Time series data intelligent fault diagnosis method based on deep learning

Abstract

The invention discloses a time series data intelligent fault diagnosis method based on deep learning. The method comprises the following steps: 1) collecting one-dimensional time series health state data of electromechanical equipment, wherein fault data is the fault data arranged at a preset measuring point of the electromechanical equipment; 2) constructing a one-dimensional time series originalfault data set according to the acquired N health state data; 3) preprocessing the data, wherein the preprocessing includes normalization and data truncation; 4) constructing a two-dimensional feature map fault data set; 5) carrying out data set division; 6) constructing a deep learning fault diagnosis model; 7) training the deep learning fault diagnosis model to obtain parameters; and 8) performing fault diagnosis on the input to-be-diagnosed sample data by using the deep learning fault diagnosis model, and outputting a final fault diagnosis result. By constructing the two-dimensional feature map and improving a traditional convolutional neural network model structure, a fault diagnosis speed and fault diagnosis accuracy are improved.

Description

technical field [0001] The present invention relates to fault diagnosis technology, in particular to an intelligent fault diagnosis method for time series data based on deep learning. Background technique [0002] With the advent of the intelligent era, more and more electromechanical equipment products are developing in the direction of intelligence, automation, multi-function and precision. Today, the complexity of the application environment of electromechanical equipment products is gradually increasing. Under such circumstances, the long-term continuous operation of electromechanical equipment is prone to various faults. If the fault cannot be diagnosed and eliminated in time, once the fault hazard spreads, it may cause heavy losses. Therefore, it is extremely necessary to provide effective intelligent fault diagnosis methods for electromechanical equipment; Nowadays, with the wide application of "Internet +", Internet of Things and advanced smart sensor technology in e...

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, Chinese invention patent (a method for intelligent diagnosis of rotating machinery fault characteristics based on deep convolutional neural network structure, application number: CN201810240234.1) and Chinese invention patent (based on vibration spectrum Diagram and deep convolutional neural network bearing fault diagnosis method, application number: CN201811567134.6) Although convolutional neural network is used, there are still three major defects. First, this method still needs to use traditional The feature extraction method (short-time Fourier transform) for pre-processing of feature extraction fails to make full use of the powerful feature extraction capabilities of convolutional neural networks, which limits the further improvement of fault diagnosis results; the second is the traditional convolutional neural network parameter The traditional convolutional neural network contains a 2-3 layer fully connected layer network structure, which is usually located between the last pooling layer and the Softmax classification output layer. Due to the existence of the fully connected layer The generated training parameters account for 80% to 90% of the total parameters of CNN. This defect greatly offsets the advantages of CNN in reducing the number of parameters through pooling dimensionality reduction. The structure of the fully connected layer not only occupies too much computing resources, At the same time, it is easy to cause over-fitting of CNN model training, especially the fully connected layer containing multiple hidden layers. Too many make the test time-consuming when used in fault online diagnosis, which is not conducive to real-time and rapid fault diagnosis

Third, the Softmax classification function used in traditional CNN is far less powerful than SVM in terms of multi-classification functions. However, the disadvantage of SVM is that its deep feature extraction and data mining capabilities are insufficient, and it is difficult to play a superior role in intelligent fault diagnosis. performance

Images