CN113392475A - Intelligent fault identification method for speed reducer of industrial robot - Google Patents

Intelligent fault identification method for speed reducer of industrial robot Download PDF

Info

Publication number
CN113392475A
CN113392475A CN202110516224.8A CN202110516224A CN113392475A CN 113392475 A CN113392475 A CN 113392475A CN 202110516224 A CN202110516224 A CN 202110516224A CN 113392475 A CN113392475 A CN 113392475A
Authority
CN
China
Prior art keywords
encoder
layer
self
wavelet
function
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
CN202110516224.8A
Other languages
Chinese (zh)
Inventor
邵海东
程军圣
钟翔
王续达
吕顺娣
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.)
Qingdao International Academician Port And Qinghu Science And Technology Collaborative Innovation Research Institute
Qingdao International Academician Port Group Intelligent Construction Development Co ltd
Hunan University
Original Assignee
Qingdao International Academician Port And Qinghu Science And Technology Collaborative Innovation Research Institute
Qingdao International Academician Port Group Intelligent Construction Development Co ltd
Hunan University
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 Qingdao International Academician Port And Qinghu Science And Technology Collaborative Innovation Research Institute, Qingdao International Academician Port Group Intelligent Construction Development Co ltd, Hunan University filed Critical Qingdao International Academician Port And Qinghu Science And Technology Collaborative Innovation Research Institute
Priority to CN202110516224.8A priority Critical patent/CN113392475A/en
Publication of CN113392475A publication Critical patent/CN113392475A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • 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
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/028—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • 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
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/10—Noise analysis or noise optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/12—Timing analysis or timing optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Medical Informatics (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses an intelligent fault identification technology for an industrial robot speed reducer, which is characterized in that a depth wavelet self-encoder is constructed based on a Morlet wavelet function and a related entropy correction loss function, the characteristics implicit in original input data which are unstable, nonlinear and strongly coupled and contain a large amount of background noise are automatically mined, and the highly nonlinear mapping relation between the original input data and various running states is directly established, so that the fault mode of the industrial robot speed reducer is accurately identified.

