CN109829538A - A kind of equipment health Evaluation method and apparatus based on deep neural network - Google Patents
A kind of equipment health Evaluation method and apparatus based on deep neural network Download PDFInfo
- Publication number
- CN109829538A CN109829538A CN201910150890.7A CN201910150890A CN109829538A CN 109829538 A CN109829538 A CN 109829538A CN 201910150890 A CN201910150890 A CN 201910150890A CN 109829538 A CN109829538 A CN 109829538A
- Authority
- CN
- China
- Prior art keywords
- neural network
- deep neural
- measuring point
- health status
- vibration frequency
- 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
Links
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The equipment health Evaluation method and apparatus based on deep neural network that the invention discloses a kind of.The described method includes: by Devices to test dry run under different working conditions, and obtain corresponding vibration frequency-region signal under different working condition；The vibration frequency-region signal of preset quantity is randomly selected as sample data, and uses DAE algorithm, to train preset deep neural network；Using trained deep neural network, to assess the health status of Devices to test.Method provided by the invention, the advantage of the characteristics of bonding apparatus big data and deep neural network, the identification of equipment big data fault signature extracted in self-adaptive and equipment health status can be completed at the same time, the fault message contained in health status signal spectrum can also adaptively be extracted, achieve higher equipment health evaluating precision, characteristic complicated and changeable hiding inside device data can be more characterized, when in face of complicated monitoring, diagnosing task, can more preparatively identify equipment health status.
Description
Technical field
The present invention relates to deep neural network technical field, in particular to a kind of equipment health based on deep neural network
Condition evaluation method and apparatus.
Background technique
Equipment health Evaluation is highly important, the especially hair of Internet technology in the Maintenance and Repair field of equipment
Exhibition, so that becoming increasingly important technological means using Internet technology come intelligent decision equipment health status.
Traditional equipment health Evaluation method is based on " the Feature extraction~+ machine learning model of signal processing ", the biography
The following deficiency of system method:
In terms of feature extraction, a large amount of signal processing technology is needed to be grasped in conjunction with engineering experience abundant to extract
Fault signature, and that isolates treats feature extraction and two links of health evaluating, does not consider the relationship between them；
In terms of model training, using mapping relations complicated between shallow Model exterior syndrome signal and health status, cause
When in face of equipment big data, the analysis evaluation capacity of model has obvious deficiency, is difficult to meet equipment health shape under big data background
The actual demand of condition assessment.
Summary of the invention
In order to solve problems in the prior art, it is strong that the embodiment of the invention provides a kind of equipment based on deep neural network
Health condition evaluation method and apparatus.The technical solution is as follows:
On the one hand, the equipment health Evaluation method based on deep neural network that the embodiment of the invention provides a kind of,
The described method includes:
By Devices to test dry run under different working conditions, and obtain corresponding vibration frequency under different working condition
Domain signal；
The vibration frequency-region signal of preset quantity is randomly selected as sample data, and uses noise reduction autocoder
(Denoised Auto Encoder, referred to as " DAE ") algorithm, to train preset deep neural network；
Using trained deep neural network, to assess the health status of Devices to test.
In the above-mentioned equipment health Evaluation method of the embodiment of the present invention, the vibration for randomly selecting preset quantity
Frequency-region signal uses DAE algorithm as sample data, to train preset deep neural network, comprising:
Using unsupervised learning mode and DAE algorithm is utilized, successively every layer of hidden layer of training deep neural network.
In the above-mentioned equipment health Evaluation method of the embodiment of the present invention, preset deep neural network is completed
Before training, the method also includes:
According to the different type of sample data health status, (Back Propagation, abbreviation are broadcast using error-duration model
" BP ") algorithm, the parameter of preset deep neural network is finely adjusted.
In the above-mentioned equipment health Evaluation method of the embodiment of the present invention, the preset deep neural network includes
Five layers of hidden layer, the neuron number of every layer of hidden layer respectively is 400,200,100,50 and 8 in five layers of hidden layer.
In the above-mentioned equipment health Evaluation method of the embodiment of the present invention, the Devices to test is nuclear power station CRF, institute
It states and obtains corresponding vibration frequency-region signal under different working condition, comprising:
From horizontal, vertical, axial three directions, to the default vibration measuring point acquisition vibration frequency-region signal of nuclear power station CRF, institute
Stating default vibration measuring point includes: motor anti-drive end measuring point, motor drive terminal measuring point, gear box input measuring point, planet carrier survey
Point, bull gear measuring point, output shaft measuring point, pump top chock closely drive end measuring point, pump top chock far to drive end measuring point.
