주파수 인지형 대칭 점 패턴 기반 베어링 결함 진단 기법

초록

Bearing fault diagnosis is essential for ensuring the reliability and safety of rotating machinery systems. Among visualization-based approaches, the symmetrized dot pattern (SDP) method has attracted attention because it converts one-dimensional vibration signals into intuitive two-dimensional geometric patterns. However, conventional SDP methods rely mainly on raw time-domain signals and therefore have limited capability to reflect fault-related frequency characteristics, particularly under low-speed and low signal-to-noise-ratio conditions. To overcome these limitations, this paper proposes a frequency-aware SDP framework that enhances fault-sensitive features prior to SDP transformation. The proposed framework consists of adaptive bandpass filtering, envelope demodulation-optimized low-pass filtering, robust normalization, and SDP image generation. A sparsity-based frequency- band-selection strategy is employed to automatically identify fault-sensitive resonance regions, and the low-pass cutoff frequency is optimized by jointly considering the signal-to-noise ratio of fault-related frequency components and signal smoothness. The proposed method was validated using the Case Western Reserve University bearing dataset and experimentally acquired low-speed datasets under 20r/min and 5r/min conditions. Experimental results demonstrate that the proposed preprocessing significantly improves SDP pattern separability and classification performance compared with conventional SDP approaches, especially under low-speed operating conditions where fault features are weak. These results confirm that incorporating frequency-domain characteristics into SDP representation can substantially enhance the robustness and discriminative capability of SDP-based bearing fault diagnosis.

키워드