Description

Intelligent fault identification method for speed reducer of industrial robot
Technical Field
The invention belongs to the field of fault identification, and particularly relates to an intelligent fault identification method for an industrial robot speed reducer.
Background
At present, industrial robots such as welding robots, assembly robots, spraying robots, and carrying robots are widely used in the fields of intelligent construction industry, machining industry, automobile part manufacturing industry, electronic and electrical industry, rubber and plastic industry, food industry, and wood and furniture manufacturing industry, and play an increasingly important role. The industrial robot has severe operating environment and heavy operation tasks, and a speed reducer which is an important part of the industrial robot is easy to break down. The method has the advantages that advanced intelligent fault identification technology is researched and developed, various fault modes of the speed reducer of the industrial robot can be distinguished accurately and reliably, and the method is an important means for enhancing fault diagnosis of the industrial robot and realizing safe operation of the industrial robot. However, the speed reducer has a complex structure, the motion is not uniform, the detection signal has strong non-stationarity, and the fault characteristic is often submerged in the coupled vibration and complex background noise of the structural component, so that the fault identification of the speed reducer has great challenge.
Some adopt machine learning models to perform intelligent recognition, but traditional shallow machine learning models, such as artificial neural networks, support vector machines, random forests and the like, have been widely applied to intelligent fault recognition research of industrial equipment, and because the input of the model is feature data which must be manually lifted, the following three disadvantages still exist: (1) due to the complexity of actually acquiring non-stationary signals, feature extraction often depends on various advanced signal processing technologies; (2) the most sensitive features selected from the original feature set are time-consuming and labor-consuming, and require more abundant engineering practice experience; (3) the portability of feature extraction and selection is not high, and features often need to be re-extracted when different diagnostic tasks are solved. A depth Auto-encoder (DAE for short) is a common unsupervised Deep learning method, and aims to minimize an error between original input data and reconstructed output data by continuously adjusting weights between layers, and finally automatically obtain a hidden layer feature representation of the original input data.
Although the DAE gets rid of the dependence on the work of extracting the artificial features to a great extent and provides an important support for realizing intelligent fault identification based on original monitoring data, because the vibration signal of the industrial robot speed reducer on the actual site has strong non-stationarity and high non-linearity and contains a large amount of interference components such as background noise, a new technology needs to be introduced to further improve the identification performance of the DAE.
Disclosure of Invention
In order to solve the problems, the invention provides an intelligent fault identification method for an industrial robot speed reducer. The invention overcomes the problems that the recognition effect of the traditional shallow machine learning method depends heavily on signal processing and artificial feature extraction and the recognition effect of the existing deep self-encoder is reduced when processing the original input data which is unstable, nonlinear and strongly coupled and contains a large amount of background noise, directly and accurately establishes the highly nonlinear mapping relation between the original complex input data and the changeable running states, and realizes reliable fault recognition of the industrial robot speed reducer.
In order to achieve the technical effects, the technical scheme of the invention is as follows:
an intelligent fault identification method for an industrial robot speed reducer comprises the following steps:
step S1: collecting original vibration acceleration time domain signals of a speed reducer of an industrial robot under different health states, and dividing the signals into training samples and testing samples after mean value removing and normalization processing;
step S2: a Morlet wavelet function is adopted as an activation function of a hidden layer of an auto-encoder, a loss function of the auto-encoder is corrected by adopting a correlation entropy, and the wavelet auto-encoder is designed;
step S3: stacking a plurality of wavelet auto-encoders and a Softmax classifier to construct a depth wavelet auto-encoder;
step S4: inputting a training sample into a depth wavelet self-encoder, training to obtain an intelligent fault recognition model, checking the fault recognition effect of the intelligent fault recognition model by using a test sample, obtaining the trained intelligent fault recognition model if the fault recognition effect is achieved, and adding the training sample to retrain the training process if the fault recognition effect is not achieved until the trained intelligent fault recognition model is obtained;
step S5: and recognizing the fault type of the speed reducer of the industrial robot by adopting a trained intelligent fault recognition model.
In a further improvement, the specific step of using the Morlet wavelet function as the hidden layer activation function of the self-encoder in step S2 is as follows:
step S21: the Morlet wavelet function is used as a hidden layer activation function of a self-encoder for nonlinear transformation of data of an input layer, and the output of a hidden layer node after transformation, namely the characteristic obtained by learning, is as follows;
