
Density-based clustering can effectively identify clusters with different shapes and densities and has therefore attracted extensive attention in the field of data mining. However, efficient estimation of local density and accurate identification of density peaks remain key challenges for existing density clustering methods. To address these issues, this paper proposes an adaptive density clustering algorithm based on neighborhood expansion. First, the dynamic reverse nearest neighbor sequence of each data object is calculated. Second, dynamic information entropy is derived from the reverse nearest neighbor sequence and used to identify density centers. Finally, adaptive clustering is achieved by iteratively expanding neighborhood and dynamically determining local density centers. Experimental results show that the proposed algorithm can accurately identify density centers and adaptively discover clusters with various shapes and densities. In addition, it shows strong robustness to noise and variable-density data. Compared with DBSCAN, DPC, and RNN-DBSCAN, the proposed algorithm achieves superior clustering performance.
Full-scale fatigue testing plays a crucial role in verifying aircraft structural design and durability; however, it involves long test cycles, numerous sensors, and complex operating conditions. Multi-channel data loss often occurs due to uncertainties in data acquisition systems and external interference, directly affecting data integrity and the reliability of subsequent evaluations. To address this issue, this study proposes a self-attention-based multi-channel imputation model for full-scale fatigue test data. The proposed model captures both long- and short-term dependencies through a two-layer diagonal masked self-attention mechanism, focusing on intra-channel local details and depicting inter-channel global correlations. It further integrates attention weights with missing masks via a weighted combination strategy to adapt to different missing mechanisms and noise conditions. In terms of learning strategy, a dual-objective joint optimization framework is adopted, combining masked imputation and observed-value reconstruction with a hierarchical weighted loss function. These designs enable the model to balance local imputation accuracy with global temporal consistency, suppress information leakage, and improve robustness. Comparative and ablation experiments were conducted on real full-scale fatigue test data under various missing patterns and missing rates. The results demonstrate that the proposed method is effective and stable, significantly enhances the integrity of fatigue test data, and provides a reliable data foundation for fatigue anomaly detection and life assessment.
Accurate prediction of the remaining useful life (RUL) of lithium batteries is crucial for predictive maintenance and health management. Existing methods often require large training datasets, exhibit low computational efficiency, and struggle to satisfy real-time application requirements due to variations in operating conditions and manufacturing processes. To address these limitations, this paper proposes a lithium battery RUL prediction method based on multi-scale feature extraction and an extreme learning machine (ELM) ensemble. To capture the spatio-temporal features of complex battery degradation data, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and dilated convolution are employed to extract multi-scale features, enabling accurate identification of modal variations in battery lifetime behavior. An integrated ELM-based RUL prediction model is then built using elastic net (EN) regularization, while incorporating concepts from the broad learning system (BLS) and ensemble learning. This design improves prediction accuracy and robustness for complex sequential data while remaining suitable for resource-constrained scenarios. Experimental results on the university of maryland center for advanced life cycle engineering CALCE dataset reveal that the proposed method yields an 81.25% reduction in RMSE, a 60.43% decrease in MAPE, and a 0.74% improvement in R2 relative to the best-performing deep learning model. Compared to the optimal conventional machine learning model, training time increases by less than 1 ms, while prediction time drops by 45%. These results demonstrate that the proposed method significantly enhances prediction accuracy and robustness, enabling lightweight models to perform better in resource-constrained circumstances.
Straddle-type monorail train gearboxes operate in complex environments where vibration signals exhibit strong non-stationarity, nonlinearity, and high noise levels. In practical operating conditions, gearbox fault samples are difficult to obtain, and critical features are often obscured by noise, leading to overfitting and poor generalization of existing deep learning models for time series prediction. To address these challenges, this paper proposes a topology-guided time-series prediction model designed for small-sample conditions. First, an adaptive period detection mechanism is constructed using the Hilbert-Huang transform and the Teager energy operator to overcome the limitations of traditional methods in extracting features from non-stationary signals. Second, a topological attention mechanism is introduced by mapping vibration signals into a topological point cloud and quantifying multi-scale structural features via persistent homology, thereby generating an attention map with physical interpretability. Finally, a topology attention-guided convolutional neural network is constructed to achieve high-precision vibration signal prediction. Experimental results show that the proposed model achieves a coefficient of determination (R2) approximately 61% higher than that of the TimesNet model, verifying its robustness and efficient predictive performance under small sample conditions.
To evaluate the safety risks associated with the construction and operation of underground engineering, this study takes the complex cavern group of the u201C818 Projectu201D as the research background. A two-level safety evaluation index system was established, including a first-level framework comprising four dimensions (human factors, materials, management, and environment) and a second level system consisting of 15 indictors, such as technical support, personnel quality, cavern seepage and leakage, safety inspection, and rainfall conditions. Four risk levels for cavern group construction and operation were defined: Level u2160 (inadmissible), Level u2161 (undesirable), Level u2162 (allowable), and Level u2163 (negligible). Based on these classifications, a two-level fuzzy comprehensive evaluation model for the operational safety of the u201C818 Projectu201D cavern group was developed. Using the fuzzy evaluation matrices of the four primary indicators together with construction and operational data, the single-factor fuzzy evaluation matrices and index weights for both levels were determined through the analytic hierarchy process (AHP). The proposed model was then applied to predict the construction and operational safety of the cavern group. The evaluation result yields a grade value of u03B4=2.3445, corresponding to risk level u2163 , indicating a low probability of risk occurrence. The findings provide a scientific basis for safety assessment and risk management in the construction and operation of underground cavern groups.