CN116467652A - Bearing fault prediction method and system based on convolution noise reduction self-encoder - Google Patents

Bearing fault prediction method and system based on convolution noise reduction self-encoder Download PDF

Info

Publication number
CN116467652A
CN116467652A CN202310315205.8A CN202310315205A CN116467652A CN 116467652 A CN116467652 A CN 116467652A CN 202310315205 A CN202310315205 A CN 202310315205A CN 116467652 A CN116467652 A CN 116467652A
Authority
CN
China
Prior art keywords
encoder
noise reduction
bearing
reduction self
vibration signal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310315205.8A
Other languages
Chinese (zh)
Inventor
魏星宇
肖罡
张蔚
赵斯杰
李端玲
万可谦
魏志宇
刘小兰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangxi Kejun Industrial Co ltd
Original Assignee
Jiangxi Kejun Industrial Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangxi Kejun Industrial Co ltd filed Critical Jiangxi Kejun Industrial Co ltd
Priority to CN202310315205.8A priority Critical patent/CN116467652A/en
Publication of CN116467652A publication Critical patent/CN116467652A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2123/00—Data types
    • G06F2123/02—Data types in the time domain, e.g. time-series data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a bearing fault prediction method and a system based on a convolution noise reduction self-encoder, wherein the method comprises the steps of collecting a vibration signal sequence of a rolling bearing, inputting a bearing fault diagnosis model, extracting vibration signal characteristics through the convolution noise reduction self-encoder comprising a gate control circulating unit GRU, and classifying by a classifier to obtain a fault detection result of the rolling bearing; the convolutional noise reduction self-encoder comprising the gate control circulation unit GRU is an encoding-decoding structure formed by connecting an encoding unit and a decoding unit, wherein the encoding unit comprises a noise adding layer, a first GRU network, a convolutional layer and a pooling layer which are connected in sequence, and the decoding unit comprises a deconvolution layer, a pooling layer and a second GRU network which are connected. The method can adaptively extract the time sequence characteristics in the bearing time sequence vibration signals, enhance the characteristic extraction capability, reduce the noise influence, and input hidden layer characteristics into the classifier for recognition so as to realize accurate and efficient bearing fault prediction and diagnosis.

