Akademik Veri Yönetim Sistemi
Rize Trade and Economy Summit and Congress, Rize, Türkiye, 22 - 23 Ekim 2025, (Tam Metin Bildiri)
Rolling element bearings are among the most critical components in rotating machinery, as their operational integrity directly affects system reliability, safety, and maintenance costs. Unexpected bearing failures may result in costly downtime, efficiency losses, and potential safety hazards. Therefore, accurate and timely fault detection is of great importance not only from a technical perspective but also from an economic and financial standpoint. Traditional diagnostic techniques, which primarily rely on time- and frequency-domain analysis, often fail to capture the transient and non-stationary nature of vibration signals. To overcome these limitations, this study introduces a deep learning-based framework that integrates advanced signal processing with state-of-the-art neural network architectures. In the proposed methodology, raw vibration signals from the Machinery Failure Prevention Technology (MFPT) dataset were transformed into two-dimensional scalogram images using the continuous wavelet transform (CWT). This conversion allows the simultaneous representation of transient and spectral features across multiple scales, thereby enhancing data discriminability. The scalograms were then used to train a customized ResNet-152 deep residual network designed for a three-class classification task: normal operating condition (NOC), outer fault condition (OFC), and inner fault condition (IFC). Experimental results demonstrated strong classification performance, achieving 90% accuracy on the training dataset and 91% on the independent test dataset. OFC achieved the highest recognition rates, while IFC exhibited relatively lower accuracy due to overlapping feature distributions. Nevertheless, high precision, recall, and specificity across all classes confirmed the robustness of the proposed framework. Beyond its technical contributions, this study also emphasizes the economic and financial benefits of predictive maintenance. By enabling fault detection prior to catastrophic failure, predictive strategies reduce unplanned downtime, optimize resource utilization, and lower maintenance costs. Thus, the proposed method offers not only a scalable and automated solution for fault diagnosis in rotating machinery but also a financially sustainable approach supporting industrial competitiveness.