Adaptive Segment-wise Filtering of Acceleration Data Based on Automatic Motion Stage Detection Adaptive Segment-wise Acceleration Data Filtering

This study presents an adaptive segment-wise filtering framework for non-stationary acceleration signals in impact-related motions. Motion phases are automatically identified using acceleration magnitude and its temporal variation, enabling stage-specific filtering through moving-average, Savitzky–Golay, and median methods. The approach effectively suppresses noise in static and micro-vibration stages, preserves low-frequency trends during acceleration, and maintains peak fidelity in impact segments. Experiments on long-duration multi-sensor data demonstrate substantial RMS reductions, high raw–filtered correlation, and minimal peak distortion. The proposed method enhances signal stability and integrity, providing a reliable preprocessing tool for downstream motion analysis and impact event detection.