Remaining useful life prediction of complex equipment based on adaptive dual-branch multi-scale feature fusion model

Precise Remaining Useful Life (RUL) prediction is critical for ensuring the reliability and operational efficiency of complex equipment. However, existing data-driven methods are often limited by their inability to capture multi-scale degradation dynamics or sustain performance under fluctuating operating conditions. To overcome these limitations, this paper proposes an Adaptive Dual-Branch Multi-Scale Feature Fusion (ADB-MSF) framework. The architecture integrates a self-attention mechanism to elucidate inter-sensor correlations, followed by a multi-scale Temporal Convolutional Network (TCN) that simultaneously captures transient anomalies and long-term degradation trends. A gated enhancement module is further incorporated to refine feature representations by highlighting informative features while suppressing irrelevant noise. Uniquely, the model features an adaptive dual-branch prediction module that dynamically activates the most appropriate prediction pathway based on the input degradation state, ensuring robustness in complex environments. Additionally, a weighted Mean Squared Error (MSE) loss function is introduced to prioritize prediction accuracy near the critical end-of-life phase. Extensive experiments on stratospheric airship and C-MAPSS datasets confirm that ADB-MSF yields superior predictive performance compared to existing state-of-the-art baselines.