. MSPAN integrates three core strategies: Centroid-Guided Expansion (CGE), Progressive Consensus Filtering (PCF), and Localized Knowledge Integration (LKI). CGE expands the label space with pseudo-target domain samples, alleviating interference from private classes in the source domain; PCF combines Dual-Measure Driven Ranking and Consensus-Aware Estimation to adaptively filter unknown classes in the target domain; LKI focuses on learning unknown-class representations while enhancing the transferability of general diagnostic knowledge across domains. Extensive experiments on the PU and NLN-ESP public datasets demonstrate that MSPAN achieves both accurate shared-class diagnosis and effective unknown-class separation under varying degrees of class overlap and openness. These results validate its robustness, generalization ability, and practical potential for deployment in complex industrial fault diagnosis scenarios.

To explore and improve the economic efficiency and nitrogen oxide (NOx) emissions of marine natural gas-diesel dual-fuel engines, this study proposes a performance optimization method integrating machine learning and multi-objective optimization based on bench tests and simulation models. Firstly, a GT-Power simulation model was established and calibrated. Four machine learning models were developed and compared, with the XGBoost model demonstrating superior prediction accuracy and generalization ability, achieving average R² values of 0.9979 and 0.9874 for the training and testing datasets, respectively. Subsequently, the MOEA/D algorithm was employed for multi-objective optimization of dual-fuel engine performance. The results indicate that MOEA/D exhibited excellent optimization performance under all load conditions, especially at high loads, effectively balancing fuel consumption (BSFC) and nitrogen oxide emissions (BS-NOx). CV-TOPSIS decision-making resulted in 17.63%–28.58% BSFC reduction and 26.8%–69.14% BS-NOx abatement across load conditions. The proposed methodology improved computational efficiency by over 10× compared to conventional simulation, providing an effective solution for low-carbon operation of marine dual-fuel engines.

The key to reducing operation and maintenance costs and improving reliability is to evaluate the condition and fault detection methods of ship ballast water systems. To reduce the impact of support vector machine (SVM) parameter uncertainty and improve the accuracy of ship ballast water system fault diagnosis models, this paper proposes a multi-strategy collaborative improved sparrow search algorithm (ISSA) to optimize the parameters of SVM for fault diagnosis. First, ISSA initializes the population using tent chaotic maps to improve spatial distribution uniformity and stabilize the population quality. Second, the discoverer in the algorithm adopts an adaptive weighting strategy to avoid falling into local optima and accelerate the convergence speed. Finally, the follower position introduces mutation strategies and chaotic perturbations to enhance local evolution capabilities and improve the utilization of search areas. The experimental results show that compared with sparrow search algorithm (SSA), particle swarm optimization algorithm (PSO), Genetic Algorithm (GA), whale optimization algorithm (WOA), and grey wolf optimization algorithm (GWO), ISSA can obtain better optimal solutions at the fastest convergence rate on the benchmark test function. On the dataset of the ship ballast water system, the ISSA-SVM fault diagnosis model proposed in this paper has the best classification performance, with an average accuracy of 97.44%. Compared with the other two models, ISSA-SVM is 7.55% and 0.83% higher than SVM’s 89.89% and SSA-SVM’s 96.61%.

The entire life cycle of an engine is an asymmetric process, and the characteristics of its internal components are different. It is of great significance to extract engine degradation features and build models for improving the RUL prediction of the main engine. In this paper, a RUL prediction method based on CBAM attention mechanism is designed. For long time series engine data, using Convolutional Neural Network (CNN) to extract features and share convolutional kernels can process high-dimensional data well. The purpose of adding a Gated Recurrent Unit (GRU) network is to use the GRU network to sequentially model the feature expressions extracted from the CNN network and learn the sequential correlation of data. Convolutional block attention module (CBAM) is introduced as a mechanism to reduce the influence of different data lengths on the final prediction by focusing on the channels and spaces. This paper validates the above prediction method on two different engine datasets. The prediction method can effectively mine the internal time and space dependence between multiple sensor data, improve the precision of RUL prediction, and prove that the method has good applicability and effectiveness.