A Slow Trend Analysis Method for Health Indicator Extraction and Remaining Useful Life Prediction of Aero-Engines

In recent years, Remaining Useful Life (RUL) prediction has become essential for ensuring operational safety and optimizing maintenance decisions. Existing research has shown that extracting a one-dimensional Health Indicator (HI) from multidimensional measurements is effective for RUL prediction through similarity analysis. However, these HI construction approaches often fail to adequately consider the slow-varying and monotonic nature, which are crucial for reflecting the gradual and irreversible decline in equipment performance. To address this issue, we propose a Slow Trend Analysis (STA) method to extract HIs that better characterize the degradation trends of equipment and apply them to RUL prediction. The STA derives a novel objective function to extract a common degradation trend from multidimensional sensor readings, aiming to accurately describe the slow-varying degradation nature and align with the monotonic trend of equipment performance decline. As an HI extraction technique, STA can be seamlessly integrated into similarity-based RUL prediction frameworks. Experimental results on the CMAPSS FD001 dataset, provided by NASA for aero-engine RUL assessment, demonstrate that our proposed method achieves superior predictive performance.