A Multi-Source Fusion Spatiotemporal Neural Network Improved by Koopman Operators for Predicting Remaining Useful Life

A Knowledge-Enhanced Spatiotemporal Framework for Remaining Useful Life Prediction of Aero-Engines

Remaining useful life (RUL) prediction is critical for improving reliability and supporting predictive maintenance in aero-engine systems. However, existing methods have limitations in jointly modeling the spatial correlations between multi-sensor signals and the temporal evolution characteristics of the degradation process. Hence, this study develops a knowledge-enhanced spatiotemporal framework for system-level aero-engine RUL prediction. Firstly, a graph based on the Pearson correlation coefficient (PCC) is constructed from monitoring data to capture data-driven dependencies among sensors. Afterwards, a thermodynamic-cycle-mechanism prior is incorporated into the PCC-based graph through the Hadamard product, forming a knowledge-enhanced graph that emphasizes physically meaningful sensor relationships. Subsequently, an enhanced graph attention module is designed to extract discriminative spatial representations from the knowledge-enhanced graph. Furthermore, relational representations between adjacent time steps are constructed to capture implicit temporal correlations and local degradation dynamics. Finally, a dual-stream GRU with an attention mechanism is employed to model the fused feature stream and relational feature stream for RUL prediction. Experiments on the CMAPSS and N-CMAPSS datasets demonstrate that the proposed method achieves competitive and overall superior performance compared with nine state-of-the-art methods. KESTF achieves the best average RMSE/Score of 12.83/580 on CMAPSS and 5.94/3468 on N-CMAPSS, validating its effectiveness and robustness.

MFSNet: lightweight multi-scale MLP-guided frequency suppression network for multivariate time series forecasting

Remaining useful life prediction of electronic power components based on stacked denoising autoencoders and temporal fusion networks.

Electronic components are prone to degradation or even failure due to the influence of environmental factors during their wide application. Accurate prediction of their remaining useful life (RUL) is therefore crucial for implementing early fault warning systems. However, traditional lifespan prediction methods face challenges in simultaneously capturing both short-term fluctuations and long-term trends in the degradation process of devices. To address this issue, this paper proposes a lifespan prediction method for Electronic components based on a stacked denoising autoencoder-temporal fusion transformer (SDAE-TFT). Firstly, 19 features are extracted from raw degradation data, and these features are subjected to dimensionality reduction and fusion through a SDAE to construct a health index (HI). Secondly, a TFT model is built to output the root mean square (RMS) voltage and power characteristics of the device, thereby predicting its degradation trend. In this study, a data acquisition system for DC-DC modules is established to carry out accelerated degradation experiments. The SDAE-TFT model is used to predict the RUL of the modules, and its accuracy is further verified using the NASA public dataset. Comparative experiments show that, compared with existing models, the proposed model reduces the mean squared error (MSE) by 34.1%.

DeRNN: Decomposed Recurrent Neural Network for Long-Term Time Series Forecasting

The Decomposed Recurrent Neural Network (DeRNN) is proposed, which decouples global trend modeling from local fluctuation extraction via an asymmetric dual-track architecture and exhibits superior robustness against noise and distribution shifts.

Hierarchical Adaptive Transformer Framework for Modeling Abrupt Short-Term Fluctuations of Hazardous Gas Concentrations in Industrial Air

Short-term prediction of hazardous gas concentrations is crucial for industrial air monitoring, but conventional approaches often fail to capture abrupt local fluctuations and nonlinear temporal dependencies, limiting prediction accuracy. To address these limitations, this study develops a multi-task residual Transformer-based framework for short-term concentration forecasting. First, historical high-frequency H2S measurements are processed using a sliding-window approach to form input sequences for the model. Next, a shared Transformer encoder extracts temporal features, while task-specific branches perform residual concentration prediction and concentration-based emission-state classification. Within this multi-task framework, an adaptive weighting mechanism emphasizes high-variation samples during training to improve sensitivity to rapid concentration changes. Experiments conducted on data from the South Coast Air Quality Management District demonstrate that, averaged over three random seeds, the model achieves an MAE of 0.133±0.001, an RMSE of 0.237±0.000, and an R2 of 0.810±0.001 for one-observation-step forecasting. These results show that the proposed framework effectively captures abrupt rises and peak concentrations, providing a reliable tool for industrial emission monitoring and early warning applications.

A Remaining Useful Life Prediction Method for Aero-Engines Based on Degradation-Aware Masked Augmentation and a CNN–Transformer Hybrid Network

Accurate remaining useful life (RUL) prediction is essential for condition-based maintenance and safe aero-engine operation. To address non-stationary monitoring data, limited informative degradation samples, and the difficulty of jointly modeling local degradation patterns and temporal dependencies, this study proposes a degradation-aware dynamic masking augmentation method combined with a multiscale CNN–Transformer network. The 21 sensor variables in the NASA C-MAPSS dataset are first grouped by physical meaning and reconstructed into four-dimensional state features. A degradation-state score integrating local variance and trend slope is then used to adapt temporal masking probabilities across degradation stages, while feature masking probabilities are assigned according to feature importance. Masked positions are filled with adjacent unmasked observations, and invalid augmented samples are removed through trend consistency verification. A multiscale CNN with channel attention extracts local degradation features, and a Transformer encoder captures temporal dependencies. Bayesian optimization is used to determine key hyperparameters. On FD001, the proposed method achieves an MSE of 348.3970, an MAE of 8.3908, and an R2 of 0.9227, reducing MSE and MAE by 4.27% and 28.42%, respectively, compared with ML-RFR. It also achieves the highest R2 of 0.9369 on FD003. Cross-dataset and multi-seed experiments further confirm its applicability and stability.