Rotary machine fault diagnosis method based on improved convolutional neural network

Abstract

The invention discloses a rotary machine fault diagnosis method based on an improved convolutional neural network, and the method comprises the following steps: (1) collecting one-dimensional time sequence fault data, and obtaining an original fault data set; (2) performing preprocessing operation on the acquired data of the original fault data set, wherein the preprocessing comprises standardization, data truncation and data reconstruction; (3) dividing the preprocessed samples of each type of faults into a training set, a verification set and a test set; (4) an improved convolutional neuralnetwork fault diagnosis model is established, the model comprises an input layer, a feature extraction layer, a dimension reduction and parameter reduction layer and a softmax classification output layer, and the dimension reduction and parameter reduction layer comprises a 1 * 1 transition convolution layer and a global mean pooling layer; (5) training and testing the model; wherein the diagnosismodel can automatically extract features and diagnose the fault data without any manual feature extraction operation, so that people can diagnose the fault of the rotating machine more conveniently and quickly.

Description

technical field [0001] The invention belongs to the technical field of fault diagnosis and detection of rotating machinery, and more specifically relates to a fault diagnosis method for rotating machinery based on an improved convolutional neural network. Background technique [0002] With the rapid development of modern industrial technology, rotating machinery equipment is increasingly developing towards high speed, precision, automation and integration. Rotating machinery mainly includes power devices, such as diesel engines, steam turbines, engines, motors, etc., and also includes rotating parts , such as bearings, spindles, etc. With the diversification of the working environment of rotating machinery, especially when it operates continuously for a long time in a complex and changeable working environment, it is often prone to various failures due to its heavy workload, variable load, and the influence of salt-alkali corrosion and high temperature. . If the fault cann...

AI Technical Summary

Problems solved by technology

Although the above-mentioned existing intelligent fault diagnosis methods have been applied and achieved certain results, there are still three major shortcomings: (1) It is necessary to master various advanced signal processing techniques for feature extraction, and feature selection must rely on engineers with experience and Completion of professional knowledge has great subjectivity and blindness; (2) Feature extraction is mainly used to solve specific fault problems, which has poor versatility and is difficult to complete in the environment of massive data samples; (3) Manually extracted fault features are not comprehensive, Features that reflect minor faults are easily deleted by mistake and covered by noise

In recent years, some scholars have applied CNN to the field of fault diagnosis. A Chinese invention patent (a method for intelligent diagnosis of fault characteristics of rotating machinery based on deep convolutional neural network structure, application number: CN201810240234.1) discloses a fault diagnosis method for rotating machinery. Although the convolutional neural network is used in the method, there are still two major defects. First, the method still needs to use the traditional feature extraction method (short-time Fourier transform) for the pre-processing of the original fault data. Making full use of the powerful feature extraction capabilities of convolutional neural networks limits the further improvement of fault diagnosis results; the second is the problem of too many parameters in traditional convolutional neural networks. Traditional convolutional neural networks contain a 2~3 The network structure part of the fully connected layer of the layer is usually located between the last pooling layer and the Softmax classification output layer. Due to the existence of the fully connected layer, the training parameters generated by the existence of the fully connected layer account for 80% to 90% of the total parameters of CNN. One defect greatly offsets the advantages of CNN to reduce the number of parameters through pooling dimensionality reduction. The structure of the fully connected layer not only occupies too many computing resources, but also easily causes CNN model training to overfit, especially if it contains multiple hidden layers. The number of fully connected layers of the CNN model will increase exponentially with the increase of the number of fully connected layers, resulting in the excessive amount of parameters of the traditional CNN model, which makes the test time-consuming when used in online fault diagnosis, which is not conducive to faults real-time rapid diagnosis

Images