Digital twin-driven bearing fault diagnosis framework for incomplete fault data scenarios

In practical industrial applications, fault data of rolling bearings are often scarce, and the available data may cover only partial fault types or limited fault sizes, posing challenges for training diagnostic models. To address the issue, this study proposes a digital twin – driven bearing fault diagnosis framework. The framework employs a digital twin model incorporating an adaptive information transfer strategy (DT-AIT) to generate high-fidelity fault samples for bearing fault diagnosis in incomplete fault data scenarios. DT-AIT simulates the missing fault samples using a bearing dynamic model and adaptively integrates relevant components extracted from available measured signals to minimise the distribution variations between simulated and real data. A multi-scale progressive convolutional neural network (MSPC) is then proposed to perform fault diagnosis by progressively extracting deep features from the input signals. Experimental results demonstrate that the integration of DT-AIT and MSPC effectively mitigates the impact of the absence of some fault samples and substantially improves bearing fault diagnostic accuracy in incomplete fault data scenarios.