Description

Bearing fault prediction method and system based on convolution noise reduction self-encoder
Technical Field
The invention belongs to the technical field of fault prediction in the operation process of industrial equipment, and particularly relates to a bearing fault prediction method and system based on a convolution noise reduction self-encoder.
Background
Rolling bearings are very important in the modern manufacturing industry, are called as industrial joints, with the advent of the age of 4.0, rotating machinery has become one of the most common categories of large-scale mechanical equipment, and because the rotating mechanical equipment has long working time and high strength, the working environment of the rolling bearings in the equipment is severe, the conditions of abrasion, fatigue and the like caused by insufficient lubrication appear, adverse effects are generated on the mechanical equipment, and potential safety hazards are caused to industrial operation. Especially, in the trend of gradually moving to automation of the current mechanical equipment, the failure of the rolling bearing can not only lead to the damage of the mechanical equipment, but also cause economic loss, and more likely to cause disastrous results. Therefore, the fault diagnosis technology of the rolling bearing has great significance for the safe operation of mechanical equipment.
The time series (or dynamic series) is a series of values of the same statistical index arranged according to the time sequence of occurrence. Vibration signals of bearings in large industrial equipment vary greatly in the case of bearings at different life time periods, and are typical time series data. Time sequence analysis is a time domain method for identifying modal parameters by processing ordered random sampling data by using a parameter model. At present, fault prediction techniques are various, and with the rise of artificial intelligence, a plurality of new techniques are emerging. In general, failure prediction techniques based on physical models, failure prediction techniques based on data driving, and failure prediction techniques based on statistical reliability can be broadly classified. The prediction model based on statistical reliability mainly searches for a fault distribution rule of a target object, and has two branches of a life distribution model and a fault tree analysis model. The prediction method based on the physical model mainly identifies failure mechanism, failure position time and the like of equipment or a system through a series of mathematical models, further analyzes and predicts the feasibility of the equipment or the system, and most commonly used methods include Paris rule and Forman rule. The fault prediction technology based on data driving mainly relies on advanced sensor technology to acquire and acquire characteristic parameters related to system attributes, correlate the characteristic parameters with useful information, detect, analyze and predict by means of intelligent algorithms and models, and further provide decision information for maintenance management tasks and system guarantee, such as a machine learning model (autoregressive moving average model, support vector machine, random forest and the like), a deep learning model (layered sparse coding, deep self-coding, deep belief network, cyclic neural network, convolutional neural network and the like), a Kalman filtering model and the like.
Under the development of computers and information technologies, the industry has entered a big data era, traditional fault diagnosis methods include wavelet transformation, modal decomposition and the like, high-dimensional feature sets can be constructed, and then classification models such as error back propagation algorithms (Error Back Propagation Training, BP), support vector machines (Support Vector Machine, SVM) and the like are trained. Although the traditional artificial neural network has achieved a certain result in the bearing fault diagnosis field, the traditional artificial neural network is limited by a shallow structure of the traditional artificial neural network and has a certain limitation. In recent years, deep learning has been developed rapidly, and particularly, an Auto-Encoder (AE) has outstanding performance in terms of data dimension reduction, and has a remarkable effect in terms of rolling bearing fault diagnosis due to its simple structure. The basic self-encoder model is too simple, is easy to be interfered by various noise and is over-fitted, and the variant of the model is generally used in practice. For example, in order to improve the feature extraction capability of the self-encoder model, a stacked sparse self-encoder structure is used, so that the feature extraction capability is improved to a certain extent. The actual bearing vibration signal generally contains noise, and the structure of the stacked noise reduction self-encoder can be used, so that the influence of the noise can be effectively reduced, and the diagnosis accuracy is improved. The stacked self-encoder may be combined with a wavelet transform approach to noise reduction of the signal by the framework of digital wavelet transform and then extracting features in the signal from the stacked self-encoder. The method of the sparse self-encoder and the nuclear extreme learning machine can be combined, and the high-dimensional data characteristic dimension reduction capability of the nuclear extreme learning machine can be effectively improved. Since the fully connected network in the self-encoder structure cannot effectively grasp the time sequence characteristics in the signal, a method for constructing the self-encoder structure by using the LSTM network is proposed for generating a video sequence, but in the aspect of fault diagnosis of a bearing, the method can extract the time sequence characteristics of the bearing signal, but has limited characteristic extraction capability, is easy to receive noise interference, and has relatively low diagnosis accuracy. And the phenomenon of over fitting is easy to occur under the condition of less sample size; slower training speed, etc.
Disclosure of Invention
The invention aims to solve the technical problems: aiming at the problems in the prior art, the invention provides a bearing fault prediction method and a system based on a convolution noise reduction self-encoder, which can adaptively extract the time sequence characteristics in the time sequence vibration signals of the bearing, enhance the characteristic extraction capability, reduce the noise influence, and input hidden layer characteristics into a classifier for recognition so as to realize accurate and efficient bearing fault prediction and diagnosis.
In order to solve the technical problems, the invention adopts the following technical scheme:
a bearing fault prediction method based on a convolution noise reduction self-encoder comprises the following steps:
s101, collecting a vibration signal sequence of a rolling bearing;
