Lingling Yang
View Full Paper
Hypergraph; Time-frequency Hypergraph; Order Alignment; Multi-sensor Fusion; Main Bearing of Wind Turbine; Fault Diagnosis.
To address the issues that the main bearings in the drive train of offshore wind turbines operate long-term under variable speed, vari- able load, strong noise, and environmental disturbances, with vi- bration energy showing intermittent and sparse distribution, cross- sensor signals suffering from asynchrony and nonlinear coupling, and limited on-site labeling, this paper proposes an order-aligned time- frequency hypergraph multi-sensor modeling framework (OT-TFH). The method first performs order tracking and multi-resolution time- frequency transformation on multi-source signals such as vibration, current, and temperature, then constructs hyperedges through co- activation events in the same time slice and fuses amplitude coher- ence with mutual information to obtain a weighted time-frequency hypergraph, extracts high-order co-occurrence relationships across sensors via attention-based hypergraph convolution, and simultane- ously introduces a lightweight 1D-CNN to capture short-term dy- namics and an operating condition gate to achieve adaptive fusion, which has been integrated into an online prototype system to meet real-time diagnostic requirements. Experiments based on CWRU bearing data and self-built wind turbine drive train data demon- strate that the proposed method achieves higher accuracy and stabil- ity than graph modeling and temporal fusion baselines under complex operating conditions, and the generalization performance of CWRU single-sensor data is improved through data augmentation techniques such as multi-speed and noise disturbance. This study verifies the effectiveness and engineering applicability of the order-aligned time- frequency hypergraph in characterizing high-order coupling across multiple sensors.
Important Links:
Go Back