On the other hand, the embodiment of the invention provides a kind of, and the equipment health Evaluation based on deep neural network fills
It sets, described device includes:
Module is obtained, for Devices to test dry run under different working conditions, and to be obtained different working condition
Under corresponding vibration frequency-region signal；
Training module, the vibration frequency-region signal for randomly selecting preset quantity are calculated as sample data, and using DAE
Method, to train preset deep neural network；
Processing module, for using trained deep neural network, to assess the health status of Devices to test.
In the above-mentioned equipment health Evaluation device of the embodiment of the present invention, the training module is also used to using nothing
Supervised learning mode simultaneously utilizes DAE algorithm, successively every layer of hidden layer of training deep neural network.
In the above-mentioned equipment health Evaluation device of the embodiment of the present invention, further includes:
Module is adjusted, for the different type according to sample data health status, using BP algorithm, to preset depth mind
Parameter through network is finely adjusted.
In the above-mentioned equipment health Evaluation device of the embodiment of the present invention, the preset deep neural network includes
Five layers of hidden layer, the neuron number of every layer of hidden layer respectively is 400,200,100,50 and 8 in five layers of hidden layer.
In the above-mentioned equipment health Evaluation device of the embodiment of the present invention, the Devices to test is nuclear power station CRF, institute
Acquisition module is stated, is also used to from horizontal, vertical, axial three directions, to the default vibration measuring point acquisition vibration frequency of nuclear power station CRF
Domain signal, the default vibration measuring point include: motor anti-drive end measuring point, motor drive terminal measuring point, gear box input measuring point,
Planet carrier measuring point, bull gear measuring point, output shaft measuring point, pump top chock closely drive end measuring point, pump top chock that end is far driven to survey
Point.
Technical solution provided in an embodiment of the present invention has the benefit that
By the way that Devices to test dry run under different working conditions, and is obtained corresponding vibration under different working condition
Dynamic frequency-region signal；The vibration frequency-region signal of preset quantity is randomly selected as sample data, and uses DAE algorithm, it is pre- to train
If deep neural network；Using trained deep neural network, to assess the health status of Devices to test.It is above-mentioned in this way to set
The advantage of the characteristics of standby health Evaluation method, bonding apparatus big data and deep neural network algorithm, can be completed at the same time
The identification of equipment big data fault signature extracted in self-adaptive and equipment health status overcomes conventional method in feature extraction and event
Deficiency in barrier identification；The fault message contained in health status signal spectrum can also be adaptively extracted, is got rid of to big
The dependence for measuring signal processing knowledge and diagnosis engineering experience, achieves higher equipment health evaluating precision；It can also more characterize and set
Hiding characteristic complicated and changeable can be identified more preparatively and be set when in face of complicated monitoring, diagnosing task inside standby data
Standby health status.In addition, according to the different type of sample data health status, using BP algorithm, to preset depth nerve net
The parameter of network is finely adjusted, in this way to the hidden layer of deep neural network using unsupervised learning training, meanwhile, and to depth mind
Parameter through network is finely tuned using supervised learning, can be complete simultaneously in this way by the combination of unsupervised learning and supervised learning
The identification of forming apparatus big data fault signature extracted in self-adaptive and equipment health status.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of equipment health Evaluation method stream based on deep neural network that the embodiment of the present invention one provides
Cheng Tu；
Fig. 2 is a kind of equipment health Evaluation device knot based on deep neural network provided by Embodiment 2 of the present invention
Structure schematic diagram.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
Embodiment one
The equipment health Evaluation method based on deep neural network that the embodiment of the invention provides a kind of is suitable for table
Characteristic complicated and changeable hiding inside device data is levied, referring to Fig. 1, this method may include:
Step S11 by Devices to test dry run under different working conditions, and is obtained corresponding under different working condition
Vibration frequency-region signal.
In the present embodiment, above equipment health state evaluation mode is (such as more using the different working condition of equipment
Kind of operating condition, various faults, normal operation etc.) under, when dry run generated vibration frequency-region signal, these vibration frequency-region signals
Characteristic complicated and changeable hiding inside device data can be more characterized, it, can be more quasi- when in face of complicated monitoring, diagnosing task
Equipment health status is identified standbyly.