s102, inputting a vibration signal sequence into a trained bearing fault diagnosis model formed by a convolution noise reduction self-encoder containing a gate control circulating unit GRU and a classifier, extracting vibration signal characteristics through the convolution noise reduction self-encoder containing the gate control circulating unit GRU, and classifying by the classifier to obtain a fault detection result of the rolling bearing; the convolutional noise reduction self-encoder comprising the gate control circulation unit GRU is an encoding-decoding structure formed by connecting an encoding unit and a decoding unit, wherein the encoding unit comprises a noise adding layer, a first GRU network, a convolution layer and a pooling layer which are connected in sequence, and the decoding unit comprises a deconvolution layer, an anti-pooling layer and a second GRU network which are connected.
Optionally, the classifier is a Softmax classifier.
Optionally, step S102 includes training a bearing fault diagnosis model:
s201, collecting vibration signal sequences of a plurality of rolling bearings to obtain a sample set S; dividing a sample set S into a training set, a verification set and a test set; respectively carrying out standardized treatment on the training set, the verification set and the test set;
s202, constructing a convolutional noise reduction self-encoder containing a gate control circulating unit GRU of a bearing fault diagnosis model, respectively mixing noise to a training set, a verification set and a test set, and training the convolutional noise reduction self-encoder containing the gate control circulating unit GRU by using the training set, the verification set and the test set after mixing noise to obtain a trained convolutional noise reduction self-encoder containing the gate control circulating unit GRU;
s203, adding a classifier for a convolution noise reduction self-encoder comprising a gate control circulating unit GRU to construct a complete bearing fault diagnosis model;
s204, respectively collecting vibration signal sequence samples of a plurality of rolling bearings according to various fault detection result types of the rolling bearings to obtain a sample set S 1 The method comprises the steps of carrying out a first treatment on the surface of the Sample set S 1 Dividing into trainingA training set, a verification set and a test set; respectively carrying out standardized processing on the training set, the verification set and the test set, adding a label of a fault detection result type, and converting the label into One-Hot codes;
s205, using sample set S 1 Training the classifier by the training set, the verification set and the test set, thereby obtaining a trained bearing fault diagnosis model.
Optionally, in step S202, a loss function is used to train the convolutional noise reduction self-encoder including the gate-control loop unit GRU, where:
in the above equation, MSE represents the mean square error as a function of loss, n is the number of samples, w i Is the weight, y i Is thatCorresponding true value, < >>For the predicted value of the convolutional noise reduction self-encoder comprising the gate-control loop unit GRU, the subscript i is the sequence number of the sample.
Optionally, in step S205, the loss function used in training the classifier is:
in the above formula, H (p||q) represents cross entropy as a loss function, p is a true probability distribution, q is a probability distribution predicted by a bearing fault diagnosis model, p (i) is a true probability distribution of an ith sample, q (i) is a probability distribution predicted by a bearing fault diagnosis model of the ith sample, and subscript i is a sequence number of the sample.
Optionally, the functional expression of the sample set S in step S201 is:
S＝{D 1 ，D 2 ，D 3 ，...，D n }，
in the above, D 1 ～D n Respectively n vibration signal sequence samples of the rolling bearings;
sample set S in step S204 1 The functional expression of (2) is:
S 1 ＝{D 1 ，D 2 ，D 3 ，...，D n }，
in the above, D 1 ～D n The vibration signal sequence samples of n rolling bearings are respectively:
D i ＝[x i ，x i+1 ，...，x i+w-1 ]，
in the above, D i For the samples of the vibration signal sequence of the ith rolling bearing, x i ～x i+w-1 W vibration signals in the vibration signal sequence are respectively.
Optionally, the functional expression of the normalization process is:
in the above, x scale To the result of the normalization processing of the input vibration signal x, x mean Std is the variance of the vibration signal x, which is the mean value of the vibration signal x.
Alternatively, mixing noise in step S202 refers to adding gaussian white noise of a variance and noise figure of a specified magnitude.
In addition, the invention also provides a bearing fault prediction system based on the convolution noise reduction self-encoder, which comprises a microprocessor and a memory which are connected with each other, wherein the microprocessor is programmed or configured to execute the bearing fault prediction method based on the convolution noise reduction self-encoder.
Furthermore, the present invention provides a computer readable storage medium having stored therein a computer program for programming or configuring by a microprocessor to perform the convolutional noise reduction self-encoder based bearing failure prediction method.
Compared with the prior art, the invention has the following advantages:
1. the bearing fault diagnosis model adopted by the invention combines the advantages of the self-encoder, such as strong feature extraction capability and the capability of the GRU network to process time sequence features, so that the neural network model can realize the fault prediction of the bearing and also can realize the fault diagnosis of the bearing.
2. The bearing fault diagnosis model adopted by the invention inputs the original vibration signal of the bearing added with noise, and the output result is also the prediction of the vibration signal, so that the model can adaptively reduce noise from the original vibration signal and extract the characteristics.
3. The bearing fault diagnosis model solves the defect that a full-connection network in a general self-encoder structure cannot process the time sequence characteristics of data, and a bearing vibration signal is time sequence data, so that the bearing fault diagnosis model can effectively extract the time sequence characteristics in the bearing vibration signal and predict the time sequence characteristics, and the characteristic extracted by the convolutional noise reduction self-encoder comprising a gate control circulating unit GRU is used as the input of a classifier, so that the fault classification of the bearing can be effectively carried out.
Drawings
FIG. 1 is a schematic diagram of a basic flow of a method according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a network structure of a convolutional noise reduction self-encoder including a gate-control loop unit GRU according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of a network structure of a GRU network according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a training flow of a bearing fault diagnosis model according to an embodiment of the present invention.
Fig. 5 is a waveform example of vibration signals before and after noise addition in the embodiment of the present invention.
FIG. 6 is a graph showing a loss function of a bearing failure diagnosis model according to an embodiment of the present invention.
Fig. 7 is a graph comparing the reconstructed vibration signal and the original vibration signal in the embodiment of the present invention.
FIG. 8 is an example of a classification confusion matrix obtained in an embodiment of the invention.
