CN111562109A - Deep learning state identification and diagnosis method for mechanical equipment - Google Patents

Deep learning state identification and diagnosis method for mechanical equipment Download PDF

Info

Publication number
CN111562109A
CN111562109A CN202010423267.7A CN202010423267A CN111562109A CN 111562109 A CN111562109 A CN 111562109A CN 202010423267 A CN202010423267 A CN 202010423267A CN 111562109 A CN111562109 A CN 111562109A
Authority
CN
China
Prior art keywords
self
layer
encoder
negative
training
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
CN202010423267.7A
Other languages
Chinese (zh)
Inventor
张焱
韦航
黄庆卿
邓钦元
冯乔琦
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.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
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 Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN202010423267.7A priority Critical patent/CN111562109A/en
Publication of CN111562109A publication Critical patent/CN111562109A/en
Pending legal-status Critical Current

Links

Images

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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Acoustics & Sound (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a method for identifying and diagnosing deep learning states of mechanical equipment, and belongs to the field of fault diagnosis. Aiming at the problem of monitoring and diagnosing the state of mechanical equipment, firstly, time-frequency transformation is carried out on monitored vibration signal data to obtain a signal time-frequency amplitude spectrum. Secondly, stacking a plurality of self-encoders and a Softmax classifier to form a deep self-coding neural network, adding a network parameter non-negative constraint limiting item into the cost function, and training and optimizing the parameters of the deep self-coding neural network by using training sample data. And performing state recognition and diagnostic analysis on the real-time monitoring data by using a deep non-negative self-coding network model to obtain a specific classification recognition result. And finally, verifying by using the real data measured on the test platform, wherein the verification result shows that the method can effectively monitor and diagnose the state.

