RUL Prediction for CT X-Ray Tube Based on Degradation Pattern Recognition and Multidomain Joint Feature Fusion

Accurate remaining useful life (RUL) prediction of X-ray tubes is critical for ensuring the reliable operation and imaging stability of computed tomography equipment. However, the diverse degradation patterns and significant signal fluctuations in X-ray tubes obscure degradation characteristics, making RUL prediction particularly challenging. To address this issue, this paper proposes a RUL prediction framework for X-ray tubes based on degradation pattern clustering and multi-domain feature fusion. First, early-stage filament current modulation and clustering are employed to identify degradation patterns, mitigating domain distribution discrepancies. Second, considering periodic fluctuations and inherent noise in X-ray tube degradation, filament signals are decomposed into trend, periodic, and residual components to extract multi-scale time-frequency features. Third, a multi-head attention mechanism is employed for deep time-frequency feature fusion, facilitating recursive RUL prediction. Finally, the proposed method is validated using realworld operational data from millions of scans collected via an internet of medical things platform. Experimental results demonstrate that the proposed method outperforms existing models in terms of RMSE, achieving values as low as 0.019. This highlights marked improvements in prediction precision, especially near end-of-life conditions.