Bearings Fault Diagnosis Based on Improved Deep Belief Network by Self-individual Adaptive Learning Rate

Abstract

The health monitoring of key component of mechanical equipment such as bearings is entering the era of big data and intelligence. The traditional method of bearing fault diagnosis depends on manual extraction, which relies on complex signal processing methods and abundant professional experience. As a new machine learning method which can learn deep features from data, the deep learning method is introduced into the field of mechanical fault diagnosis, and the operation efficiency and fault recognition accuracy of mechanical fault diagnosis are improved, which will further improve the practicability of deep learning method in fault diagnosis field. A novel self-adaptive deep belief network with Nesterov momentum (NM-based ADDBN) is proposed. Nesterov momentum is used to replace the traditional momentum method to predict the next position of the parameters, and control parameters' speed towards the optimal values, avoiding missing the optimal values caused by the traditional momentum method. Self-individual adaptive learning rate is used to adaptively select decrease step length for the gradient update, and to speed up the model training and improve the generalization ability of the model. Experimental results show that compared with support vector machine and standard depth belief network, the proposed method obtain the best precision performance for bearing fault recognition under different load conditions. For operational efficiency, the optimization model can effectively and steadily speed up the model training, improve generalization ability of deep belief network, and realize the rolling bearing fault diagnosis effectively compared with current optimization methods.