Description

Deep learning state identification and diagnosis method for mechanical equipment
Technical Field
The invention belongs to the field of fault diagnosis, and relates to a method for identifying and diagnosing deep learning states of mechanical equipment.
Background
The identification and diagnosis of the state of the mechanical equipment are important links for achieving the above-mentioned goals, and the indirect determination of the state of the mechanical equipment according to the mapping relationship between the working condition, measurable signals (such as acoustic emission, vibration and the like) and the state of the equipment has been widely researched. For example, Lei et al have been applied to engineering for diagnosing early faults of locomotive bearings based on an adaptive stochastic resonance method, and huang et al have been applied to diagnosing complex faults of planetary gear boxes by using a resonance sparse decomposition method based on an adaptive optimization quality factor. In recent years, the deep learning theory makes a major breakthrough, and achieves remarkable results in a plurality of recognition tasks, including the fields of voice recognition, visual object recognition and detection, equipment state monitoring and diagnosis and the like. In the field of monitoring and diagnosis, Lei et al developed a mechanical equipment big data health monitoring study based on deep learning theory. Kong et al established a deep learning model based on time-frequency fusion and attention mechanism suitable for planetary gearbox applications. Li et al performed synchronous generator fault diagnosis based on a deep belief network. Zhong et al have conducted intelligent fault diagnosis of marine diesel engines based on deep belief networks.
At present, the work of deep learning in the field of equipment state monitoring and diagnosis mainly focuses on diagnosis and judgment research under typical fault modes and constant working conditions, and state identification and diagnosis related to variable working conditions and early weak faults are less. In addition, the deep learning is realized on the basis of data driving, and the obtained deep learning model has certain expected characteristics by adding relevant constraints in the deep network training process, such as limiting the sparsity of a hidden layer. There are psychological and physiological related studies that suggest the presence of a part-based representation in the brain that conforms to the human perception that things are "local to the whole", while Lee et al studies indicate that part-based representations are often sparse and contribute to revealing hidden structures of the original data. The method is inspired by the idea of partial representation, realizes partial feature representation by adding a non-negative constraint rule in the training process of the deep network, strengthens the sparsity of the deep network, improves the performance of the deep network, and further meets the requirements of mechanical equipment state identification and diagnosis application.
Disclosure of Invention
In view of the above, the present invention provides a method for identifying and diagnosing a deep learning state of a mechanical device.
In order to achieve the purpose, the invention provides the following technical scheme:
a deep learning state identification and diagnosis method for mechanical equipment comprises the following steps:
s1: monitoring working condition information of mechanical equipment in the operation process, and acquiring equipment vibration signal data by using an acceleration sensor, wherein the sensor data are vibration signals in one or more directions of an x-axis direction, a y-axis direction and a z-axis direction so as to realize the on-line state identification and diagnosis of the mechanical equipment as a target;
s2: constructing a depth non-negative self-coding network, and setting a depth non-negative self-coding network structure and initial parameters;
s3: dividing the vibration signal obtained in the step S1 into a training sample and a testing sample;
s4: performing time-frequency transformation on the training sample and the test sample vibration signals obtained in the step S3 to obtain a signal time-frequency amplitude spectrum;
s5: taking the training sample time-frequency amplitude spectrum obtained in the step S4 as the input of the depth non-negative self-coding network obtained in the step S1, and pre-training each layer of self-encoders in the depth non-negative self-coding network by adopting an unsupervised layer-by-layer pre-training method;
s6: carrying out supervised fine adjustment on the pre-training depth non-negative self-encoding network obtained in the step S5 to obtain an optimized depth non-negative self-encoding network;
s7: and obtaining a mechanical equipment state identification and diagnosis result according to the depth non-negative self-coding network obtained in the step S6.
Optionally, in step S2, specifically, the method includes:
s21: in step S2, the self-encoder and the classifier are connected in a layer-by-layer stacking method to construct a depth non-negative self-encoding network, and the hidden layer output of the self-encoder at the l-th layer is used as the input of the self-encoder at the l + 1-th layer and is expressed as
