Fault diagnosis of belt conveyor drum bearings based on digital twin data augmentation

In data-driven methods for bearing fault diagnosis, the lack or insufficiency of fault data is a key factor limiting diagnostic accuracy. To address this issue, this paper proposes a data augmentation approach based on digital twin data. By augmenting the dataset with digital twin data and comparing it with dataset augmented by traditional simulated signal, we demonstrate that the proposed method can effectively substitute for real data and achieves superior performance to traditional simulation approaches. First, a dynamic model of bearings is established based on Hertz contact theory. Then, by incorporating environmental variables, a digital-twin model of the bearing is constructed. Using this twin model, fault data for missing categories are generated. Finally, four datasets are used to train the same convolutional neural network: the complete dataset, the dataset with missing fault types, the dataset augmented with simulated signal, and the dataset augmented with DT-signal. Experimental results show that the complete dataset achieves an average diagnostic accuracy of 96.00%, the dataset with missing fault types achieves 84.22%, the simulation-augmented dataset achieves 66.56%, and the DT-signal augmented dataset achieves 93.89%. These results indicate that the proposed DT-signal-based method for fault sample supplementation effectively mitigates the degradation in training performance caused by missing fault samples, and that its performance is significantly better than using simulated signal for supplementation.