Specifically, Devices to test can be nuclear power station CRF, and above-mentioned steps S11 can be accomplished in that
From horizontal, vertical, axial three directions, to the default vibration measuring point acquisition vibration frequency-region signal of nuclear power station CRF, in advance
If vibration measuring point includes: motor anti-drive end measuring point, motor drive terminal measuring point, gear box input measuring point, planet carrier measuring point, big
Gear ring measuring point, output shaft measuring point, pump top chock closely drive end measuring point, pump top chock far to drive end measuring point.
In the present embodiment, nuclear power station CRF (i.e. nuclear power station circulation), by motor, gear-box, pump group and shaft coupling
Composition, is mainly used for conventional island cold source, provides cold source to condenser and supplement heat rejecter water system by two pipe networks.From it is horizontal,
Vertically, axial three directions can fully and effectively adopt the default vibration measuring point acquisition vibration frequency-region signal of nuclear power station CRF
Collect required vibration frequency-region signal.
Step S12 randomly selects the vibration frequency-region signal of preset quantity as sample data, and uses DAE algorithm, to instruct
Practice preset deep neural network.
In the present embodiment, DAE (i.e. noise reduction autocoder) is on the basis of autocoder, and training data is added
Noise, Lai Xunlian whole network, because noise is inevitable in actual test data, using noisy training
Data train network, and neural network can learn the main feature of input feature vector and noise to not plus noise.It can make net
Network has stronger generalization ability in test data, it is of course also possible to understand are as follows: autocoder, which will learn to obtain to remove, makes an uproar
Sound obtains the ability of noise-free picture, and therefore, this just forces encoder to go to learn the more robust expression of input signal, it
Generalization ability it is also just more stronger than general encoder.The core of DAE algorithm is that coding network will contain making an uproar for certain statistical property
Sample data is added in sound, then encodes to sample, and decoding network is further according to noise statistics from undisturbed data
In estimate the primitive form for being disturbed sample.Pre-training technology of the DAE as deep neural network, can more accurate building number
According to model, facilitate the fault signature in effective excavating equipment signal.Health status signal frequency can be adaptively extracted in this way
The fault message contained in spectrum gets rid of the dependence to a large amount of signal processing knowledge and diagnosis engineering experience, achieves higher
Equipment health evaluating precision.
In practical applications, health of the CRF under various working, various faults, great amount of samples can be simulated by test
Situation substantially obtains 20350 samples, randomly chooses 25% sample for training, remaining sample is for testing.In order to reduce
The influence of enchancement factor, test repeat 15 times totally.
Specifically, above-mentioned steps S12 can be accomplished in that
Using unsupervised learning mode and DAE algorithm is utilized, successively every layer of hidden layer of training deep neural network.
In the present embodiment, the hidden layer number N for determining deep neural network, successively trains N number of DAE in unsupervised mode,
Will the hidden layer of each DAE export input as next layer of DAE, until the training of the N number of DAE of completion, and trained by using
DAE initialization deep neural network corresponds to the parameter of hidden layer.
Optionally, preset deep neural network includes five layers of hidden layer, the nerve of every layer of hidden layer in five layers of hidden layer
First number respectively is 400,200,100,50 and 8.
Step S13, according to the different type of sample data health status, using BP algorithm, to preset depth nerve net
The parameter of network is finely adjusted.
In the present embodiment, BP algorithm also makes error-duration model broadcast, and is traditional neural network (relative to deep neural network)
The training mechanism of middle use.This method calculates output of the input under current network by the initial value of random setting network parameter
As a result, then according to the difference between output and true value, using iterative algorithm to the parameter integrated regulation of whole network, Zhi Daoshou
It holds back.
The hidden layer of deep neural network is trained using unsupervised learning, meanwhile, and to the parameter of deep neural network
It is finely tuned using supervised learning, in this way by the combination of unsupervised learning and supervised learning, equipment big data can be completed at the same time
The identification of fault signature extracted in self-adaptive and equipment health status.
Step S14, using trained deep neural network, to assess the health status of Devices to test.
In the present embodiment, the characteristics of above equipment health Evaluation method, bonding apparatus big data and depth nerve
The advantage of network algorithm can be completed at the same time the identification of equipment big data fault signature extracted in self-adaptive and equipment health status,
Overcome deficiency of the conventional method in feature extraction and fault identification.
