CN109858345A - A kind of intelligent failure diagnosis method suitable for pipe expanding equipment - Google Patents
A kind of intelligent failure diagnosis method suitable for pipe expanding equipment Download PDFInfo
- Publication number
- CN109858345A CN109858345A CN201811589368.0A CN201811589368A CN109858345A CN 109858345 A CN109858345 A CN 109858345A CN 201811589368 A CN201811589368 A CN 201811589368A CN 109858345 A CN109858345 A CN 109858345A
- Authority
- CN
- China
- Prior art keywords
- data
- pipe expanding
- neural network
- expanding equipment
- deep neural
- 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.)
- Granted
Links
Landscapes
- Testing And Monitoring For Control Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention belongs to pipe expanding equipment intelligent trouble diagnosis correlative technology fields, and it discloses a kind of intelligent failure diagnosis methods suitable for pipe expanding equipment, method includes the following steps: (1) acquires the pressure data of pipe expanding equipment in real time;(2) original pressure data is pre-processed, by treated, data are divided into training set and test set;(3) sparse autocoder is denoised based on the improved stack of Leacky line rectification function, constructs deep neural network fault diagnosis model, the activation primitive using Softmax function as the BP classifier of deep neural network fault diagnosis model;Deep neural network fault diagnosis model is trained using training set again, and then test set is inputted in deep neural network fault diagnosis model, deep neural network fault diagnosis model carries out diagnostic classification to test set to predict fault type, thus completes the fault diagnosis to pipe expanding equipment.The present invention improves production efficiency, reduces costs.
Description
Technical field
The invention belongs to pipe expanding equipment intelligent trouble diagnosis correlative technology fields, are suitable for more particularly, to one kind swollen
The intelligent failure diagnosis method of tube apparatus.
Background technique
It is well known that the core of air conditioner refrigerating technology includes refrigeration compressor and heat exchanger (condenser and evaporator), and
Expansion tube process as air-conditioning heat exchanger produce in critical process, it is of large quantities, it is higher to the coherence request of quality, once set
It is standby exception occur, if it find that not, often will cause bulk product quality exception, even scrap of the product;And right
Air-conditioning heat exchanger carries out being primarily present following problems in the actual production process of tube expansion process:
(1) sampling observation quality guarantee mode is not perfect.Presently mainly by the conventional means of sampling observation in advance in the quality to product
Guaranteed, but inspects method by random samples and not can guarantee the quality of all over products, and subsequent processing compensation can bring biggish manpower wave
Take.
(2) personal monitoring is not in time.It is accidentally and to happen suddenly, once equipment goes wrong, such as since equipment fault occurs
Fruit worker can not active observation product it is abnormal, then will appear bulk abnormal quality, or even cause bulk waste.
(3) pipe expanding equipment is not intelligent.Itself for tube expansion relevant device, existing equipment can not to Key Mold and
Spare and accessory parts realize automatic, intelligent real-time monitoring, to cannot handle it in time to the problems such as equipment fault, product quality.
But since most of pipe expanding mechanicals lack intelligentized status monitoring and health control module, need to pass through biography
The mode of the artificial carry out corrective maintenance of system and product sampling observation, just can guarantee properties of product, this has resulted in expansion tube process production
The failure occurred in the process, which is unable to get, timely to be handled, and serious quality conformance problem is be easy to cause.Correspondingly, this field
There is the technical needs for developing a kind of timeliness and being preferably suitable for the intelligent failure diagnosis method of pipe expanding equipment.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of intelligence suitable for pipe expanding equipment
Method for diagnosing faults is based on existing failure prediction method, studies and devises a kind of timeliness and preferably set suitable for tube expansion
Standby intelligent failure diagnosis method.The intelligent failure diagnosis method is replaced manually using the fault prediction model of pipe expanding equipment
The way of initiating to find failure and mass defect in time, can on-call maintenance and a degree of failure predication, so most
It ensure that product quality to big degree, saved human cost and cost of equipment maintenance, improved production efficiency.