Figure BDA0002497748590000021
Wherein: h is(l)Representing the hidden layer output, x, of the l-th layer self-encoder(l)And x(l+1)Respectively input to the l and l +1 layer encoders,
Figure BDA0002497748590000022
self-encoder parameters for layer I;
s22: and adding a Softmax classification layer at the hidden layer output of the last automatic encoder to form a deep non-negative self-encoding network with classification capability.
Optionally, in step S4, performing time-frequency analysis on the vibration signal acquired by the sensor by using a short-time fourier transform method to obtain a time-frequency amplitude spectrum; the short-time fourier transform STFT is defined as:
Figure BDA0002497748590000023
where γ (t) is a window function with a very short width.
Optionally, in step S5, with respect to pre-training each layer of self-encoder in the deep non-negative self-encoding network by using the training sample time-frequency amplitude spectrum, the pre-training step is as follows:
s51: the time-frequency amplitude spectrum is normalized into a column vector and an amplitude range [0,1 ];
s52: performing layer-by-layer pre-training on each layer of self-encoder in the depth non-negative self-encoding network in an unsupervised mode, adding non-negative constraints and sparse constraints to parameters of each layer of self-encoder, and constructing a cost function in a pre-training stage as follows:
Figure BDA0002497748590000031
wherein: j. the design is a squareE(W, b) are reconstruction error terms,
Figure BDA0002497748590000032
for sparse constraint terms, β for coefficients that balance the sparse constraint terms,
Figure BDA0002497748590000033
for the weight constraint terms, α for the coefficients that balance the weight constraint terms,
Figure BDA0002497748590000034
a weight parameter between the jth neuron at the l th layer and the ith neuron at the l +1 th layer is set;
s53: constructing reconstruction errors based on squared errors
Figure BDA0002497748590000035
Wherein S isfAnd SgActivation functions of an encoder and a decoder in an autoencoder, respectively, m being the number of samples, xrInputting data;
s54: the model sparsification is realized by applying sparse constraint based on Kullback-Leibler, namely KL, divergence function,
Figure BDA00024977485900000318
is defined as
Figure BDA0002497748590000036
In the formula (I), the compound is shown in the specification,
Figure BDA0002497748590000037
in order to hide the mean activation vector of the neurons,
Figure BDA0002497748590000038
if and only if
Figure BDA0002497748590000039
When the temperature of the water is higher than the set temperature,
Figure BDA00024977485900000310
s55: construct the following non-negative constraint terms
Figure BDA00024977485900000311
S56: the self-encoder parameters W and b are updated and optimized in an iterative manner using a gradient descent algorithm,
Figure BDA00024977485900000312
and
Figure BDA00024977485900000313
the formula is as follows:
Figure BDA00024977485900000314
Figure BDA00024977485900000315
wherein the content of the first and second substances,
Figure BDA00024977485900000316
for the bias term of the ith cell at layer l +1, η > 0 is the learning rate, which determines the speed of parameter update.
Optionally, in step S6, a supervised manner is adopted to perform fine tuning on the pre-training deep non-negative self-encoding network, where the supervised fine tuning step includes:
s61: solving the error classification cost of the Softmax classifier by the following formula
Figure BDA00024977485900000317
Wherein m is the number of samples, k is the number of classes, yrIs a sample xrW is the input weight matrix of all nodes in the Softmax layer, WpA pth column vector of the W matrix corresponding to the pth Softmax node input weight;
s62: solving the optimization problem of the deep nonnegative self-coding network model, and constructing a fine-tuning stage cost function as follows
Figure BDA0002497748590000041
Wherein, WDNInput weight parameters including respective encoder and Softmax classifier, bDNIncluding the offset inputs of the respective encoders,
Figure BDA0002497748590000042
is a non-negative constraint term, s, with respect to Softmax classifier input weightsLNumber of hidden layer nodes for the last auto-encoder;
s63: all layers of the depth non-negative self-coding network are regarded as a whole, and the gradient descent algorithm is adopted to update and optimize the depth non-negative self-coding network model parameters W in an iterative mode by utilizing the labeled samplesDNAnd bDN。
Optionally, in the step S7, the test sample time-frequency amplitude spectrum obtained in the step S4 is used as an input of the depth non-negative self-encoding network, so as to obtain a corresponding classification recognition node.
The invention has the beneficial effects that: the invention can well utilize vibration signal data of mechanical equipment and a structured deep non-negative self-coding network to realize the automation of self-learning from state characteristics to state recognition and diagnosis and carry out high-precision state recognition and diagnosis on different fault types and different damage degrees of the same type.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the means of the instrumentalities and combinations particularly pointed out hereinafter.