In practical applications, to the nuclear power station CRF experiment used it is found that the accuracy rates of diagnosis of 15 tests fill for 100%
Bright the present embodiment of defending oneself proposes that method can exclude operating condition interference, accurately identifies the health status under different operating conditions.In order to compare,
Using more hidden layer reverse transmittance nerve network BPNNs identical with DNN (i.e. deep neural network) structure and there is shallow structure
Method 15 times tests of single hidden layer BPNN (i.e. error use reverse neural network) assessment accuracys rate concentrate on 44.38%~
In 50.26% range, accuracy rate is integrally relatively low.For the CRF health Evaluation precision of three kinds of methods of quantitative contrast, 15 are calculated
The accuracy rate of secondary test and corresponding standard deviation, it is known that, the method highest that the present embodiment proposes, average Evaluation accuracy is
98%；Single hidden layer BPNN method is minimum, and the Evaluation accuracy that is averaged is only 42.23%；In terms of algorithm stability, the method for proposition
Most stable, accuracy rate standard deviation is 0；More hidden layer BPNN methods are most unstable, and the standard deviation of accuracy rate of diagnosis is up to
24.4%.The above result shows that: the present embodiment proposes that the Evaluation accuracy of method and compared to two kinds BPNN methods of Generalization Capability are equal
It has a clear superiority, single hidden layer BPNN method causes Evaluation accuracy limited due to using shallow Model；More hidden layer BPNN methods due to
It is trained using BP algorithm, causes the Generalization Capability of method poor.The present embodiment proposes method using deep learning to having
The neural network of depth structure is trained, so that the method for proposition has higher accuracy.
The embodiment of the present invention is by under different working conditions, and obtaining different operating shape for Devices to test dry run
Corresponding vibration frequency-region signal under state；The vibration frequency-region signal of preset quantity is randomly selected to calculate as sample data, and using DAE
Method, to train preset deep neural network；Using trained deep neural network, to assess the healthy shape of Devices to test
State.The advantage of the characteristics of above equipment health Evaluation method in this way, bonding apparatus big data and deep neural network algorithm,
It can be completed at the same time the identification of equipment big data fault signature extracted in self-adaptive and equipment health status, conventional method is overcome and exist
Deficiency in feature extraction and fault identification；The failure letter contained in health status signal spectrum can also adaptively be extracted
Breath, gets rid of the dependence to a large amount of signal processing knowledge and diagnosis engineering experience, achieves higher equipment health evaluating precision；
Characteristic complicated and changeable hiding inside device data can be also more characterized, when in face of complicated monitoring, diagnosing task, Ke Yigeng
Preparatively identify equipment health status.In addition, according to the different type of sample data health status, using BP algorithm, to default
The parameter of deep neural network be finely adjusted, in this way to the hidden layer of deep neural network using unsupervised learning training, together
When, and the parameter of deep neural network is finely tuned using supervised learning, in this way by organic knot of unsupervised learning and supervised learning
It closes, the identification of equipment big data fault signature extracted in self-adaptive and equipment health status can be completed at the same time.
Embodiment two
The embodiment of the invention provides a kind of equipment health Evaluation device based on deep neural network, performs reality
Apply method described in example one, referring to fig. 2, the apparatus may include: obtain module 100, training module 200, adjustment module 300,
Processing module 400.
Module 100 is obtained, for Devices to test dry run under different working conditions, and to be obtained different operating shape
Corresponding vibration frequency-region signal under state.
In the present embodiment, above equipment health state evaluation mode is (such as more using the different working condition of equipment
Kind of operating condition, various faults, normal operation etc.) under, when dry run generated vibration frequency-region signal, these vibration frequency-region signals
Characteristic complicated and changeable hiding inside device data can be more characterized, it, can be more quasi- when in face of complicated monitoring, diagnosing task
Equipment health status is identified standbyly.
Training module 200 for randomly selecting the vibration frequency-region signal of preset quantity as sample data, and uses DAE
Algorithm, to train preset deep neural network.