To achieve the above object, the present invention provides a kind of intelligent failure diagnosis method suitable for pipe expanding equipment, the intelligence
Can method for diagnosing faults the following steps are included:
(1) signature analysis is carried out with clear failure type to pipe expanding equipment, and determines the critical component for causing failure, together
When, the pressure data of the critical component is acquired in real time;
(2) missing values cleaning is carried out to collected original pressure data, and is looked for using the outlier detection based on cluster
The error dot of original pressure data bias data race and discarding out, with the signal data after being arranged;
(3) high-frequency noise in the signal data is washed by the way of wavelet analysis, smoothly to be worked
Signal data;
(4) using 01 standardized modes will the smooth working signal hough transformation in (0,1) section, to optimize
Data structure, and obtained signal data is divided into training set and test set;
(5) sparse autocoder is denoised based on the improved stack of Leacky line rectification function and constructs deep neural network
Fault diagnosis model, the activation using Softmax function as the BP classifier of the deep neural network fault diagnosis model
Function;The deep neural network fault diagnosis model is trained using the training set again, and the test set is defeated
Enter into the trained deep neural network fault diagnosis model, the deep neural network fault diagnosis model is according to institute
It states test set and carries out diagnostic classification to predict fault type, thus complete the fault diagnosis to pipe expanding equipment.
Further, the stack denoises sparse autocoder in the training process, first to initial data XnAdd and makes an uproar, then
Allow contaminated dataReproduction is initial data Xn, so that the intermediate features vector extracted is to data under noise circumstance
Feature extraction has robustness.
Further, what step (5) used adds function of making an uproar to add function of making an uproar, expression formula for Gauss are as follows:
In formula, σ2I is the covariance matrix of the noise, and σ takes 0.05;XnFor original input data.
Further, the layer-by-layer extracting mode of feature is used in the training process of step (5), used by feature is successively extracted
Weight activation primitive is Leaky-ReLU function.
Further, the expression formula of Leaky-ReLU function are as follows:
In formula, α is constant as defined in artificial experience.
Further, in step (4), standardization processing is carried out to pressure signal data using zero-mean normalization.
Further, the mean value of data is 0, and standard deviation is the normalized conversion formula of 1, Zero-Mean are as follows:
In formula,It is the mean value of initial data;σ is the standard deviation of initial data.
Further, the denoising by the way of wavelet analysis to the signal data is the following steps are included: (a) decomposes original
Beginning pressure data;(b) the characteristics of judging pressure data is to determine high frequency threshold value;(c) weight is carried out to the signal that step (b) obtains
Structure.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, it is provided by the invention suitable
Intelligent failure diagnosis method for pipe expanding equipment mainly has the advantages that
1. denoising sparse autocoder based on the improved stack of Leacky line rectification function constructs deep neural network
Fault diagnosis model realizes pipe expanding equipment data adaptive feature extraction, reduces the need of a large amount of specialized engineering practical experiences
It asks, reduces human cost and signal analysis time, improve efficiency.
2. the activation letter using Softmax function as the BP classifier of the deep neural network fault diagnosis model
Number, is more suitable big data processing environment, improves the quality of feature extraction, further improves the recognition correct rate of model.
3. in such a way that sDSAE model is distinctive plus reproduction of making an uproar, so that repellence of the model to actual production environment noise
Enhancing, further improves the robustness of model.
4. feature successively extracts the Leaky-ReLU function that related weight activation primitive is modified to specific weight values, the letter
Number largely solves while merging non-linear factor for model training and uses Sigmoid function in conventional model
Bring gradient disappearance problem, improves the iteration speed of the feature extraction to tube expansion pressure data, reduces operation time, together
When complete the sparse expression of information, improve the generalization of model.
Detailed description of the invention
Fig. 1 is the flow diagram of the intelligent failure diagnosis method provided by the invention suitable for pipe expanding equipment.
Fig. 2 is the schematic diagram of the pipe expanding equipment being related to suitable for the intelligent failure diagnosis method of pipe expanding equipment in Fig. 1.
Fig. 3 is that the process of the data prediction being related to suitable for the intelligent failure diagnosis method of pipe expanding equipment in Fig. 1 is shown
It is intended to.
Fig. 4 is the sDSAE's being related to suitable for the intelligent failure diagnosis method of pipe expanding equipment in Fig. 1 plus the showing of mode of making an uproar
It is intended to.
Fig. 5 is the schematic network structure of the sDSAE in Fig. 4.
Fig. 6 is the Constructed wetlands figure of the deep neural network fault diagnosis model of the sDSAE in Fig. 4.
Fig. 7 is the realization schematic diagram of the deep neural network fault diagnosis model in Fig. 6.