Drawings
For the purposes of promoting a better understanding of the objects, aspects and advantages of the invention, reference will now be made to the following detailed description taken in conjunction with the accompanying drawings in which:
FIG. 1 is a schematic flow chart of a deep learning state identification and diagnosis method for mechanical equipment according to the present invention;
FIG. 2 is a schematic diagram of a deep non-negative self-coding network with classifiers according to the present invention;
FIG. 3 is a multi-class confusion matrix of the state recognition result according to the embodiment of the present invention.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention in a schematic way, and the features in the following embodiments and examples may be combined with each other without conflict.
Wherein the showings are for the purpose of illustrating the invention only and not for the purpose of limiting the same, and in which there is shown by way of illustration only and not in the drawings in which there is no intention to limit the invention thereto; to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numerals in the drawings of the embodiments of the present invention correspond to the same or similar components; in the description of the present invention, it should be understood that if there is an orientation or positional relationship indicated by terms such as "upper", "lower", "left", "right", "front", "rear", etc., based on the orientation or positional relationship shown in the drawings, it is only for convenience of description and simplification of description, but it is not an indication or suggestion that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes, and are not to be construed as limiting the present invention, and the specific meaning of the terms may be understood by those skilled in the art according to specific situations.
A method for identifying and diagnosing deep learning state of mechanical equipment is shown in figure 1. FIG. 2 is a schematic diagram of a deep non-negative self-coding network with a classifier according to the present invention. Firstly, working condition information of mechanical equipment in the operation process is monitored, and vibration signal data of the equipment are obtained by using an acceleration sensor, wherein the sensor data can be vibration signals in one or more directions of an x-axis direction, a y-axis direction and a z-axis direction. And then constructing a depth non-negative self-coding network, setting a structure and initial parameters of the depth non-negative self-coding network, and performing time-frequency analysis on the vibration signals acquired by the sensor by adopting a short-time Fourier transform method to obtain a time-frequency amplitude spectrum. The method comprises the following steps:
1) the self-encoder and the classifier are connected front and back by adopting a layer-by-layer stacking method to construct a depth non-negative self-encoding network, and the hidden layer output of the self-encoder at the l-th layer is used as the input of the self-encoder at the l + 1-th layer and is expressed as
Figure BDA0002497748590000051
Wherein: h is(l)Representing the hidden layer output, x, of the l-th layer self-encoder(l)And x(l+1)Respectively input to the l and l +1 layer encoders,
Figure BDA0002497748590000052
is the l-th layer self-encoder parameter.
2) Adding a Softmax classification layer to form a depth nonnegative self-encoding network with classification capability at the hidden layer output of the last automatic encoder;
3) calculating a signal time-frequency amplitude spectrum by using short-time Fourier transform (STFT), wherein the STFT is defined as:
Figure BDA0002497748590000053
where γ (t) is a window function with a very short width.
Obtaining a depth non-negative self-coding network with a classifier and a signal time-frequency amplitude spectrum, and further pre-training each layer of self-coder in the depth non-negative self-coding network by using the training sample time-frequency amplitude spectrum, wherein the pre-training step is as follows:
1) the time-frequency amplitude spectrum is normalized into a column vector and an amplitude range [0,1 ];
2) performing layer-by-layer pre-training on each layer of self-encoder in the depth non-negative self-encoding network in an unsupervised mode, adding non-negative constraints and sparse constraints to parameters of each layer of self-encoder, and constructing a cost function in a pre-training stage as follows:
Figure BDA0002497748590000061
wherein: j. the design is a squareE(W, b) are reconstruction error terms,
Figure BDA0002497748590000062
for sparse constraint terms, β for coefficients that balance the sparse constraint terms,
Figure BDA0002497748590000063
for the weight constraint terms, α for the coefficients that balance the weight constraint terms,
Figure BDA0002497748590000064
is the weight parameter between the jth neuron at the l < th > layer and the ith neuron at the l +1 < th > layer.
3) Constructing reconstruction errors based on squared errors
Figure BDA0002497748590000065