In the present embodiment, DAE (i.e. noise reduction autocoder) is on the basis of autocoder, and training data is added
Noise, Lai Xunlian whole network, because noise is inevitable in actual test data, using noisy training
Data train network, and neural network can learn the main feature of input feature vector and noise to not plus noise.It can make net
Network has stronger generalization ability in test data, it is of course also possible to understand are as follows: autocoder, which will learn to obtain to remove, makes an uproar
Sound obtains the ability of noise-free picture, and therefore, this just forces encoder to go to learn the more robust expression of input signal, it
Generalization ability it is also just more stronger than general encoder.The core of DAE algorithm is that coding network will contain making an uproar for certain statistical property
Sample data is added in sound, then encodes to sample, and decoding network is further according to noise statistics from undisturbed data
In estimate the primitive form for being disturbed sample.Pre-training technology of the DAE as deep neural network, can more accurate building number
According to model, facilitate the fault signature in effective excavating equipment signal.Health status signal frequency can be adaptively extracted in this way
The fault message contained in spectrum gets rid of the dependence to a large amount of signal processing knowledge and diagnosis engineering experience, achieves higher
Equipment health evaluating precision.
Module 300 is adjusted, for the different type according to sample data health status, using BP algorithm, to preset depth
The parameter of degree neural network is finely adjusted.
In the present embodiment, BP algorithm also makes error-duration model broadcast, and is traditional neural network (relative to deep neural network)
The training mechanism of middle use.This method calculates output of the input under current network by the initial value of random setting network parameter
As a result, then according to the difference between output and true value, using iterative algorithm to the parameter integrated regulation of whole network, Zhi Daoshou
It holds back.
The hidden layer of deep neural network is trained using unsupervised learning, meanwhile, and to the parameter of deep neural network
It is finely tuned using supervised learning, in this way by the combination of unsupervised learning and supervised learning, equipment big data can be completed at the same time
The identification of fault signature extracted in self-adaptive and equipment health status.
Processing module 400, for using trained deep neural network, to assess the health status of Devices to test.
In the present embodiment, the characteristics of above equipment health Evaluation method, bonding apparatus big data and depth nerve
The advantage of network algorithm can be completed at the same time the identification of equipment big data fault signature extracted in self-adaptive and equipment health status,
Overcome deficiency of the conventional method in feature extraction and fault identification.
Specifically, Devices to test can be nuclear power station CRF, obtain module 100, be also used to from horizontal, vertical, three axial
Direction, to the default vibration measuring point acquisition vibration frequency-region signal of nuclear power station CRF, presetting vibration measuring point includes: motor anti-drive end
Measuring point, motor drive terminal measuring point, gear box input measuring point, planet carrier measuring point, bull gear measuring point, output shaft measuring point, axis on pump
Holding seat closely drives end measuring point, pump top chock far to drive end measuring point.
In the present embodiment, nuclear power station CRF (i.e. nuclear power station circulation), by motor, gear-box, pump group and shaft coupling
Composition, is mainly used for conventional island cold source, provides cold source to condenser and supplement heat rejecter water system by two pipe networks.From it is horizontal,
Vertically, axial three directions can fully and effectively adopt the default vibration measuring point acquisition vibration frequency-region signal of nuclear power station CRF
Collect required vibration frequency-region signal.
In practical applications, health of the CRF under various working, various faults, great amount of samples can be simulated by test
Situation substantially obtains 20350 samples, randomly chooses 25% sample for training, remaining sample is for testing.In order to reduce
The influence of enchancement factor, test repeat 15 times totally.
Optionally, training module 200 are also used to using unsupervised learning mode and utilize DAE algorithm, successively train depth
Every layer of hidden layer of neural network.
In the present embodiment, the hidden layer number N for determining deep neural network, successively trains N number of DAE in unsupervised mode,
Will the hidden layer of each DAE export input as next layer of DAE, until the training of the N number of DAE of completion, and trained by using
DAE initialization deep neural network corresponds to the parameter of hidden layer.
Optionally, preset deep neural network includes five layers of hidden layer, the nerve of every layer of hidden layer in five layers of hidden layer
First number respectively is 400,200,100,50 and 8.