Fig. 8 be using one embodiment of the invention provide be suitable for pipe expanding equipment intelligent failure diagnosis method acquisition with
Machine testing failure rate of correct diagnosis contrast schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
Please refer to Fig. 1, Fig. 2, Fig. 3 and Fig. 4, the intelligent failure diagnosis method provided by the invention suitable for pipe expanding equipment,
The method for diagnosing faults the following steps are included:
Step 1 carries out signature analysis to pipe expanding equipment with clear failure type, and determines the critical component for causing failure,
Meanwhile the pressure data of the critical component is acquired in real time.
Specifically, signature analysis is carried out to expansion tube process using known apparatus and process information and product report, with accurate
Judge the critical component " swollen head and expanding rod " for leading to tube expansion failure, and records related fault type.Meanwhile in the key portion
Pressure sensor is set on part to acquire pressure data in real time, and guarantees accurately to contain equipment fault hair in collected data
The information of signal intensity when raw.
Wherein, the pipe expanding equipment includes expanding rod, A is swollen head, B is swollen head, expansion set, moves back die holder, receives seat, reamer, jam plate
And copper pipe mold.By being combined to the experience of product analysis and operator after tube expansion, being made to product quality consistency
The following is summarised as at the equipment fault reason of larger impact and the fault type of generation:
(1) the inconsistent copper pipe welding quality that will cause of reaming size is bad, and especially reaming is excessive, produces fracture, then
Scrap of the product can be directly contributed.
(2) abrasion of swollen head is excessive, it will cause the tension force of copper pipe is inadequate, so that the interference fit of cooling fin and copper pipe is not
Foot, causes heat dissipation effect bad, can affect to properties of product.
(3) swollen head falls off, then directly contributes copper pipe blocking, and need replacing swollen head, influence production efficiency.
(4) expanding rod is bent, and may cause expanding rod, tube expansion fractures and copper pipe damages by pressure toward causing under a certain side pressure.
(5) expansion tube process parameter and standard is still inaccurate, and horn mouth one-pass molding low qualified, secondary operation causes working hour
Loss.
These fifth types caused in the main reason for failure, and the failure due to caused by " expanding rod " and " swollen head " self-defect accounts for greatly
Majority illustrates that expanding rod and the normal working condition of swollen head are particularly important to the quality control of tube expansion finished product, so being selected as tube expansion
Expanding rod on machine and the swollen head of described expanding rod one end is connected to as critical component, pressure sensor is arranged in tube expansion received block
On, to collect the pressure signal data of critical component.
Step 2 carries out missing values cleaning to collected original pressure data, and using the outlier inspection based on cluster
It surveys and finds out error dot and the discarding of original pressure data bias data race, with the signal data after being arranged.
Specifically, missing values processing is carried out to collected original pressure data by the way of ignoring deletion, recycled
Outlier detection based on K-Means cluster is found out in the original pressure data there is exceptional value (isolated point), these are lonely
Vertical point is likely to be caused by the quality of data there is a problem, is understood abnormal point, is further promoted the quality of data.Wherein, it isolates
The distance between point formula uses Euclidean distance (Euclidean distance):
In formula, xijIt is the idealization coordinate of a data point.Outlier detection exactly is used to find and others data not phase
Pass or weak relevant point, it is possible to the outlier far from other races is found out using clustering algorithm.
Step 3 washes the high-frequency noise in the signal data by the way of wavelet analysis, smooth to obtain
Working signal data.
Specifically, in the collection environment of signal, since there are a variety of noises, the accuracy of signal also receives certain
It influences, improves the difficulty for therefrom obtaining valuable information.Therefore, signal is handled in the way of wavelet analysis,
To eliminate the noise in pressure data, it is beneficial to improve the value and accuracy of the information got.In Wavelet Denoising Method, one
The one-dimensional data series model of a noisy pressure signal, generally are as follows:
C (x)=l (x)+ε n (x), x=0,1 ..., n-1 (2)
In formula, c (x) is to have noise cancellation signal;L (x) is desired signal;N (x) is source noise signal;ε is noise intensity.
In expansion tube process production field, l (x) is in general the more stationary signal of a low frequency or one, institute
With assume n (x) for white Gaussian noise, generally high-frequency signal.Under normal conditions, wavelet decomposition can be carried out to data-signal, and
Noise signal is different with the frequency of desired signal, therefore in analysis wavelet coefficient, utilizes the limit values gimmicks such as setting high frequency threshold value
It is filtered, small echo signal is reconstructed later, to play the role of filtering high-frequency noise.