Wherein S isfAnd SgActivation functions of an encoder and a decoder in an autoencoder, respectively, m being the number of samples, xrInputting data;
4) applying sparse constraint based on Kullback-Leibler (KL) divergence function to realize model sparseness,
Figure BDA0002497748590000066
is defined as
Figure BDA0002497748590000067
In the formula (I), the compound is shown in the specification,
Figure BDA0002497748590000068
in order to hide the mean activation vector of the neurons,
Figure BDA0002497748590000069
if and only if
Figure BDA00024977485900000610
When the temperature of the water is higher than the set temperature,
Figure BDA00024977485900000611
since it is desirable that most hidden layer neurons be "inactive", the parameter p should be as small as possible.
5) Construct the following non-negative constraint terms
Figure BDA00024977485900000612
6) The self-encoder parameters W and b are updated and optimized in an iterative manner using a gradient descent algorithm,
Figure BDA00024977485900000613
and
Figure BDA00024977485900000614
the formula is as follows:
Figure BDA00024977485900000615
Figure BDA00024977485900000616
wherein the content of the first and second substances,
Figure BDA00024977485900000617
for the bias term of the ith cell at layer l +1, η > 0 is the learning rate, which determines the speed of parameter update.
Obtaining a pre-training network, further adopting a supervision mode to carry out fine adjustment on the pre-training deep non-negative self-coding network, wherein the supervision fine adjustment comprises the following steps:
1) solving the error classification cost of the Softmax classifier by the following formula
Figure BDA0002497748590000071
Wherein m is the number of samples, k is the number of classes, yrIs a sample xrW is the input weight matrix of all nodes in the Softmax layer, WpIs the p column vector of the W matrix corresponding to the p Softmax node input weight.
2) Solving the optimization problem of the deep nonnegative self-coding network model, and constructing a fine-tuning stage cost function as follows
Figure BDA0002497748590000072
Wherein, WDNInput weight parameters including respective encoder and Softmax classifier, bDNIncluding the offset inputs of the respective encoders,
Figure BDA0002497748590000073
is a non-negative constraint term, s, with respect to Softmax classifier input weightsLThe number of hidden layer nodes for the last auto-encoder.
3) All layers of the depth non-negative self-coding network are regarded as a whole, and the gradient descent algorithm is adopted to update and optimize the depth non-negative self-coding network model parameters W in an iterative mode by utilizing the labeled samplesDNAnd bDN。
In order to verify the feasibility and accuracy of the method, test experiments are carried out, and the model is compared with a plurality of similar machine learning models. The data source is bearing vibration data measured on a bearing test platform, and the bearing vibration data comprises 8 bearings in different states (normal, 5% abrasion of a retainer, 10% abrasion of the retainer, 15% abrasion of the retainer, fracture of a middle cross beam of the retainer, pitting damage of a ball, pitting damage of an inner ring, mixed pitting damage of an inner ring and an outer ring, which are sequentially numbered as Sl-S8). In the aspect of data sampling, the rotation speed of a bearing is set to be 1000rpm, the load is 1kg of axial load, and the sampling frequency of a vibration signal is 25600Hz during data acquisition. 204800 points are respectively sampled for each state bearing, a vibration signal is intercepted according to the length of 2048 points, the overlapping rate of two adjacent sections of signals is 0.75, and 397 samples can be obtained for each state bearing.
And mixing all bearing samples in different states, randomly drawing 2400 samples (about 300 samples per bearing state) as training samples, and taking the rest samples as test samples for prediction. Fig. 3 is a multi-class confusion matrix of the state identification result, the multi-class confusion matrix includes classification information and misclassification information, and the element values on the main diagonal line of the confusion matrix sequentially represent the classification identification accuracy of Sl-S8 samples in different states. The prediction process is repeated for 5 times, and the average accuracy of the classification and identification of the samples in different states is calculated as the following table.
Figure BDA0002497748590000074
The experimental result shows that the model can effectively distinguish different types of faults, has higher identification capability for different degradation degrees of the same type, and has average identification accuracy of 97.10% for the wear state of 3 types of retainers. The correctness and the accuracy of the model in the process of identifying and diagnosing the state of the mechanical equipment are proved, and the model is suitable for monitoring and diagnosing the state of the mechanical equipment.
Finally, the above embodiments are only intended to illustrate the technical solutions of the present invention and not to limit the present invention, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions, and all of them should be covered by the claims of the present invention.