In practical applications, to the nuclear power station CRF experiment used it is found that the accuracy rates of diagnosis of 15 tests fill for 100%
Bright the present embodiment of defending oneself proposes that method can exclude operating condition interference, accurately identifies the health status under different operating conditions.In order to compare,
Use more hidden layer reverse transmittance nerve network BPNNs identical with DNN structure and single hidden layer BPNN method with shallow structure
The assessment accuracy rate of 15 tests concentrates in 44.38%~50.26% range, and accuracy rate is integrally relatively low.For quantitative contrast
The CRF health Evaluation precision of three kinds of methods calculates the accuracy rate and corresponding standard deviation of 15 tests, it is known that, this reality
The method highest of example proposition is applied, the Evaluation accuracy that is averaged is 98%；Single hidden layer BPNN method is minimum, and average Evaluation accuracy is only
42.23%；In terms of algorithm stability, the method for proposition is most stable, and accuracy rate standard deviation is 0；More hidden layer BPNN methods are most
Unstable, the standard deviation of accuracy rate of diagnosis is up to 24.4%.The above result shows that: the Evaluation accuracy of the present embodiment proposition method
It has a clear superiority with compared to two kinds BPNN methods of Generalization Capability, single hidden layer BPNN method causes due to using shallow Model
Evaluation accuracy is limited；More hidden layer BPNN methods cause the Generalization Capability of method poor due to using BP algorithm to be trained.This reality
It applies example and proposes that method is trained the neural network with depth structure using deep learning, so that the method for proposition is with higher
Accuracy.
The embodiment of the present invention is by under different working conditions, and obtaining different operating shape for Devices to test dry run
Corresponding vibration frequency-region signal under state；The vibration frequency-region signal of preset quantity is randomly selected to calculate as sample data, and using DAE
Method, to train preset deep neural network；Using trained deep neural network, to assess the healthy shape of Devices to test
State.The advantage of the characteristics of above equipment health Evaluation method in this way, bonding apparatus big data and deep neural network algorithm,
It can be completed at the same time the identification of equipment big data fault signature extracted in self-adaptive and equipment health status, conventional method is overcome and exist
Deficiency in feature extraction and fault identification；The failure letter contained in health status signal spectrum can also adaptively be extracted
Breath, gets rid of the dependence to a large amount of signal processing knowledge and diagnosis engineering experience, achieves higher equipment health evaluating precision；
Characteristic complicated and changeable hiding inside device data can be also more characterized, when in face of complicated monitoring, diagnosing task, Ke Yigeng
Preparatively identify equipment health status.In addition, according to the different type of sample data health status, using BP algorithm, to default
The parameter of deep neural network be finely adjusted, in this way to the hidden layer of deep neural network using unsupervised learning training, together
When, and the parameter of deep neural network is finely tuned using supervised learning, in this way by organic knot of unsupervised learning and supervised learning
It closes, the identification of equipment big data fault signature extracted in self-adaptive and equipment health status can be completed at the same time.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
It should be understood that the equipment health Evaluation device provided by the above embodiment based on deep neural network exists
When realizing the equipment health Evaluation method based on deep neural network, only illustrated with the division of above-mentioned each functional module
Illustrate, in practical application, can according to need and be completed by different functional modules above-mentioned function distribution, i.e., it will be in equipment
Portion's structure is divided into different functional modules, to complete all or part of the functions described above.In addition, above-described embodiment mentions
The equipment health Evaluation device based on deep neural network supplied is commented with the equipment health status based on deep neural network
Estimate embodiment of the method and belong to same design, specific implementation process is detailed in embodiment of the method, and which is not described herein again.
Those of ordinary skill in the art will appreciate that realizing that all or part of the steps of above-described embodiment can pass through hardware
It completes, relevant hardware can also be instructed to complete by program, the program can store in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only memory, disk or CD etc..
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of equipment health Evaluation method based on deep neural network, which is characterized in that the described method includes:
By Devices to test dry run under different working conditions, and obtain corresponding vibration frequency domain letter under different working condition
Number；
The vibration frequency-region signal of preset quantity is randomly selected as sample data, and uses DAE algorithm, to train preset depth
Neural network；
Using trained deep neural network, to assess the health status of Devices to test.
2. the method according to claim 1, wherein the vibration frequency-region signal for randomly selecting preset quantity is made
For sample data, and DAE algorithm is used, to train preset deep neural network, comprising:
Using unsupervised learning mode and DAE algorithm is utilized, successively every layer of hidden layer of training deep neural network.
3. according to the method described in claim 2, it is characterized in that, before the training for completing preset deep neural network,
The method also includes:
The parameter of preset deep neural network is carried out using BP algorithm according to the different type of sample data health status
Fine tuning.