Figure BDA0003062306640000031
wherein h isjIs the output from the encoder hidden layer jth node, j is 1,2,.. p, p is the number of hidden layer nodes, m is the number of input layer nodes from the encoder, i.e. the length of the input sample,
Figure BDA0003062306640000032
for m-dimensional input samples without labels, xkThe k-th dimension element representing the input sample x, i.e., the data of the input layer node k, k is 1, 2.
Figure BDA0003062306640000033
Representing an m-dimensional real vector space, wjkRepresenting the connection weight between the input level node k and the jth hidden level node, ajAnd cjRespectively a scaling factor and a translation factor of the wavelet activation function of the jth hidden layer node;
step S22: the hyperbolic tangent function is used as a self-encoder hidden layer activation function for nonlinear transformation of hidden layer characteristic data, the transformed output of a layer node is output, namely reconstruction data, and the reconstruction data form a reconstruction sample:
Figure BDA0003062306640000034
wherein the content of the first and second substances,
Figure BDA0003062306640000035
is the output from the ith node of the output layer of the encoder, i.e. the ith term x of the input sample xiThe reconstruction data of (a) is reconstructed,
Figure BDA0003062306640000036
as reconstructed samples
Figure BDA0003062306640000037
I-th dimension data of wijRepresenting the connection weight between the ith node of the output layer and the jth node of the hidden layer; tanh represents the hyperbolic tangent function.
In a further improvement, in the step S2, for correcting the loss function of the self-encoder by using the correlation entropy, the specific steps are as follows:
step S23: measuring a reconstruction error between an input sample and an output sample of the self-encoder by adopting a correlation entropy function;
Figure BDA0003062306640000041
wherein σ is the Gaussian kernel width; x is the number ofiAn ith dimension element representing an input sample x;
step S24: on the basis of the correlation entropy function, introducing a sparsity constraint term and a weight attenuation term, reducing the complexity of the model, and avoiding overfitting:
Figure BDA0003062306640000042
where L represents a loss function, ρ represents a sparse coefficient,
Figure BDA0003062306640000043
representing the average activation value of the jth node in the hidden layer, beta representing a sparse penalty factor, lambda representing a weight attenuation coefficient, slThe number of neuron nodes at the l layer of the self-encoder, the number of neuron nodes at the first layer s1Number of neuron nodes in second layer s2P, number of neuron nodes in third layer s3＝s1＝m，
Figure BDA0003062306640000044
The connection weight between the l layer and the l +1 layer of the self-encoder is obtained;
step S25: adopting a stochastic gradient descent algorithm introducing momentum terms to iteratively update wavelet self-encoder model parameters, minimizing a loss function, and obtaining Wij，Wjk，aj，cjA value of (d);
Figure BDA0003062306640000045
wherein θ ═ Wij,Wjk,aj,cjThe parameters of the wavelet self-encoder model are theta (t +1), if the meaning of theta is different from that of theta, the 4 parameters are all in the range of-1, 1 before iteration]Internal random initialization, t represents iteration times, eta belongs to (0,1) and represents learning rate, alpha is momentum factor and takes the value of [0.8,1]L (t) represents the model loss at the t-th iteration; theta (t) represents the wavelet self-encoder model parameter at the t iteration, theta (t +1) represents the wavelet self-encoder model parameter at the t +1 iteration, and t represents the iteration times，
Figure BDA0003062306640000046
Representing a partial derivative calculation.
The invention has the advantages that:
the invention overcomes the problems that the recognition effect of the traditional shallow machine learning method depends heavily on signal processing and artificial feature extraction and the recognition effect of the existing deep self-encoder is reduced when processing the original input data which is unstable, nonlinear and strongly coupled and contains a large amount of background noise, directly and accurately establishes the highly nonlinear mapping relation between the original complex input data and the changeable running states, and realizes reliable fault recognition of the industrial robot speed reducer.
Drawings
FIG. 1 is a flow chart of an intelligent fault identification technique for a speed reducer of an industrial robot according to the invention;
FIG. 2 is a graph of the time domain signals of the original vibration acceleration for various fault conditions of the speed reducer of the present invention;
FIG. 3 is a diagram of vibration acceleration time domain signals of various fault states of the reducer according to the present invention, which are subjected to mean value removal and normalization;
FIG. 4 is a multi-classification confusion matrix diagram for intelligent fault identification of the reducer of the present invention;
FIG. 5 is a F-measurement diagram for intelligent fault identification of the speed reducer of the present invention.
Detailed Description
The technical means of the present invention will be specifically described below by way of specific embodiments.
Example 1
In the embodiment, a speed reducer Data set PHM 2009Challenge Data issued by the International Prediction and Health Management (PHM) Association is adopted to verify the effectiveness of the technology in intelligent fault identification of the speed reducer, wherein the effectiveness comprises health states and various fault states of a gear, a bearing and a rotating shaft. Three shafts (input shaft IS, intermediate shaft ID, output shaft OS) are arranged in the testing device. Two groups of meshing gears adopted in the experiment have a straight gear mode and a helical gear mode. There are 8 complex failure modes in the helical gear (Spur) meshing mode (see table 1), the rotation speed is 1800rpm, and the load-bearing mode is low load. Vibration sensors are respectively arranged on two sides of the box body and used for collecting data, the sampling frequency is 66.67kHz, the sampling time is 4s, and the number of sampling points is 266656.
