A New Convolutional Neural Network Model and Its Application

Abstract

The invention discloses a new convolutional neural network model and its application. The model is based on the convolutional neural network, uses an improved threshold function as the activation function tReLU, and introduces residual neurons into the middle layer of the convolutional neural network. , use convolution and pooling to alternately connect, softmax classification, and generate a novel convolutional neural network model RLCNN throughout the entire structure; the activation function tReLU is shown in formula (1). The method of using this model for bearing fault diagnosis is as follows: firstly, the vibration signal from the bearing is converted into a two-dimensional vibration image, and the image is processed to obtain the pixel intensity matrix of the gray image, and the pixel intensity matrix is ​​input into the new convolutional neural network model, Take the existing public bearing data set as the training set, perform convolution and batch normalization processing; obtain the diagnosis result of the bearing fault. The invention solves the problem of low accuracy of bearing fault diagnosis caused by model gradient disappearance and mean value shift caused by traditional activation function.

Description

Technical field [0001] The present invention belongs to the technical field of computer convolutional neural network, and specific relates to a new type of convolutional neural network model and application. Background technique [0002] Modern machinery equipment is being carried out under complex and high-intensity work conditions. Once failed, it may be catastrophic, so mechanical failure hidden huge risks and economic losses. Bearings are important components of mechanical equipment, in order to make some equipment can function properly, the diagnosis of bearings is essential, and 30% is statistically caused by the damage of the bearing. Now the machinery equipment fault data is quite the "big data" era. [0003] How to automatically conduct feature mining in massive data, in order to replace artificial extraction, real-time detection of the bearing to ensure the accuracy and efficiency of fault diagnosis. Therefore, the intelligent diagnosis of the processing industry "big d...

AI Technical Summary

Problems solved by technology

[0004] In view of the deficiencies in the prior art, the present invention solves the problem of low diagnostic accuracy caused by gradient disappearance by improving the activation function linear correction unit; two layers of residual neuron layers are added to the convolutional neural network to deepen Network depth to facilitate the extraction of potential features that are not easy to be explored; thereby constructing a new convolutional neural network model and applying it to bearing fault diagnosis

Images