4. according to the method described in claim 3, it is characterized in that, the preset deep neural network includes five layers hiding
Layer, the neuron number of every layer of hidden layer respectively is 400,200,100,50 and 8 in five layers of hidden layer.
5. the acquisition is different the method according to claim 1, wherein the Devices to test is nuclear power station CRF
Corresponding vibration frequency-region signal under working condition, comprising:
It is described pre- to the default vibration measuring point acquisition vibration frequency-region signal of nuclear power station CRF from horizontal, vertical, axial three directions
If vibration measuring point includes: motor anti-drive end measuring point, motor drive terminal measuring point, gear box input measuring point, planet carrier measuring point, big
Gear ring measuring point, output shaft measuring point, pump top chock closely drive end measuring point, pump top chock far to drive end measuring point.
6. a kind of equipment health Evaluation device based on deep neural network characterized by comprising
Obtain module, for by Devices to test dry run under different working conditions, and obtain it is right under different working condition
The vibration frequency-region signal answered；
Training module, the vibration frequency-region signal for randomly selecting preset quantity are come as sample data, and using DAE algorithm
The preset deep neural network of training；
Processing module, for using trained deep neural network, to assess the health status of Devices to test.
7. device according to claim 6, which is characterized in that the training module is also used to using unsupervised learning side
Formula simultaneously utilizes DAE algorithm, successively every layer of hidden layer of training deep neural network.
8. device according to claim 7, which is characterized in that further include:
Module is adjusted, for the different type according to sample data health status, using BP algorithm, to preset depth nerve net
The parameter of network is finely adjusted.
9. device according to claim 8, which is characterized in that the preset deep neural network includes five layers and hides
Layer, the neuron number of every layer of hidden layer respectively is 400,200,100,50 and 8 in five layers of hidden layer.
10. device according to claim 6, which is characterized in that the Devices to test is nuclear power station CRF, the acquisition mould
Block, is also used to from horizontal, vertical, axial three directions, vibrates frequency-region signal to the default vibration measuring point acquisition of nuclear power station CRF,
The default vibration measuring point includes: motor anti-drive end measuring point, motor drive terminal measuring point, gear box input measuring point, planet carrier
Measuring point, bull gear measuring point, output shaft measuring point, pump top chock closely drive end measuring point, pump top chock far to drive end measuring point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910150890.7A CN109829538A (en) | 2019-02-28 | 2019-02-28 | A kind of equipment health Evaluation method and apparatus based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910150890.7A CN109829538A (en) | 2019-02-28 | 2019-02-28 | A kind of equipment health Evaluation method and apparatus based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109829538A true CN109829538A (en) | 2019-05-31 |
Family
ID=66864885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910150890.7A Pending CN109829538A (en) | 2019-02-28 | 2019-02-28 | A kind of equipment health Evaluation method and apparatus based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829538A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680806A (en) * | 2020-06-11 | 2020-09-18 | 河南财政金融学院 | Building equipment management system |
| CN112686506A (en) * | 2020-12-18 | 2021-04-20 | 海南电网有限责任公司电力科学研究院 | Distribution network equipment comprehensive evaluation method based on multi-test method asynchronous detection data |
| CN113591792A (en) * | 2021-08-19 | 2021-11-02 | 国网吉林省电力有限公司四平供电公司 | Transformer fault identification method based on self-organizing competitive neural network algorithm |
| CN113742638A (en) * | 2021-08-30 | 2021-12-03 | 南通大学 | Kurtosis-based STLBO motor bearing fault diagnosis method based on FastICA and approximation solution domain |
| CN114684217A (en) * | 2022-03-16 | 2022-07-01 | 武汉理工大学 | Rail transit health monitoring system and method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016132468A1 (en) * | 2015-02-18 | 2016-08-25 | 株式会社日立製作所 | Data evaluation method and device, and breakdown diagnosis method and device |
| CN107957551A (en) * | 2017-12-12 | 2018-04-24 | 南京信息工程大学 | Stacking noise reduction own coding Method of Motor Fault Diagnosis based on vibration and current signal |