Table 1 speed reducer 8 composite failure modes
Figure BDA0003062306640000061
FIG. 2 is an original vibration acceleration time domain signal of various fault states of the speed reducer, the sampling frequency is 66.67kHz, the sampling time is 4s, the number of sampling points is 266656, and the abscissa in the figure represents time in units of s; the ordinate represents the amplitude in the unit of 10 mv/g. Fig. 3 is a vibration acceleration time domain signal of various fault states of the speed reducer after mean value removal and normalization, wherein the abscissa represents time, the unit is s, the ordinate is dimensionless, the range is-1 to 1, the first 265800 points are selected to divide training samples and test samples, each sample is 4800 points, 3000 points are overlapped between the front sample and the rear sample, 86 training samples of each state are included, and 60 test samples of each state are included.
An intelligent fault identification model based on a depth wavelet self-encoder is constructed, and in the example, the hyper-parameters are set as follows: the gaussian kernel width sigma is 1.5, the sparse coefficient rho is 0.1, the sparse penalty term factor beta is 5, the weight attenuation coefficient lambda is 0.002, the learning rate eta is 0.01, the momentum factor alpha is 0.86, the iteration time t is 100, and the depth structure is 4800 + 2000 + 800 + 300-8'. Table 2 shows the comparison results of the method of the present invention and intelligent recognition methods such as a depth autoencoder, an artificial neural network, a random forest, a support vector machine, etc., after ten independent operations. The input of the depth self-encoder is the original time domain signal, the input of the artificial neural network, the random forest and the support vector machine is 16 time domain characteristic values, and the time domain characteristic values specifically comprise an average value, a root mean square, a square root amplitude value, an absolute average value, a maximum value, a minimum value, a peak-to-peak value, a variance, a skewness, a kurtosis, a waveform index, a peak index, a pulse index, a margin index, a skewness index and a kurtosis index. Specifically analyzing a certain recognition result in 10 runs of the invented method, the overall classification accuracy of intelligent fault recognition is 98.33% (472/480), and the multi-classification confusion matrix and the F-measured value are respectively shown in FIG. 4 and FIG. 5. As can be seen from table 2, fig. 4 and fig. 5, with the method of the present invention, the average recognition accuracy is the highest, and various fault states of the speed reducer can be effectively distinguished based on the original signal.
TABLE 2 comparison of predicted results
Figure BDA0003062306640000071
Referring to fig. 1, the present invention can be mainly divided into three parts. The first part is to collect original vibration acceleration time domain signals of a speed reducer of an industrial robot under different health states, and the signals are divided into training samples and testing samples after mean value removing and normalization processing; the second part adopts a Morlet wavelet function as a hidden layer activation function of the self-encoder, adopts related entropy to correct a loss function of the self-encoder, designs a plurality of wavelet self-encoders and adds a Softmax classifier, and finally constructs the depth wavelet self-encoder; the third part is that the test sample checks the trained intelligent fault recognition model based on the depth wavelet self-encoder.
Referring to fig. 2, the original vibration acceleration time domain signals of various fault states of the speed reducer comprise health states and various fault states of gears, bearings and rotating shafts. The sampling frequency is 66.67kHz, the sampling time is 4s, the number of sampling points is 266656, and the abscissa in the figure represents time in units of s; the ordinate represents the amplitude in the unit of 10 mv/g.
Referring to fig. 3, in the vibration acceleration time domain signals of various types of fault states of the speed reducer subjected to mean value removal and normalization, the abscissa represents time, the unit is s, the ordinate is dimensionless, the range is-1 to 1, the first 265800 points are selected to divide training samples and test samples, each sample is 4800 points, 3000 points are overlapped between the front sample and the rear sample, namely the first sample is the 1 st point to the 4800 th point, the 2 nd sample is the 2401 st point to the 7200 th point, and 146 samples are obtained in total, wherein 86 training samples of each type of state are obtained, and 60 test samples of each type of state are obtained.
Referring to fig. 4, the multi-classification confusion matrix diagram for intelligent fault identification of the speed reducer is characterized in that the abscissa is a predicted state label, the ordinate is a real state label, and the main diagonal number represents the identification accuracy rate of the category
Referring to fig. 5, in an F-measurement value diagram for intelligent fault identification of a speed reducer, the abscissa is a state label, the ordinate is an F-measurement value, the range is 0 to 1, and the larger the value is, the better the identification effect of the category is represented.
The above description is only one specific guiding embodiment of the present invention, but the design concept of the present invention is not limited thereto, and any insubstantial modification of the present invention using this concept shall fall within the scope of the invention.

