As the key component of power transmission of mechanical equipment, gearbox is widely used in wind power, automobile, aviation and other mechanical equipment. However, the bad working environment makes the gear, rolling bearing and other components in the gearbox prone to failure, which affects the overall safety and reliability of the mechanical system.
In view of the importance and complexity of gearbox bearing, scholars at home and abroad have carried out a lot of research on bearing fault diagnosis. However, in the process of using these classical methods, it is usually necessary to extract and analyze the signal features artificially. For multi classification problems such as complex equipment structure, only relying on traditional pattern recognition methods can not accurately represent the mapping relationship between data.
For these reasons, scholars at home and abroad have proposed a fault diagnosis method based on deep learning. Jia et al. Pointed out that big data and deep learning model are very promising development directions in the field of fault diagnosis, and used stacked self encoder to diagnose the motor bearing data of Western Reserve University (CWRU). This method not only can adaptively mine the available fault features from the measured signals, but also has higher diagnosis accuracy than traditional methods. Singh et al. Decomposed the original signal by ensemble empirical mode decomposition (EEMD), and used the combined mode function (CMF) algorithm to select the appropriate intrinsic mode function (IMF) input convolution neural network to accurately identify the bearing rolling element compound fault. Li Huan proposed a fault diagnosis method based on short-time Fourier transform and convolution neural network for bearing signals in non-stationary and noise environment. The results show that it has higher recognition accuracy than taking time series and time chart as input samples.
It is worth pointing out that most of the current methods are aimed at bearing faults without strong interference. In engineering practice, the initial signal characteristics of bearing fault are often weak, which are easy to be affected by strong interference signals such as gears, and are easy to be modulated, showing typical nonlinear and non-stationary characteristics. Although convolutional neural network has achieved good results in the field of fault recognition, it still has some problems such as feature redundancy. Therefore, if the bearing signal component can be separated from the gearbox vibration signal, the analysis and diagnosis accuracy of the gearbox bearing can be improved.
In order to solve the problem of weak feature of initial fault signal of gearbox bearing, which is interfered by gear signal and difficult to express by shallow feature and feature redundancy of convolution neural network (CNN), the signal separation ability of Sanc and the strong feature extraction and learning ability of CNN are combined. This paper presents a fault diagnosis method for gearbox bearing based on Sanc and 1d-cnn. Firstly, the original signal of the gearbox is filtered adaptively by Sanc to suppress the redundant strong interference components which are independent of the bearing features. Then, the fault features of the gearbox bearing are extracted and identified by combining the powerful feature automatic extraction and fault identification ability of 1d-cnn. Compared with the traditional machine learning algorithm support vector machine (SVM), recurrent neural network (RNN) and one-dimensional convolutional neural network (1d-cnn) without Sanc, the proposed method has higher recognition accuracy and anti-interference.