Carrying out small echo signal denoising to the pressure data of pipe expanding equipment mainly has following steps:
(1) original pressure data is decomposed.Decomposition basic function appropriate, decomposition rank etc. are set, and to one-dimensional pressure signal
Carry out decomposition computation.
(2) the characteristics of judging pressure signal determines high frequency threshold value.The signal that different levels can be obtained after wavelet decomposition,
Different threshold values is arranged in the signal level of each high frequency, to filter a part of detail signal coefficient.
(3) reconstruct of denoised signal.The basic ideas of reconstruction signal are exactly the low frequency coefficient knot retained before utilizing
It closes processed high frequency coefficient and is characterized and carry out reconstruction signal.
Step 4, using 01 standardized modes by the smooth working signal hough transformation in (0,1) section, with
Optimize data structure, and obtained signal data is divided into training set and test set.
Specifically, standardization processing is carried out to pressure signal using zero-mean normalization, passes through this processing mode, number
According to mean value be 0, standard deviation is 1, and the normalized conversion formula of Zero-Mean are as follows:
In formula,It is the mean value of initial data;σ is the standard deviation of initial data.Input data is standardized, it can be effectively
The speed for improving e-learning, improves the accuracy of classification.
Step 5 denoises sparse autocoder (sDSAE) building based on the improved stack of Leacky line rectification function
Deep neural network fault diagnosis model, the BP using Softmax function as the deep neural network fault diagnosis model
The activation primitive of classifier;The deep neural network fault diagnosis model is trained using the training set again, and will
The test set is input in the trained deep neural network fault diagnosis model, and the deep neural network failure is examined
Disconnected model carries out diagnostic classification to the test set to predict fault type, thus completes the fault diagnosis to pipe expanding equipment.
Fig. 5, Fig. 6 and Fig. 7 are please referred to, realizes that the non-supervisory foundation structure for extracting feature of depth artificial neural network is exactly single
A autocoder (AE), each autocoder are considered as independent three-layer network model, are divided into Encoder again
With two parts Decoder.For an autocoder, its input layer is input data Xn, the target of output layer is also
Output data Xn, also just form the neural network of ' X-C-X ' formula, in order to by way of coding and decoding, go as possible
Reappear former data, to obtain the coding characteristic C of trained middle layern, thus successfully reappear former data, while illustrating Cn
The most information of former data is contained, so feature CnIt can be used as the input of next autocoder, and then realize former
Input data is to each layer of coding characteristic CnConversion, that is, the layer-by-layer extraction process of feature.
Specifically, for signal collection X collectednThe output layer constituted, autocoder utilize coding function by f
() is by XnIt is converted into middle layer value Cn, indicate are as follows:
Cn=f (Xn)
Then, by decoding functions g () to CnInverse transformation is carried out to obtain reconstruct output Yn, indicate are as follows:
Yn=g (Cn)
Finally, making the reconstruct of this layer of autocoder export Y by the training method that gradient declinesnX is inputted with original as far as possiblen
It approaches.By guaranteeing reconstructed error LnMinimum guarantee that data reappear successfully, the calculation formula of reconstructed error are as follows:
If reconstructed error is less than set desired value (minimum), just illustrate the characteristic vector C of middle layernIt can answer well
Existing original data Xn, and CnItself contain former data XnMost of information, can be by CnIt is extracted as the conversion of this layer data
Characteristic value out.
It is the former data X in autocoder training that stack, which denoises sparse autocoder (sDSAE),nIn, it is added specific
The noise signal N of distribution form, makes former input data become X 'n, but the target of output data is still Xn, so as to form " X 'n-C-
Xn" AE structure.By way of gradient decline identical with AE, so that reconstructed error LnLess than desired value (minimum), to make
Contaminated data X 'nReproduction is X as far as possiblen, so that characteristic vector C has centainly the feature extraction of data under noise circumstance
Robustness so that fault diagnosis model avoids causing diagnostic result inaccurate small noise signal tetchiness.Together
When, data point deletion is strengthened into the generalization ability of data to form sparse data structure at random using Bernoulli Jacob's distribution, into
One step improves classification accuracy, has been thusly-formed single DSAE unit.
