DegraDiff-Transformer: A Degradation Trajectory Generation Framework to RUL Prediction under Variable Operating Conditions

Accurate prediction of Remaining Useful Life (RUL) for critical equipment, such as industrial pumps, is paramount for ensuring system safety; however, this task faces significant challenges under variable operating conditions (VOCs), such as fluctuating frequencies and loads. The primary difficulty lies in the deep entanglement of signal fluctuations caused by varying operating conditions with the intrinsic degradation features of the equipment, which hinders traditional models from achieving reliable predictions. To address this issue, this paper introduces DegraDiff-Transformer, a probabilistic degradation trajectory generation framework. Initially, the framework leverages the Hilbert-Schmidt Independence Criterion (HSIC) as a novel regularization constraint to explicitly disentangle and extract purified degradation representations from raw, multi-modal sensor data. Subsequently, we reformulate RUL prediction into a conditional generation task based on these decoupled features. A generative model is constructed to learn and simulate the complete stochastic degradation process from the current health state to failure, thereby quantifying the prediction’s uncertainty. To further enhance the model, we integrate physical prior knowledge into the core Transformer architecture. This is achieved by designing a novel, decay-style positional encoding mechanism inspired by the Weibull distribution. This mechanism simulates the physical property of ”memory decay” in degradation processes, guiding the model to focus more on recent health states and thus enhancing the physical plausibility and accuracy of the predictions. The proposed method’s effectiveness was validated on two datasets: a real-world industrial submersible pump dataset (ZN-Pump) and the public C-MAPSS turbofan engine dataset. Experimental results demonstrate that the proposed method maintains its effectiveness in RUL prediction for rotating machinery when subjected to abrupt changes in operating conditions.