Claims (3)

1. An intelligent fault identification method for an industrial robot speed reducer is characterized by comprising the following steps:
step S1: collecting original vibration acceleration time domain signals of a speed reducer of an industrial robot under different health states, and dividing the signals into training samples and testing samples after mean value removing and normalization processing;
step S2: a Morlet wavelet function is adopted as an activation function of a hidden layer of an auto-encoder, a loss function of the auto-encoder is corrected by adopting a correlation entropy, and the wavelet auto-encoder is designed;
step S3: stacking a plurality of wavelet auto-encoders and a Softmax classifier to construct a depth wavelet auto-encoder;
step S4: inputting a training sample into a depth wavelet self-encoder, training to obtain an intelligent fault recognition model, checking the fault recognition effect of the intelligent fault recognition model by using a test sample, obtaining the trained intelligent fault recognition model if the fault recognition effect is achieved, and adding the training sample to retrain the training process if the fault recognition effect is not achieved until the trained intelligent fault recognition model is obtained;
step S5: and recognizing the fault type of the speed reducer of the industrial robot by adopting a trained intelligent fault recognition model.
2. The intelligent fault identification method for the industrial robot speed reducer according to claim 1, wherein the specific steps of adopting the Morlet wavelet function as the hidden layer activation function of the self-encoder in the step S2 are as follows:
step S21: the Morlet wavelet function is used as a hidden layer activation function of a self-encoder for nonlinear transformation of data of an input layer, and the output of a hidden layer node after transformation, namely the characteristic obtained by learning, is as follows;
Figure FDA0003062306630000011
wherein h isjIs the output from the encoder hidden layer jth node, j is 1,2,.. p, p is the number of hidden layer nodes, m is the number of input layer nodes from the encoder, i.e. the length of the input sample,
Figure FDA0003062306630000012
for m-dimensional input samples without labels, xkThe k-th dimension element representing the input sample x, i.e., the data of the input layer node k, k is 1, 2.
Figure FDA0003062306630000021
Representing an m-dimensional real vector space, wjkRepresenting the connection weight between the input level node k and the jth hidden level node, ajAnd cjRespectively a scaling factor and a translation factor of the wavelet activation function of the jth hidden layer node;
step S22: the hyperbolic tangent function is used as a self-encoder hidden layer activation function for nonlinear transformation of hidden layer characteristic data, the transformed output of a layer node is output, namely reconstruction data, and the reconstruction data form a reconstruction sample:
Figure FDA0003062306630000022
wherein the content of the first and second substances,
Figure FDA0003062306630000023
is output from the encoderThe output of the ith node of the layer, i.e. the ith term x of the input sample xiThe reconstruction data of (a) is reconstructed,
Figure FDA0003062306630000024
as reconstructed samples
Figure FDA0003062306630000025
I-th dimension data of wijRepresenting the connection weight between the ith node of the output layer and the jth node of the hidden layer; tanh represents the hyperbolic tangent function.
3. The intelligent fault identification method for the industrial robot speed reducer according to claim 2, wherein in the step S2, the method for correcting the loss function of the self-encoder by using the correlation entropy comprises the following specific steps:
step S23: measuring a reconstruction error between an input sample and an output sample of the self-encoder by adopting a correlation entropy function;
Figure FDA0003062306630000026
wherein σ is the Gaussian kernel width; x is the number ofiAn ith dimension element representing an input sample x;
step S24: on the basis of the correlation entropy function, introducing a sparsity constraint term and a weight attenuation term, reducing the complexity of the model, and avoiding overfitting:
Figure FDA0003062306630000027
where L represents a loss function, ρ represents a sparse coefficient,
Figure FDA0003062306630000028
representing the average activation value of the jth node in the hidden layer, beta representing a sparse penalty factor, lambda representing a weight attenuation coefficient, slNumber of neuron nodes at layer I of self-encoder, layer I spiritNumber of warp meta-nodes s1Number of neuron nodes in second layer s2P, number of neuron nodes in third layer s3＝s1＝m，
Figure FDA0003062306630000031
The connection weight between the l layer and the l +1 layer of the self-encoder is obtained;
step S25: adopting a stochastic gradient descent algorithm introducing momentum terms to iteratively update wavelet self-encoder model parameters, minimizing a loss function, and obtaining Wij，Wjk，aj，cjA value of (d);
Figure FDA0003062306630000032
wherein θ ═ Wij,Wjk,aj,cjThe parameters of the wavelet self-encoder model are theta (t +1), if the meaning of theta is different from that of theta, the 4 parameters are all in the range of-1, 1 before iteration]Internal random initialization, t represents iteration times, eta belongs to (0,1) and represents learning rate, alpha is momentum factor and takes the value of [0.8,1]L (t) represents the model loss at the t-th iteration; theta (t) represents the wavelet self-encoder model parameter at the t iteration, theta (t +1) represents the wavelet self-encoder model parameter at the t +1 iteration, t represents the iteration number,
Figure FDA0003062306630000033
representing a partial derivative calculation.
CN202110516224.8A 2021-05-12 2021-05-12 Intelligent fault identification method for speed reducer of industrial robot Pending CN113392475A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110516224.8A CN113392475A (en) 2021-05-12 2021-05-12 Intelligent fault identification method for speed reducer of industrial robot

