Selective Feature Reinforcement Network for Robust Remote Fault Diagnosis of Wind Turbine Bearing Under Non-Ideal Sensor Data

Tan, Jinbiao; Wan, Jiafu; Chen, Baotong; Safran, Mejdl; AlQahtani, Salman A.; Zhang, Rui

doi:10.1109/tim.2024.3375958

Cited by 4 publications

(1 citation statement)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, bearings operating under strong interference conditions often encounter challenges such as random impacts and influences from other components, making data acquisition a complex task. The collected data typically contains noise and lacks a clear degradation trend, necessitating RUL prediction algorithms capable of effectively resisting random noise and analyzing intricate features that are highly correlated with future bearing degradation trends from longterm historical data [4]. Existing bearing RUL prediction methods can be classified into physics-based and data-driven [5].…”

Section: Introductionmentioning

confidence: 99%

Ultra-long-term spatiotemporal feature fusion network for bearing remaining life prediction under strong interference

Yin,

Tan,

Wan

2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

Under high noise conditions and random impacts, which constitute strong interference, models often exhibit limited capability in capturing long-term dependencies, leading to lower accuracy in predicting the Remaining Useful Life (RUL) of bearings. To address this issue, a spatiotemporal fusion network capable of ultra-long-term feature analysis is proposed to enhance the accuracy of bearing RUL prediction under substantial interference. This network utilizes a dilated convolution-based lightweight vision transformer encoder to extract spatial features reflecting the short-term degradation state of the bearing. Then, these features are sequentially fed into an adaptive tiered memory unit, based on the multiple attention mechanism and the neuron layering mechanism, to analyze temporal features indicative of long-term degradation. Subsequently, short-term spatial and long-term temporal features are fused for RUL prediction. To validate the robustness and predictive accuracy of the proposed approach under strong interference, a gearbox-rolling bearing accelerated platform is constructed, simulating high noise and random impact conditions. Experiments confirm the high robustness and predictive accuracy of the proposed method under strong interference conditions.

show abstract

Section: Introductionmentioning

confidence: 99%