| CN108062572A (en) * | 2017-12-28 | 2018-05-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on DdAE deep learning models |
-
2019
- 2019-02-28 CN CN201910150890.7A patent/CN109829538A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016132468A1 (en) * | 2015-02-18 | 2016-08-25 | 株式会社日立製作所 | Data evaluation method and device, and breakdown diagnosis method and device |
| CN107957551A (en) * | 2017-12-12 | 2018-04-24 | 南京信息工程大学 | Stacking noise reduction own coding Method of Motor Fault Diagnosis based on vibration and current signal |
| CN108062572A (en) * | 2017-12-28 | 2018-05-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on DdAE deep learning models |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680806A (en) * | 2020-06-11 | 2020-09-18 | 河南财政金融学院 | Building equipment management system |
| CN112686506A (en) * | 2020-12-18 | 2021-04-20 | 海南电网有限责任公司电力科学研究院 | Distribution network equipment comprehensive evaluation method based on multi-test method asynchronous detection data |
| CN112686506B (en) * | 2020-12-18 | 2022-06-17 | 海南电网有限责任公司电力科学研究院 | Distribution network equipment comprehensive evaluation method based on multi-test method asynchronous detection data |
| CN113591792A (en) * | 2021-08-19 | 2021-11-02 | 国网吉林省电力有限公司四平供电公司 | Transformer fault identification method based on self-organizing competitive neural network algorithm |
| CN113591792B (en) * | 2021-08-19 | 2023-11-28 | 国网吉林省电力有限公司四平供电公司 | Transformer fault identification method based on self-organizing competitive neural network algorithm |
| CN113742638A (en) * | 2021-08-30 | 2021-12-03 | 南通大学 | Kurtosis-based STLBO motor bearing fault diagnosis method based on FastICA and approximation solution domain |
| CN114684217A (en) * | 2022-03-16 | 2022-07-01 | 武汉理工大学 | Rail transit health monitoring system and method |
| CN114684217B (en) * | 2022-03-16 | 2024-03-01 | 武汉理工大学 | Rail transit health monitoring system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829538A (en) | A kind of equipment health Evaluation method and apparatus based on deep neural network | |
| CN109000930B (en) | Turbine engine performance degradation evaluation method based on stacking denoising autoencoder | |
| CN113935460B (en) | Intelligent diagnosis method for mechanical faults under unbalanced-like data set | |
| CN106408088B (en) | A kind of rotating machinery method for diagnosing faults based on deep learning theory | |
| CN110702411B (en) | Residual error network rolling bearing fault diagnosis method based on time-frequency analysis | |
| CN105160678A (en) | Convolutional-neural-network-based reference-free three-dimensional image quality evaluation method | |
| CN101303764B (en) | Method for self-adaption amalgamation of multi-sensor image based on non-lower sampling profile wave | |
| CN104954778B (en) | Objective stereo image quality assessment method based on perception feature set | |
| CN110033417A (en) | A kind of image enchancing method based on deep learning | |
| CN105208374A (en) | Non-reference image quality objective evaluation method based on deep learning | |
| CN108398268A (en) | A kind of bearing performance degradation assessment method based on stacking denoising self-encoding encoder and Self-organizing Maps | |
| CN108305619A (en) | Voice data collection training method and apparatus | |
| CN111429931B (en) | Noise reduction model compression method and device based on data enhancement | |
| CN113379601A (en) | Real world image super-resolution method and system based on degradation variational self-encoder | |
| CN114970628B (en) | Rotating part fault diagnosis method based on generation countermeasure network under condition of unbalanced samples | |
| CN116417013A (en) | Underwater propeller fault diagnosis method and system | |
| CN104767993B (en) | A kind of stereoscopic video objective quality evaluation based on matter fall time domain weighting | |
| CN102609764A (en) | CPN neural network-based fault diagnosis method for stream-turbine generator set | |
| CN106156877A (en) | Predict the drive method of risk, Apparatus and system | |
| CN115808312A (en) | Rolling bearing fault diagnosis method based on deformable volume and Transformer | |
| CN116399588A (en) | Rolling bearing fault diagnosis method based on WPD and AFRB-LWUNet under small sample | |
| Xu et al. | Deep noise suppression with non-intrusive pesqnet supervision enabling the use of real training data | |
| CN113673397A (en) | Local area adaptive mechanical fault diagnosis method based on class weighting alignment | |
| CN110160778A (en) | Gearbox fault state identification method based on sequential hypothesis testing | |
| CN117725419A (en) | Small sample unbalanced rotor fault diagnosis method and 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190531 |
|
| RJ01 | Rejection of invention patent application after publication |