Robust Bearing Fault Diagnosis Under Noisy Conditions Using Envelope Analysis and Support Vector Machines
III. International Conference on Engineering Sciences (ICES2026), Konya, Türkiye, 4 - 06 Eylül 2026, ss.597-609, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Konya
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.597-609
- Bursa Uludağ Üniversitesi Adresli: Evet
Özet
The continuous and safe operation of rotating machinery is critical to the efficiency and reliability of industrial production processes. Bearing faults are among the most common mechanical failures that can lead to unscheduled downtime and substantial economic losses when not detected in a timely and accurate manner. Although vibration signals are widely used in fault diagnosis, severe background noise in industrial environments poses significant challenges to identifying fault-specific characteristic frequency components. While machine learning algorithms offer effective solutions for automated diagnosis, the classification performance of the models fundamentally depends on the quality and robustness of the applied signal preprocessing techniques. In this study, the direct effects of different preprocessing techniques on the classification performance of a support vector machine (SVM) model for bearing fault diagnosis are examined. Fast Fourier transform (FFT), short-time Fourier transform (STFT), cepstrum analysis, and envelope analysis were applied together to process vibration data across the frequency domain. To simulate harsh industrial conditions and evaluate model robustness, different levels of artificial noise were synthetically added to the raw vibration signals. The SVM model was trained and tested separately using feature sets with and without envelope analysis under different noise levels. The obtained model classification accuracy metrics were comparatively analyzed. The findings demonstrate that Envelope-analysis-supported preprocessing significantly enhances model accuracy, especially in high-noise scenarios, by accentuating masked weak periodic fault impulses. Additionally, the multidimensional feature representations provided by STFT, cepstrum, and FFT analyses further enhance the model’s generalization capability and classification performance. This research clearly demonstrates the critical role of envelope analysis in improving the fault-diagnostic accuracy and noise tolerance of machine learning models in bearing condition-monitoring systems.