FIG. 9 is a schematic diagram showing the experimental results of universality in the embodiment of the present invention.
Detailed Description
Aiming at the defects of the prior art, the invention aims to solve the technical problems that the LSTM self-encoder model is optimized, the time sequence characteristics of the bearing signals are learned, the model characteristic extraction capability is enhanced, and the noise influence is reduced. As shown in fig. 1, the present embodiment provides a bearing failure prediction method based on a convolutional noise reduction self-encoder, which includes:
s101, collecting a vibration signal sequence of a rolling bearing;
s102, inputting a vibration signal sequence into a trained bearing fault diagnosis model formed by a convolution noise reduction self-encoder containing a gate control circulating unit GRU and a classifier, extracting vibration signal characteristics through the convolution noise reduction self-encoder containing the gate control circulating unit GRU, and classifying by the classifier to obtain a fault detection result of the rolling bearing; as shown in fig. 2, the present embodiment designates a fault diagnosis model as a GRU-CDAE (Gated Recurrent Units-Convolutional Denoising Auto-Encoder) fault diagnosis model, and the core of the fault diagnosis model is a convolutional noise reduction self-Encoder comprising a gate-control loop unit GRU, the convolutional noise reduction self-Encoder is an encoding-decoding structure formed by connecting an encoding unit and a decoding unit, the encoding unit is formed by a noise adding layer, a first GRU network, a convolution layer and a pooling layer which are connected in sequence, and the decoding unit is formed by a deconvolution layer, a pooling layer and a second GRU network which are connected.
In order to solve the disadvantages that the LSTM self-encoder has long training time and general model feature extraction effect, and is prone to the occurrence of the over-fitting phenomenon, the embodiment adopts a convolutional noise reduction self-encoder including a gate-control loop unit GRU as a model for feature extraction, as shown in fig. 2. The convolutional noise-reducing self-Encoder comprising the gate-control loop unit GRU combines the gate-control loop unit (Gated Recurrent Unit, GRU) and the convolutional noise-reducing self-Encoder (Convolution Denoising Auto-Encoder, CDAE), i.e. inputs a sequence x to generate an output sequence y, and also has better noise-reducing capability and feature extraction capability, which can map the sequence of bearing vibration signals into a fixed-length representation and decode it into a reconstructed sequence through several decoding layers. The convolution noise reduction self-encoder comprising the gate control circulation unit GRU can learn the characteristics of bearing vibration time sequence signals, can effectively avoid certain noise interference and overfitting, has certain improvement on the characteristic extraction capacity, encodes input time sequence data x by the encoding unit, decodes the encoded information by the decoding unit, outputs a section of reconstructed time sequence information y, has better noise reduction capacity and characteristic extraction capacity, can map the sequence of the bearing vibration signals into a fixed-length representation, and decodes the sequence into a reconstruction sequence by a plurality of decoding layers. It should be noted that the classifier may adopt a required classifier model according to needs, for example, as an alternative implementation, the classifier in this embodiment is a Softmax classifier. The embodiment is a deep learning method based on a GRU-CDAE (Gated Recurrent Units-Convolutional Denoising Auto-Encoder) self-Encoder and a Sofmax classifier, which can adaptively extract the characteristics of a bearing time sequence vibration signal, and input hidden layer characteristics into the Softmax classifier for identification, so that accurate bearing fault prediction and diagnosis can be provided.
The gated loop unit (Gated Recurrent Unit, GRU) is a variant of the LSTM network, the network structure of which is shown in fig. 3. The GRU network mainly improves two points on the basis of the LSTM network, namely, the unit state and output in the LSTM network are integrated into a unified hidden state h t By means of hidden state h t To communicate the required information; and combining the forgetting gate layer and the input gate layer in the LSTM network to unify the forgetting gate layer and the input gate layer into an updated gate layer. Compared with an LSTM network, the GRU network reduces the parameters by nearly one third by the two-point modification, so that the GRU network reduces the training time, accelerates the model iteration speed, is not easy to generate the phenomenon of over-fitting, and is favorable for constructing a model with large data volume. Update Gate (Update Gate), z in FIG. 3 t Part(s). This part is used to determine how much status information was brought into the status information at this point in time before. Reset Gate (Reset Gate), i.e. r in FIG. 3 t Part(s). Its function andforgetting gate layer similarity in LSTM networks for controlling how much unimportant information to select is forgotten and how much state information of previous moment is written to current candidate setThe larger the value of the update gate, the more state information is brought to the current at the previous time, and the smaller the value of the reset gate, the more information is forgotten, and the less state information is written to the current candidate set at the previous time. And has the following steps:
z t ＝σ(W z ·[h t-1 ，x t ])，
r t ＝σ(W r ·[h t-1 ，x t ])，
in the above, z t To update the gate output at time t, σ is the activation function, W z To update the weight matrix of the gate, h t-1 Is the hidden state at the moment t-1, x t Input at time t; r is (r) t To reset the output of the gate at time t, W r A weight matrix for the reset gate;for the candidate set at time t +.>As a weight matrix, h t Is the hidden state at time t.
The trained convolutional noise reduction self-encoder comprising the gated loop unit GRU can be used for supervised learning tasks, such as predicting the vibration signal of the following bearing. Such sequence-to-sequence learning may find application in a variety of situations, such as extracting features from time series signal data; and a reasonable threshold value can be set according to the characteristics of the signals, so that abnormality detection is performed, namely whether the signals are abnormal signals or not is determined according to whether the distance between the input signals and the reconstructed signals (the signals output by the convolutional noise reduction self-encoder containing the gate control loop unit GRU) is larger than the threshold value or not. As shown in fig. 4, step S102 in the present embodiment includes training a bearing failure diagnosis model:
s201, collecting vibration signal sequences of a plurality of rolling bearings to obtain a sample set S; dividing a sample set S into a training set, a verification set and a test set; and respectively carrying out standardization treatment on the training set, the verification set and the test set.
