A COMPARATIVE STUDY OF DIVERSE AUTOENCODER MODELS IN LOCAL GEAR PITTING FAULT DIAGNOSIS

Gearbox, which is one of the most important and frequently used components among mechanical power transmission systems, has often been observed to occur in gear surface pitting faults in industrial applications that require high torque. For the diagnosis of gear pitting faults, vibration analysis is one of the commonly utilized techniques. Recently, there has been an increasing interest in applying deep learning approaches for classification and learning feature representations. Deep learning provides an excellent opportunity to integrate vibration signals for gear pitting fault diagnosis. Therefore, in this study, autoencoder models Contractive Autoencoder (CAE), Sparse Autoencoder (SAE) and Variational Autoencoder (VAE) are used to extract deep feature representations of gear pitting data. Without using any additional feature extraction techniques, in this study uses the raw vibrational data directly to identify the local gear pitting faults. Experimental results have shown that Sparse Autoencoder is a viable and efficient feature extraction method and provides a new research method for gear pit fault diagnosis.