Gear fault diagnosis research based on GAF-TFR-2DCV with small sample size

Abstract

To solve the problem of insufficient precision of gear fault diagnosis with small sample size, an adaptive floating convolutional sequence pattern neural network recognition method based on the combination of simplified Gramian Angle field and time–frequency feature images is proposed. One-dimensional data is converted to two-dimensional real-time data by simplifying Gramian Angle field. The time domain data and time frequency data are mixed enhanced by fast Fourier transform to obtain feature enhanced image-like data. The feature-enhanced image dataset generated by this method has a high degree of intra-group consistency in data features, which is beneficial to fault identification. The adaptive floating convolutional sequence model is used to construct neural network to identify image-like data and improve the accuracy of gear fault diagnosis with small samples. Moreover, the diagnosis method has strong robustness and high engineering application value.

Keywords

• Gear fault diagnosis
• Gramian angle field
• Time-frequency diagram
• Convolutional Neural Networks
• Vision algorithms