In this embodiment, a section of a long signal in a section of data set with a signal length w is intercepted backwards each time to obtain an acquired sample set S; taking 0.7 of the final standardized sample set S as a training set, 0.2 as a verification set and 0.1 as a test set.
S202, constructing a convolutional noise reduction self-encoder containing a gate control circulating unit GRU of a bearing fault diagnosis model, respectively mixing noise of a training set, a verification set and a test set, and training the convolutional noise reduction self-encoder containing the gate control circulating unit GRU by using the training set, the verification set and the test set after mixing noise to obtain the trained convolutional noise reduction self-encoder containing the gate control circulating unit GRU.
When training the convolutional noise-reducing self-encoder comprising the gate-controlled circulating unit GRU by utilizing the training set, the verification set and the test set after mixed noise, after a certain training round number is completed, a section of fault signal can be randomly intercepted, the signal reconstructed by the convolutional noise-reducing self-encoder comprising the gate-controlled circulating unit GRU is compared with a real signal, and the closer the signal reconstructed by the convolutional noise-reducing self-encoder comprising the gate-controlled circulating unit GRU is to the real signal, the better the feature extraction effect is proved, so that whether the training of the convolutional noise-reducing self-encoder comprising the gate-controlled circulating unit GRU is completed can be judged according to the feature extraction effect.
S203, adding a classifier for a convolution noise reduction self-encoder comprising a gating circulating unit GRU to construct a complete bearing fault diagnosis model.
S204, aiming at the rolling shaftVarious fault detection result types of the bearing respectively collect vibration signal sequence samples of a plurality of rolling bearings to obtain a sample set S 1 The method comprises the steps of carrying out a first treatment on the surface of the Sample set S 1 Dividing the training set, the verification set and the test set; and respectively carrying out standardized processing on the training set, the verification set and the test set, adding a label of the fault detection result type, and converting into One-Hot coding (single-Hot coding).
S205, using sample set S 1 Training the classifier by the training set, the verification set and the test set, thereby obtaining a trained bearing fault diagnosis model.
In this embodiment, the functional expression of the sample set S in step S201 is:
S＝{D 1 ，D 2 ，D 3 ，...，D n }，
in the above, D 1 ～D n Respectively n vibration signal sequence samples of the rolling bearings;
sample set S in step S204 1 The functional expression of (2) is:
S 1 ＝{D 1 ，D 2 ，D 3 ，...，D n }，
in the above, D 1 ～D n The vibration signal sequence samples of n rolling bearings are respectively:
D i ＝[x i ，x i+1 ，...，x i+w-1 ]，
in the above, D i For the samples of the vibration signal sequence of the ith rolling bearing, x i ～x i+w-1 W vibration signals in the vibration signal sequence are respectively.
In this embodiment, the function expression of the normalization process is:
in the above, x scale To the result of the normalization processing of the input vibration signal x, x mean Std is the variance of the vibration signal x, which is the mean value of the vibration signal x.
In step S202 of this embodiment, when training a convolutional noise reduction self-encoder including a gate-controlled cyclic unit GRU, a loss function is used as a mean-square error (MSE), and the function expression is as follows:
in the above equation, MSE represents the mean square error as a function of loss, n is the number of samples, w i Is the weight, y i Is thatCorresponding true value, < >>For the predicted value of the convolutional noise reduction self-encoder comprising the gate-control loop unit GRU, the subscript i is the sequence number of the sample.
When a label of the fault detection result type is added in step S204 in this embodiment, each vibration signal sequence sample D is obtained from the bearing signals collected under different fault states and normal conditions i Labels are made, for example 0 for samples in normal conditions, 1, 2, 3, etc. for samples in different faults. In the training of the classifier in step S205 of this embodiment, the loss function used is Cross Entropy (Cross-Entropy), and the function expression is:
in the above formula, H (p||q) represents cross entropy as a loss function, p is a true probability distribution, q is a probability distribution predicted by a bearing fault diagnosis model, p (i) is a true probability distribution of an ith sample, q (i) is a probability distribution predicted by a bearing fault diagnosis model of the ith sample, and subscript i is a sequence number of the sample. After training the classifier by using the training set and the verification set, a trained bearing fault diagnosis model formed by a trained convolution noise reduction self-encoder containing a gate control circulating unit GRU and the classifier can be obtained, then the test set is input into the bearing fault diagnosis model for fault diagnosis, and loss and accuracy of the bearing fault diagnosis model can be obtained.
In this embodiment, the mixing of noise in step S202 is to add gaussian white noise with variance and noise figure of a specified size, which is specifically accomplished by a noise adding layer.
After the training of the bearing fault diagnosis model is completed, collecting time sequence signals of rolling bearing vibration, adding certain noise into time sequence data of original bearing vibration to obtain signal data mixed with noise, taking the data mixed with noise as input, inputting the data into a convolution noise reduction self-encoder comprising a gate control circulation unit GRU, and carrying out self-adaptive feature extraction; and inputting the extracted features into a Softmax classifier for fault classification, and adjusting the model according to the result until the expected effect is achieved.
To verify the effect of the bearing failure diagnosis model of this embodiment, a bearing public data set of university of kesixi, usa was selected for experimental verification in this embodiment. Selecting 10 different states under the driving end of the rolling bearing: in a normal state, the rolling element has a pitting diameter of 0.1778mm, the rolling element has a pitting diameter of 0.3556mm, the rolling element has a pitting diameter of 0.5334mm, the inner ring has a pitting diameter of 0.1778mm, the inner ring has a pitting diameter of 0.3556mm, the inner ring has a pitting diameter of 0.5334mm, the outer ring has a pitting diameter of 0.1778mm (in the 6-point direction at the center position), the outer ring has a pitting diameter of 0.3556mm (in the 6-point direction at the center position), and the outer ring has a pitting diameter of 0.5334mm (in the 6-point direction at the center position). The sampling frequency of the 10 types of data is 48kHz, the motor load is 0, and the approximate rotating speed of the motor is 1797r/min.
