عیب یابی جعبه دنده براساس سیگنال صوتی با استفاده از شبکه عصبی پیچشی و تبدیل کاکلیاگرام

Non-contact sound signal-based fault diagnosis is more practical compared to vibration-based fault diagnosis using contact sensors, especially in conditions where attaching sensors to the equipment is not feasible. One of the fundamental challenges in utilizing acoustic signals is their relatively high susceptibility to environmental noise. In this paper, two proposed methods are presented to reduce the impact of environmental noise in the process of gearbox fault diagnosis, using sound signals for four different faults and varying severity levels. In the first method, Variational Mode Decomposition (VMD) is used to eliminate noise and irrelevant data from the sound signal, and then the signal is transferred from the time domain to the time-frequency domain using Wavelet Transform (WT). In the second method, the sound signals containing noise are transferred from the time domain to the time-frequency domain using Cochleagram transformation. The generated data in the time-frequency domain from both methods are fed into a Convolutional Neural Network (CNN) for feature extraction and fault classification. The average fault diagnosis accuracy using Cochleagram is 98%, while the average accuracy based on VMD and WT is 90%.