Multi-feature extraction and fusion intelligent fault diagnosis method

Abstract

The invention discloses a multi-feature extraction and fusion intelligent fault diagnosis method. The method comprises the following steps: firstly, acquiring data in the operation process of mechanical equipment by using a data acquisition system and a sensor; secondly, intercepting an original signal which is not processed according to the determined length, dividing the original signal into a training sample and a test sample, and performing frequency spectrum normalization on each section of signal; constructing a multi-feature extractor based on an auto-encoder, constructing a multi-feature fusion device based on a dynamic routing algorithm, and constructing a health state classifier based on softmax; then, using a training sample training model for extracting effective features for distinguishing various health states and a self-adaptive learning feature fusion method; and finally inputting the test sample into the model, and verifying the effectiveness of the model. According tothe invention, self-adaptive extraction of equipment fault characteristics and intelligent diagnosis of fault states under the condition of small samples are realized, the training time is short, andthe result is accurate and reliable.

Description

technical field [0001] The invention belongs to the technical field of fault diagnosis, in particular to an intelligent fault diagnosis method for multi-feature extraction and fusion. Background technique [0002] With the advancement of technology, modern machinery is developing towards high speed, high precision and intelligence, but it must meet the requirements of reliability and availability; as an important part of mechanical equipment, key components are important to ensure the smooth operation of equipment It is very important, once it breaks down, the reliability of the equipment will be reduced, and it will cause unexpected shutdown of the equipment, resulting in huge production costs and loss of production capacity; therefore, based on the operating data of key components, an effective A troubleshooting approach to identify health states is necessary. [0003] In the field of fault diagnosis, the identification of the health status of components is mainly divided...

AI Technical Summary

Problems solved by technology

[0004] After decades of development, intelligent fault diagnosis has gone through two stages, namely traditional intelligent fault diagnosis and intelligent fault diagnosis based on deep learning; traditional intelligent fault diagnosis extracts scalar features in various fields, such as time domain, frequency Domain, time-frequency domain, and then select some sensitive features or all features to input into the shallow model, such as support vector machine, artificial neural network or a mixture of various methods, etc., through these methods to realize the identification of equipment health status; however, traditional intelligence The diagnosis method has the following two shortcomings: 1) The relationship between feature extraction and fault identification is isolated, which increases the difficulty of intelligent fault diagnosis; 2) When faced with more complex fault identification tasks, such as many fault categories and large data volumes, etc. , Artificially extracting effective features will consume a lot of time, and the feature generalization ability is low, which narrows the application range of intelligent diagnosis; due to the powerful data mining and adaptive feature extraction capabilities, deep learning has shown the ability to overcome the inherent defects of traditional intelligent diagnosis The potential has greatly promoted the development and application of intelligent fault diagnosis; researchers have constructed a deep learning by stacking multiple layers of basic neural networks, such as restricted Boltzmann machines, autoencoders, or their variants. model, which enables it to adaptively learn effective scalar features from time-domain data, frequency-domain data, and time-frequency domain data; the classifier of the final model uses these features to identify the health status of the bearing

[0005] However, the intelligent diagnosis method based on deep learning still has the following two key points that need to be solved urgently: 1) training deep learning models usually requires a large number of training books, but in practice there are few fault samples, which cannot meet this requirement; 2) training depth Learning the model requires a lot of time, which causes the model to lack the ability to update quickly; the two problems are sometimes contradictory; therefore, solving both problems at the same time requires a specially designed smart diagnostic model

Images