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110516224.8A CN113392475A (en) 2021-05-12 2021-05-12 Intelligent fault identification method for speed reducer of industrial robot

Publications (1)

Publication Number Publication Date
CN113392475A true CN113392475A (en) 2021-09-14

Family

ID=77616943

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110516224.8A Pending CN113392475A (en) 2021-05-12 2021-05-12 Intelligent fault identification method for speed reducer of industrial robot

Country Status (1)

Country Link
CN (1) CN113392475A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114279728A (en) * 2021-12-07 2022-04-05 郑州大学 Fault diagnosis method and system for vibrating screen body
CN114571459A (en) * 2022-03-22 2022-06-03 浙江钱江机器人有限公司 Industrial robot fault diagnosis method and system

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110991424A (en) * 2019-12-25 2020-04-10 安徽工业大学 Fault diagnosis method based on minimum entropy deconvolution and stacking sparse self-encoder

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110991424A (en) * 2019-12-25 2020-04-10 安徽工业大学 Fault diagnosis method based on minimum entropy deconvolution and stacking sparse self-encoder

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
SHAO HAIDONG: "Intelligent fault diagnosis of rolling bearing using deep wavelet auto-encoder with extreme learning machine", 《KNOWLE DGE-BASE D SYSTEMS》 *
杜小磊: "改进深层小波自编码器的轴承故障诊断方法", 《计算机工程与应用》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114279728A (en) * 2021-12-07 2022-04-05 郑州大学 Fault diagnosis method and system for vibrating screen body
CN114279728B (en) * 2021-12-07 2023-07-25 郑州大学 Fault diagnosis method and system for vibrating screen body
CN114571459A (en) * 2022-03-22 2022-06-03 浙江钱江机器人有限公司 Industrial robot fault diagnosis method and system
CN114571459B (en) * 2022-03-22 2023-07-21 浙江钱江机器人有限公司 Industrial robot fault diagnosis method and system

Similar Documents

Publication Publication Date Title
Zou et al. Bearing fault diagnosis method based on EEMD and LSTM
CN108171263B (en) Based on the Fault Diagnosis of Roller Bearings for improving variation mode decomposition and extreme learning machine
Xiang et al. Fault diagnosis of rolling bearing under fluctuating speed and variable load based on TCO spectrum and stacking auto-encoder
Kong et al. Attention recurrent autoencoder hybrid model for early fault diagnosis of rotating machinery
Liang et al. Rolling bearing fault diagnosis based on one-dimensional dilated convolution network with residual connection
Zhao et al. Model-driven deep unrolling: Towards interpretable deep learning against noise attacks for intelligent fault diagnosis
CN105973594A (en) Rolling bearing fault prediction method based on continuous deep belief network
Gecgel et al. Gearbox fault diagnostics using deep learning with simulated data
CN113392475A (en) Intelligent fault identification method for speed reducer of industrial robot
CN111753891B (en) Rolling bearing fault diagnosis method based on unsupervised feature learning
CN113923104B (en) Network fault diagnosis method, equipment and storage medium based on wavelet neural network
Huang et al. Deep residual networks-based intelligent fault diagnosis method of planetary gearboxes in cloud environments
CN109060347A (en) Based on the planetary gear fault recognition method for stacking de-noising autocoder and gating cycle unit neural network
Li et al. Multi-scale CNN for multi-sensor feature fusion in helical gear fault detection
CN115993248A (en) Bearing fault detection method, system, equipment and storage medium based on impulse neural network
Deng et al. Application of BP neural network and convolutional neural network (CNN) in bearing fault diagnosis
Han et al. An intelligent fault diagnosis method of variable condition gearbox based on improved DBN combined with WPEE and MPE
CN115290326A (en) Rolling bearing fault intelligent diagnosis method
CN113919220A (en) Intelligent fault diagnosis method for rolling bearing driven by vibration image
Peng et al. Research on fault diagnosis method of rolling bearing based on 2DCNN
CN113627268A (en) Model training method, and method and device for detecting fault of speed reducer for mine hoist
Liao et al. The fault diagnosis for centrifugal compressor based on time series analysis with neutral network
Sun et al. Fault diagnosis of planetary gearbox based on signal denoising and convolutional neural network
CN114818825A (en) Rotary machine fault diagnosis method based on convolution kernel multilayer distribution residual error network
Peng et al. A temporal convolutional network based method for fault diagnosis of DEH system

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210914