The data set is constructed in a sliding window manner. The sliding window length is 1024, the step length is 256, 10000 sample data are constructed in total, and the total number of the sample data is 1000 in each type of state. Constructing a training set, a verification set and a test set according to the proportion of 7:2:1. To reduce the chance of end results, the model test will be run 10 times with the average of the diagnostic results of the 10 test sets as the end diagnostic result. Examples of two different faults and normal bearing raw vibration signal (raw Data) and noise added vibration signal (noise Data) are shown in fig. 5.
In the bearing failure diagnosis model, the individual network layer parameters are shown in the following table.
When the bearing fault diagnosis model is trained, the batch size is set to be 64, the iteration times are 350, the optimizer is Adam, the programming language is python3.8, a keras (based on a Tensorfow2.0 environment) is used as a deep learning framework, the whole bearing fault diagnosis model is constructed to carry out self-adaptive feature extraction and classification, and the training process is shown in fig. 6. As can be seen from fig. 6, although the Loss function of the bearing failure diagnosis model (the value of the Training Loss function, the value of the Validation Loss function) fluctuates with the number of iterations (epochs), the overall trend is declining due to the addition of noise Training, and gradually becomes stable after a certain number of iterations.
Samples in the test set are randomly selected as inputs, resulting in a curve of the reconstructed vibration signal (prediction_data) from the encoder containing convolutional noise reduction of the gated loop unit GRU and a curve of the original vibration signal (read_data) as shown in fig. 7. As can be seen from fig. 7, the curve of the vibration signal (prediction_data) reconstructed from the encoder by the convolutional noise reduction self-encoder including the gate-control unit GRU, which is indicated by the dashed line, is very close to the curve shape of the original vibration signal (read_data) which is indicated by the solid line, which indicates that the convolutional noise reduction self-encoder model including the gate-control unit GRU can adaptively reconstruct the shape of the original signal from the original data including noise, and the model learns the pattern implicit in the original signal, so that the timing characteristics in the noise-added signal can be extracted well. And taking the output of the last layer of the convolution noise reduction self-encoder model containing the gate control circulating unit GRU, recording the labels of the features extracted from different types of faults, and inputting the extracted features and the corresponding labels into a Softmax classifier for classification.
To avoid the occurrence of accidental classification, the average value of 10 diagnostic results is taken as the final diagnostic result, wherein the confusion matrix of a certain classification is shown in fig. 8. In fig. 8, the vertical axis represents the number of the actual fault type, and the horizontal axis represents the number of the diagnosed fault type. It can be seen that the fault diagnosis accuracy of the rolling element pitting diameter 0.3556mm is 100%, the fault diagnosis accuracy of the rest faults is 100%, and the average accuracy of final diagnosis of 1000 samples in the test set is 98.5%.
In order to verify that the bearing fault diagnosis model has certain universality, a German Pade Boen data set is selected for verification, and four types of normal bearing vibration data (K001), bearing outer ring fault data (KA 05), bearing inner ring fault data (KI 01) and bearing data (KB 23) of accelerated life test damage in the data set are selected, wherein the working conditions and conditions are the same, and the fault diagnosis result is shown in fig. 9. As can be seen from fig. 9, in 800 test samples, the fault diagnosis accuracy can reach 98.6%, and the fault diagnosis method has a good fault diagnosis effect, and proves that the model is applicable to similar bearing data sets and has a certain universality.
In summary, the bearing fault prediction method based on the convolutional noise reduction self-encoder of the embodiment includes collecting a vibration signal sequence of a rolling bearing and inputting the vibration signal sequence into a bearing fault diagnosis model, extracting vibration signal characteristics from the convolutional noise reduction self-encoder containing a gate control circulation unit (GRU) therein, and classifying by a classifier to obtain a fault detection result of the rolling bearing; the convolutional noise reduction self-encoder comprising the gate control circulation unit GRU is an encoding-decoding structure formed by connecting an encoding unit and a decoding unit, wherein the encoding unit comprises a noise adding layer, a first GRU network, a convolutional layer and a pooling layer which are connected in sequence, and the decoding unit comprises a deconvolution layer, a pooling layer and a second GRU network which are connected. The bearing fault prediction method based on the convolution noise reduction self-encoder is characterized by combining the advantages of the self-encoder that the deep learning model has strong feature extraction capability and the GRU network can process time sequence features, so that the neural network model can realize the fault prediction of the bearing and also can realize the fault diagnosis of the bearing. The bearing fault prediction method based on the convolution noise reduction self-encoder can achieve the effects that a trained bearing fault diagnosis model can input a section of time sequence data of the intercepted bearing vibration signal to obtain a section of time sequence data for predicting the next bearing vibration signal, and when the distance between the predicted vibration signal and the vibration signal in the normal state of the bearing is larger than a threshold value, faults are proved to occur. The trained bearing fault diagnosis model can input a section of time sequence data of the obtained bearing vibration signal to obtain a predicted fault state of the bearing, which is beneficial to maintenance of mechanical equipment. According to the bearing fault prediction method based on the convolution noise reduction self-encoder, the time sequence features in the bearing time sequence vibration signals can be adaptively extracted, the feature extraction capacity is enhanced, the noise influence is reduced, and hidden layer features are input into the classifier for recognition so as to realize accurate and efficient bearing fault prediction and diagnosis.
In addition, the embodiment also provides a bearing fault prediction system based on the convolution noise reduction self-encoder, which comprises a microprocessor and a memory which are connected with each other, wherein the microprocessor is programmed or configured to execute the bearing fault prediction method based on the convolution noise reduction self-encoder. Furthermore, the present embodiment also provides a computer-readable storage medium having stored therein a computer program for being programmed or configured by a microprocessor to perform the convolutional noise reduction self-encoder based bearing failure prediction method.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the present invention may occur to one skilled in the art without departing from the principles of the present invention and are intended to be within the scope of the present invention.

