Vibration-based crack diagnosis for asymmetric gears under time-varying operational conditions using a 1-D CNN-LSTM model

Kalay O. C., KARPAT F.

Mechanism and Machine Theory, cilt.220, 2026 (SCI-Expanded, Scopus) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 220
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.mechmachtheory.2025.106347
  • Dergi Adı: Mechanism and Machine Theory
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, zbMATH
  • Anahtar Kelimeler: Asymmetric gears, Convolutional neural network, Long short-term memory, Time-varying operating condition, Tooth root crack, Vibration signal
  • Bursa Uludağ Üniversitesi Adresli: Evet

Özet

Gearboxes oftentimes operate under time-varying operating conditions (TVOC). Some studies on TVOC assessed the fluctuations in vibration data through computer simulations, while others appraised healthy gearboxes or piecewise constant operating conditions. Besides, these studies all focus on gears with symmetric tooth profiles. From this standpoint, the present experimental study combined a one-dimensional convolutional neural network (1-D CNN) with a long short-term memory (LSTM) algorithm for diagnosing divergent crack degrees that vary from 0% to 100% with an increment of 25% under TVOC for spur gear pairs with symmetric (20°/20°) and asymmetric (20°/30°) teeth. A series of vibration experiments was performed, considering two TVOC scenarios: (1) variable speed and constant load, and (2) constant speed and variable load. Using an asymmetric profile amplified the amplitude difference between the vibration response of healthy and cracked gears, facilitating fault diagnosis. For different TVOC scenarios, overall accuracies calculated for symmetric gears ranged between 90.005% and 98.654% and between 93.932% and 99.908% for asymmetric ones. The results revealed that the overall classification accuracy could be improved by up to 4.633% using gears with asymmetrical teeth.