Claims (6)

1. A mechanical equipment deep learning state identification and diagnosis method is characterized in that: the method comprises the following steps:
s1: monitoring working condition information of mechanical equipment in the operation process, and acquiring equipment vibration signal data by using an acceleration sensor, wherein the sensor data are vibration signals in one or more directions of an x-axis direction, a y-axis direction and a z-axis direction so as to realize the on-line state identification and diagnosis of the mechanical equipment as a target;
s2: constructing a depth non-negative self-coding network, and setting a depth non-negative self-coding network structure and initial parameters;
s3: dividing the vibration signal obtained in the step S1 into a training sample and a testing sample;
s4: performing time-frequency transformation on the training sample and the test sample vibration signals obtained in the step S3 to obtain a signal time-frequency amplitude spectrum;
s5: taking the training sample time-frequency amplitude spectrum obtained in the step S4 as the input of the depth non-negative self-coding network obtained in the step S1, and pre-training each layer of self-encoders in the depth non-negative self-coding network by adopting an unsupervised layer-by-layer pre-training method;
s6: carrying out supervised fine adjustment on the pre-training depth non-negative self-encoding network obtained in the step S5 to obtain an optimized depth non-negative self-encoding network;
s7: and obtaining a mechanical equipment state identification and diagnosis result according to the depth non-negative self-coding network obtained in the step S6.
2. The deep learning state identification and diagnosis method for mechanical equipment according to claim 1, wherein: in step S2, specifically, the method includes:
s21: in step S2, the self-encoder and the classifier are connected in a layer-by-layer stacking method to construct a depth non-negative self-encoding network, and the hidden layer output of the self-encoder at the l-th layer is used as the input of the self-encoder at the l + 1-th layer and is expressed as
Figure FDA0002497748580000011
Wherein: h is(l)Representing the hidden layer output, x, of the l-th layer self-encoder(l)And x(l+1)Respectively input to the l and l +1 layer encoders,
Figure FDA0002497748580000012
self-encoder parameters for layer I;
s22: and adding a Softmax classification layer at the hidden layer output of the last automatic encoder to form a deep non-negative self-encoding network with classification capability.
3. The deep learning state identification and diagnosis method for mechanical equipment according to claim 1, wherein: in the step S4, a short-time fourier transform method is used to perform time-frequency analysis on the vibration signal collected by the sensor to obtain a time-frequency amplitude spectrum; the short-time fourier transform STFT is defined as:
Figure FDA0002497748580000013
where γ (t) is a window function with a very short width.
4. The deep learning state identification and diagnosis method for mechanical equipment according to claim 1, wherein: in step S5, with respect to pre-training each layer of self-encoder in the deep non-negative self-encoding network by using the training sample time-frequency amplitude spectrum, the pre-training steps are as follows:
s51: the time-frequency amplitude spectrum is normalized into a column vector and an amplitude range [0,1 ];
s52: performing layer-by-layer pre-training on each layer of self-encoder in the depth non-negative self-encoding network in an unsupervised mode, adding non-negative constraints and sparse constraints to parameters of each layer of self-encoder, and constructing a cost function in a pre-training stage as follows:
Figure FDA0002497748580000021
wherein: j. the design is a squareE(W, b) are reconstruction error terms,
Figure FDA0002497748580000022
for sparse constraint terms, β for coefficients that balance the sparse constraint terms,
Figure FDA0002497748580000023
for the weight constraint terms, α for the coefficients that balance the weight constraint terms,
Figure FDA0002497748580000024
a weight parameter between the jth neuron at the l th layer and the ith neuron at the l +1 th layer is set;
s53: constructing reconstruction errors based on squared errors
Figure FDA0002497748580000025
Wherein S isfAnd SgActivation functions of an encoder and a decoder in an autoencoder, respectively, m being the number of samples, xrInputting data;
s54: the model sparsification is realized by applying sparse constraint based on Kullback-Leibler, namely KL, divergence function,
Figure FDA0002497748580000026
is defined as
Figure FDA0002497748580000027
In the formula (I), the compound is shown in the specification,
Figure FDA0002497748580000028
in order to hide the mean activation vector of the neurons,
Figure FDA0002497748580000029
if and only if
Figure FDA00024977485800000210
When the temperature of the water is higher than the set temperature,
Figure FDA00024977485800000211
s55: construct the following non-negative constraint terms
Figure FDA00024977485800000212
S56: the self-encoder parameters W and b are updated and optimized in an iterative manner using a gradient descent algorithm,
Figure FDA00024977485800000213
and
Figure FDA00024977485800000214
the formula is as follows:
Figure FDA00024977485800000215
Figure FDA00024977485800000216
wherein the content of the first and second substances,
Figure FDA00024977485800000217
for the bias term of the ith cell at layer l +1, η > 0 is the learning rate, which determines the speed of parameter update.
5. The deep learning state identification and diagnosis method for mechanical equipment according to claim 1, wherein: in the step S6, a supervised manner is adopted to perform fine tuning on the pre-training deep non-negative self-encoding network, and the supervised fine tuning step includes:
s61: solving the error classification cost of the Softmax classifier by the following formula
Figure FDA0002497748580000031
Wherein m is the number of samples, k is the number of classes, yrIs a sample xrW is the input weight matrix of all nodes in the Softmax layer, WpA pth column vector of the W matrix corresponding to the pth Softmax node input weight;
s62: solving the optimization problem of the deep nonnegative self-coding network model, and constructing a fine-tuning stage cost function as follows
Figure FDA0002497748580000032
Wherein, WDNInput weight parameters including respective encoder and Softmax classifier, bDNIncluding the offset inputs of the respective encoders,
Figure FDA0002497748580000033
is a non-negative constraint term, s, with respect to Softmax classifier input weightsLNumber of hidden layer nodes for the last auto-encoder;
s63: all layers of the depth non-negative self-coding network are regarded as a whole, and the gradient descent algorithm is adopted to update and optimize the depth non-negative self-coding network model parameters W in an iterative mode by utilizing the labeled samplesDNAnd bDN。
6. The deep learning state identification and diagnosis method for mechanical equipment according to claim 1, wherein: in step S7, the test sample time-frequency amplitude spectrum obtained in step S4 is used as an input of the depth non-negative self-encoding network, so as to obtain a corresponding classification recognition node.
CN202010423267.7A 2020-05-19 2020-05-19 Deep learning state identification and diagnosis method for mechanical equipment Pending CN111562109A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010423267.7A CN111562109A (en) 2020-05-19 2020-05-19 Deep learning state identification and diagnosis method for mechanical equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010423267.7A CN111562109A (en) 2020-05-19 2020-05-19 Deep learning state identification and diagnosis method for mechanical equipment