Claims (10)

1. The bearing fault prediction method based on the convolution noise reduction self-encoder is characterized by comprising the following steps of:
s101, collecting a vibration signal sequence of a rolling bearing;
s102, inputting a vibration signal sequence into a trained bearing fault diagnosis model formed by a convolution noise reduction self-encoder containing a gate control circulating unit GRU and a classifier, extracting vibration signal characteristics through the convolution noise reduction self-encoder containing the gate control circulating unit GRU, and classifying by the classifier to obtain a fault detection result of the rolling bearing; the convolutional noise reduction self-encoder comprising the gate control circulation unit GRU is an encoding-decoding structure formed by connecting an encoding unit and a decoding unit, wherein the encoding unit comprises a noise adding layer, a first GRU network, a convolution layer and a pooling layer which are connected in sequence, and the decoding unit comprises a deconvolution layer, an anti-pooling layer and a second GRU network which are connected.
2. The method of claim 1, wherein the classifier is a Softmax classifier.
3. The method for predicting bearing failure based on convolutional noise reduction self-encoder of claim 1 or 2, wherein step S102 is preceded by training a bearing failure diagnosis model:
s201, collecting vibration signal sequences of a plurality of rolling bearings to obtain a sample set S; dividing a sample set S into a training set, a verification set and a test set; respectively carrying out standardized treatment on the training set, the verification set and the test set;
s202, constructing a convolutional noise reduction self-encoder containing a gate control circulating unit GRU of a bearing fault diagnosis model, respectively mixing noise to a training set, a verification set and a test set, and training the convolutional noise reduction self-encoder containing the gate control circulating unit GRU by using the training set, the verification set and the test set after mixing noise to obtain a trained convolutional noise reduction self-encoder containing the gate control circulating unit GRU;
s203, adding a classifier for a convolution noise reduction self-encoder comprising a gate control circulating unit GRU to construct a complete bearing fault diagnosis model;
s204, respectively collecting vibration signal sequence samples of a plurality of rolling bearings according to various fault detection result types of the rolling bearings to obtain a sample set S 1 The method comprises the steps of carrying out a first treatment on the surface of the Sample set S 1 Dividing the training set, the verification set and the test set; respectively carrying out standardized processing on the training set, the verification set and the test set, adding a label of a fault detection result type, and converting the label into One-Hot codes;
s205, using sample set S 1 Training the classifier by the training set, the verification set and the test set, thereby obtaining a trained bearing fault diagnosis model.
4. The method for predicting bearing failure based on convolutional noise reduction self-encoder as defined in claim 3, wherein the training of convolutional noise reduction self-encoder comprising gate-control unit GRU in step S202 uses a loss function of:
in the above equation, MSE represents the mean square error as a function of loss, n is the number of samples, i is the weight, y i Is thatCorresponding true value, < >>For the predicted value of the convolutional noise reduction self-encoder comprising the gate-control loop unit GRU, the subscript i is the sequence number of the sample.
5. The method for predicting bearing failure based on convolutional noise reduction self-encoder as recited in claim 3, wherein the loss function used in training the classifier in step S205 is:
in the above formula, H (p||q) represents cross entropy as a loss function, p is a true probability distribution, q is a probability distribution predicted by a bearing fault diagnosis model, p (i) is a true probability distribution of an ith sample, q (i) is a probability distribution predicted by a bearing fault diagnosis model of the ith sample, and subscript i is a sequence number of the sample.
6. The method for predicting bearing failure based on convolutional noise reduction self-encoder as recited in claim 3, wherein the functional expression of the sample set S in step S201 is:
S＝{D 1 ,D 2 ,D 3 ,...,D n }，
in the above, D 1 ～D n Respectively n vibration signal sequence samples of the rolling bearings;
sample set S in step S204 1 The functional expression of (2) is:
S 1 ＝{D 1 ,D 2 ,D 3 ,...,D n }，
in the above, D 1 ～D n The vibration signal sequence samples of n rolling bearings are respectively:
D i ＝[x i ,x i+1 ,...,x i+w-1 ]，
in the above, D i For the samples of the vibration signal sequence of the ith rolling bearing, x i ～x i+w-1 W vibration signals in the vibration signal sequence are respectively.
7. A method of predicting bearing failure based on a convolutional noise reduction self-encoder as defined in claim 3, wherein the functional expression of the normalization process is:
in the above, x scale To the result of the normalization processing of the input vibration signal x, x mean Std is the variance of the vibration signal x, which is the mean value of the vibration signal x.
8. The method for predicting bearing failure based on convolutional noise reduction self-encoder as recited in claim 3, wherein the step S202 of mixing noise means adding gaussian white noise of a variance and noise figure of a specified magnitude.
9. A convolutional noise reduction self-encoder based bearing failure prediction system comprising a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to perform the convolutional noise reduction self-encoder based bearing failure prediction method of any one of claims 1-8.
10. A computer readable storage medium having a computer program stored therein, wherein the computer program is for programming or configuring by a microprocessor to perform the convolutional noise reduction self-encoder based bearing failure prediction method of any one of claims 1-8.
CN202310315205.8A 2023-03-28 2023-03-28 Bearing fault prediction method and system based on convolution noise reduction self-encoder Pending CN116467652A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310315205.8A CN116467652A (en) 2023-03-28 2023-03-28 Bearing fault prediction method and system based on convolution noise reduction self-encoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310315205.8A CN116467652A (en) 2023-03-28 2023-03-28 Bearing fault prediction method and system based on convolution noise reduction self-encoder

