Classification of Healthy and Cracked Gear Conditions in a Two-Stage Gearbox Using Machine Learning

İşci M., Yıldırım B.

ZEUGMA 16TH INTERNATIONAL CONGRESS ON SCIENTIFIC RESEARCH, Gaziantep, Türkiye, 26 - 28 Haziran 2026, ss.1, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Gaziantep
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.1
  • Kayseri Üniversitesi Adresli: Evet

Özet

This study addresses the classification of the structural condition of gearboxes, a critical component in rotating machinery, with high accuracy using machine learning methods. The analysis used the “Gearbox Fault Diagnosis Data” set from the Open Energy Data Initiative (OEDI) data center to examine healthy and cracked gear conditions in a two-stage system. Data were obtained on the SpectraQuest simulator using accelerometers positioned in four directions across a load spectrum ranging from 0% to 90%. As a method, raw vibration signals were divided into windows of 128 samples; a statistical feature set consisting of mean, standard deviation, maximum, minimum, RMS, and peak-to-peak values reflecting the signal's characteristics was derived from these segments. In the classification phase, 25% of the data was reserved for testing, and K-Nearest Neighbors, Decision Tree, Support Vector Machine, and Random Forest models were trained. In the performance evaluation, the healthy gear condition was defined as the positive class; accuracy, precision, recall, and F1-score were used as the evaluation metrics. The experimental results showed that the SVM model achieved the highest success rate. This model exhibited the most stable performance with 96.50% accuracy, 96.49% precision, 96.68% recall, and 96.58% F1-score. The Random Forest algorithm produced the second-best result with 95.97% accuracy and 96.00% F1-score; KNN followed with 95.19% accuracy, and Decision Tree with 93.14% accuracy. The findings demonstrate that features derived from multi-channel sensor data are decisive for fault diagnosis and that the SVM-based approach, in particular, provides a reliable basis for predictive maintenance strategies.