Publications (1)

Publication Number Publication Date
CN111562109A true CN111562109A (en) 2020-08-21

Family

ID=72071108

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010423267.7A Pending CN111562109A (en) 2020-05-19 2020-05-19 Deep learning state identification and diagnosis method for mechanical equipment

Country Status (1)

Country Link
CN (1) CN111562109A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112132102A (en) * 2020-09-28 2020-12-25 天津工业大学 Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization
CN112347885A (en) * 2020-10-27 2021-02-09 西安科技大学 Intelligent ferrographic image identification method based on self-coding network
CN113485261A (en) * 2021-06-29 2021-10-08 西北师范大学 CAEs-ACNN-based soft measurement modeling method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868770A (en) * 2016-03-23 2016-08-17 国网山东省电力公司电力科学研究院 High-voltage circuit breaker fault diagnosis method based on unsupervised learning model
CN108805802A (en) * 2018-06-05 2018-11-13 东北大学 A kind of the front face reconstructing system and method for the stacking stepping self-encoding encoder based on constraints
CN109142976A (en) * 2018-09-10 2019-01-04 国网辽宁省电力有限公司电力科学研究院 Cable fault examination method and device
CN110779746A (en) * 2019-10-24 2020-02-11 西安理工大学 Diagnosis method for improving composite fault of deep sparse self-encoder network rotating machinery