Publications (1)

Publication Number Publication Date
CN116467652A true CN116467652A (en) 2023-07-21

Family

ID=87181601

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310315205.8A Pending CN116467652A (en) 2023-03-28 2023-03-28 Bearing fault prediction method and system based on convolution noise reduction self-encoder

Country Status (1)

Country Link
CN (1) CN116467652A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117195105A (en) * 2023-11-08 2023-12-08 北京科技大学 Gear box fault diagnosis method and device based on multilayer convolution gating circulation unit

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117195105A (en) * 2023-11-08 2023-12-08 北京科技大学 Gear box fault diagnosis method and device based on multilayer convolution gating circulation unit
CN117195105B (en) * 2023-11-08 2024-03-19 北京科技大学 Gear box fault diagnosis method and device based on multilayer convolution gating circulation unit

Similar Documents

Publication Publication Date Title
CN112784965B (en) Large-scale multi-element time series data anomaly detection method oriented to cloud environment
CN113935406B (en) Mechanical equipment unsupervised fault diagnosis method based on countermeasure flow model
CN110942101B (en) Rolling bearing residual life prediction method based on depth generation type countermeasure network
CN106682688B (en) Particle swarm optimization-based stacked noise reduction self-coding network bearing fault diagnosis method
Cheng et al. Autoencoder quasi-recurrent neural networks for remaining useful life prediction of engineering systems
Wang et al. Self-supervised signal representation learning for machinery fault diagnosis under limited annotation data
Chen et al. Machine fault classification using deep belief network
CN106124212A (en) Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine
Sun et al. Stacked denoising autoencoder with density-grid based clustering method for detecting outlier of wind turbine components
CN111459144A (en) Airplane flight control system fault prediction method based on deep cycle neural network
CN110991471B (en) Fault diagnosis method for high-speed train traction system
Wang et al. A modified generative adversarial network for fault diagnosis in high-speed train components with imbalanced and heterogeneous monitoring data
CN115791174B (en) Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium
CN116467652A (en) Bearing fault prediction method and system based on convolution noise reduction self-encoder
CN115290326A (en) Rolling bearing fault intelligent diagnosis method
CN113485863B (en) Method for generating heterogeneous imbalance fault samples based on improved generation of countermeasure network
Liu et al. A novel health prognosis method for system based on improved degenerated Hidden Markov model
CN116821730B (en) Fan fault detection method, control device and storage medium
Zhao et al. A capsnet-based fault diagnosis method for a digital twin of a wind turbine gearbox
Chen et al. Reliability analysis using deep learning
Xu et al. Bearing Fault Diagnosis in the Mixed Domain Based on Crossover‐Mutation Chaotic Particle Swarm
CN116910559A (en) Index anomaly detection method for intelligent operation and maintenance application of power grid supercomputer center
Kibrete et al. Applications of artificial intelligence for fault diagnosis of rotating machines: A review
CN114781450B (en) Train rolling bearing state identification method based on parameter optimization MOMEDA-MIA-CNN
CN116007937A (en) Intelligent fault diagnosis method and device for mechanical equipment transmission part

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination