Multivariate time series forecasting requires modeling both hierarchical temporal dynamics and complex inter-variable dependencies, a dual requirement that often degrades predictive performance and incurs high computational costs in standard Transformer architectures. Unlike current channel-independent models that ignore vital cross-variable synergies, or dense-attention frameworks that suffer from quadratic computational noise, our approach extracts structurally sparse dependencies. To address these specific limitations, this study introduces the FCA-Transformer. The proposed framework integrates a Feature Pyramid Network (FPN) to isolate macroscopic trends from high-frequency localized fluctuations via hierarchical downsampling. Concurrently, a structured Transformer-based Cross-Attention (TCA) mechanism employs Dimensional Segmentation with Weighting (DSW) and a Two-Stage Attention (TSA) layer to map topological variable interactions, effectively extracting robust cross-variable pathways and mitigating distributional noise. Extensive empirical evaluations across three real-world multivariate benchmarks (ETTh1, Electricity, and Exchange Rate) demonstrate that the FCA-Transformer achieves an average reduction of up to 4.39% in MSE and 5.11% in MAE compared to leading baselines. These findings indicate that the proposed architecture successfully reconciles multi-scale feature extraction with lightweight dependency modeling, enhancing structural generalization and providing a scalable framework for real-time temporal analysis in complex industrial environments.
Fault diagnosis is a cornerstone of mechanical equipment stability. With accurate health identification dictating system reliability, achieving high-reliability diagnosis is imperative. However, in practical scenarios, mechanical equipment often operates under variable speeds, fluctuating loads, and severe noise, leading to non-stationary vibration characteristics and feature distribution shifts that impede conventional diagnostic approaches. To address these challenges, a multi-scale frequency-temporal network (MSFTNet) is proposed. The framework begins by integrating frequency-domain transformation with large-kernel convolutions to extract features with a broad receptive field in the frequency domain. This design enhances sensitivity to weak fault signatures and improves robustness against noise. Subsequently, the network employs multi-scale convolutions coupled with a bidirectional long short-term memory network applied along the frequency dimension. This allows the model to concurrently capture fine-grained local patterns within specific frequency bands and model global contextual dependencies across different bands, effectively addressing the feature distribution shift induced by varying operational conditions. Finally, a residual enhancement module and a deep classifier are utilized to stabilize feature fusion and achieve precise fault classification. Extensive experimental results on several bearing and gear datasets, including SDUST and HUST, demonstrate the model’s powerful ability to extract weak fault features and excellent anti-interference performance. Specifically, MSFTNet achieves an average accuracy of 99.58% under variable speeds and maintains a robust accuracy of 70.14% even under extreme noise conditions (SNR = −6 dB), outperforming baseline methods across a range of complex scenarios.
Deguang Li, Zhen Ding, Yixin Chen et al.· Engineering Research Express· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