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868770A (en) * 2016-03-23 2016-08-17 国网山东省电力公司电力科学研究院 High-voltage circuit breaker fault diagnosis method based on unsupervised learning model
CN108805802A (en) * 2018-06-05 2018-11-13 东北大学 A kind of the front face reconstructing system and method for the stacking stepping self-encoding encoder based on constraints
CN109142976A (en) * 2018-09-10 2019-01-04 国网辽宁省电力有限公司电力科学研究院 Cable fault examination method and device
CN110779746A (en) * 2019-10-24 2020-02-11 西安理工大学 Diagnosis method for improving composite fault of deep sparse self-encoder network rotating machinery

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张焱 等: "非负自编码网络基于部分特征表示的变工况滚动轴承状态识别", 《仪器仪表学报》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112132102A (en) * 2020-09-28 2020-12-25 天津工业大学 Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization
CN112132102B (en) * 2020-09-28 2024-04-19 天津工业大学 Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization
CN112347885A (en) * 2020-10-27 2021-02-09 西安科技大学 Intelligent ferrographic image identification method based on self-coding network
CN112347885B (en) * 2020-10-27 2023-06-23 西安科技大学 Self-coding network-based ferrograph image intelligent recognition method
CN113485261A (en) * 2021-06-29 2021-10-08 西北师范大学 CAEs-ACNN-based soft measurement modeling method
CN113485261B (en) * 2021-06-29 2022-06-28 西北师范大学 CAEs-ACNN-based soft measurement modeling method

Similar Documents

Publication Publication Date Title
Chen et al. Deep neural networks-based rolling bearing fault diagnosis
Wang et al. ReLTanh: An activation function with vanishing gradient resistance for SAE-based DNNs and its application to rotating machinery fault diagnosis
CN112084974B (en) Multi-label rolling bearing fault diagnosis method based on meta-learning
CN111562109A (en) Deep learning state identification and diagnosis method for mechanical equipment
CN112183581A (en) Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network
CN107066759B (en) Steam turbine rotor vibration fault diagnosis method and device
Huang et al. Application of wavelet neural networks on vibration fault diagnosis for wind turbine gearbox
Pei et al. Rotating machinery fault diagnosis through a transformer convolution network subjected to transfer learning
CN108256556A (en) Wind-driven generator group wheel box method for diagnosing faults based on depth belief network
CN112161784A (en) Mechanical fault diagnosis method based on multi-sensor information fusion migration network
Guo et al. Deep convolution feature learning for health indicator construction of bearings
Pan et al. A deep learning network via shunt-wound restricted Boltzmann machines using raw data for fault detection
Di et al. Ensemble deep transfer learning driven by multisensor signals for the fault diagnosis of bevel-gear cross-operation conditions
Jaber et al. Industrial robot backlash fault diagnosis based on discrete wavelet transform and artificial neural network
Gai et al. A Parameter‐Optimized DBN Using GOA and Its Application in Fault Diagnosis of Gearbox
Wang et al. Gear fault diagnosis based on variational modal decomposition and wide+ narrow visual field neural networks
CN114722952B (en) Bearing fault detection and classification integrated method based on representation learning
Saravanakumar et al. Hierarchical symbolic analysis and particle swarm optimization based fault diagnosis model for rotating machineries with deep neural networks
Wu et al. Design a degradation condition monitoring system scheme for rolling bearing using EMD and PCA
Zhao et al. A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition
Mousavi et al. Dictionary learning-based damage detection under varying environmental conditions using only vibration responses of numerical model and real intact State: Verification on an experimental offshore jacket model
CN112949402A (en) Fault diagnosis method for planetary gear box under minimum fault sample size
Wei et al. WSAFormer-DFFN: A model for rotating machinery fault diagnosis using 1D window-based multi-head self-attention and deep feature fusion network
CN113865872B (en) Bearing fault diagnosis method based on wavelet packet reconstruction imaging and CNN
CN109946076A (en) A kind of planet wheel bearing fault identification method of weighted multiscale dictionary learning frame

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200821