In order to avoid single AE structure is to the deficiency of ability in feature extraction, multiple AE structures can generally be connected, it will
The feature vector C of previous AE structurenAs the input of the latter AE structure, it is constantly decoded coding, so that each layer
Feature vector CnFeature always comprising former data, it can initial data is reconstructed back, so multiple DSAE structures are accumulated
Get up, to form the profound structure that successively can downwards, constantly extract feature, which is known as stack and goes
Make an uproar autocoder (stacked Denoising Sparse Auto Encoder, sDSAE).
Include two big steps based on the deep neural network fault diagnosis model that stack denoises sparse autocoder:
The first step, the bottom of from and on layer-by-layer unsupervised pre-training (layer-wise unsuperwised pre-
Training) the stage.The stage is each layer of adjusting and optimizing its weight using DSAE unit in network, realizes data with this
Reconstruct reduces reconstructed error.Unlike shallow-layer neural network, which is successively carried out, upper one layer of DSAE unit
After selecting good characteristic vector, this feature vector is added to the network training of next layer of DASE unit, it is possible to lower
Diffusion problem caused by due to direction of error is propagated.The output of each DSAE unit is the feature representation of initial data, real
The pre-training purpose that data characteristics is extracted layer by layer is showed.
Second step, the top-down fine tuning (fune-tuning) for having supervision.For the deep neural network of classification task
For, it will after obtaining the characteristic parameter after first step pre-training, in the top layer of entire sDSAE stack network, then add one
A classifier, such as BP neural network, support vector machines, the function of this classifier are finely tuned by way of having supervision
Parameter is reduced through the error between the sDSAE initial data feature exported and objective function to be sorted, to realize point of network
Class requirement.And the trim process, both parameters of adjustable single DSAE unit, the also parameter of adjustable a network, target
Be make reality output with it is expected that output error be less than desired value.
In each DAE unit every layer of input data X can be given using noise functionnCarry out plus make an uproar, obtain it is noisy after number
According toAfterwards, then coding and decoding is carried out.
Formula (4) is that Gauss adds the function expression made an uproar, σ2I is the covariance matrix of the noise.And the first layer network is
By adding data of making an uproarFirst characteristic vector C for meeting minimum reconstructed error is calculated by first layer DAE unit1。
In formula (5), in formula, f1() is the coding function that parameter is first layer DAE.C is obtained1Later, by C1As
The input data and output target of second DAE unit, to train second characteristic vector C2, and so on, until most
The latter DAE unit terminates, and exports n-th of characteristic vector Cn。
Cn=fn(Cn-1) (6)
After n-th layer DAE model training terminates, addition BP neural network algorithm is finely adjusted network, and to feature
Vector CnClassification processing is carried out between output target O.
On=fn+1(Cn) (7)
In formula, fn+1() is coding function constructed by the parameter of output layer.Neural metwork training is finally carried out, instruction is worked as
Practice and proceed to the set training objective of satisfaction, then output data, and calculate error, thus completes deep neural network fault diagnosis mould
The training of type.
In addition, the characteristics of mining data feature unsupervised using deep neural network fault diagnosis model, it can be to expanding rod
Fault data carries out adaptive fault signature and extracts, to realize the failure predication of pipe expanding equipment.
The present invention sets up deep neural network fault diagnosis model without feature extraction using SDAE progress, is to utilize
The monitoring signals being collected into;It is transferred in the sDSAE network for building parameter, extracts characteristic vector using it;It recycles and swashs
The BP classifier that function living is softmax carries out failure modes, to realize fault diagnosis.Particularly, in sDSAE training process
In, using Leaky-ReLU function replace original sigmoid activation primitive, be deep neural network fault diagnosis model into
One step adds sparsity, and instrument accelerates the iteration speed of algorithm, Leaky-ReLU function is defined as:
In formula, α is constant smaller as defined in artificial experience.
Embodiment
Referring to Fig. 8, with one embodiment, the present invention is further described in detail below, use is collected
Based on the original data signal that electric tube expander pressure sensor measures, based on method for diagnosing faults proposed by the invention come real
The fault diagnosis of existing expansion tube process.
1. tube expansion data explanation
Using to pipe expanding equipment qualitative analysis as a result, by specific faulty equipment introduce respectively it is related to critical component
Nine kinds of malfunctions, and the special pressure sensor by being mounted on equipment received block collects under the fault condition
The pressure signal of expanding rod during tube expansion, and respective labels are sticked after being corresponded to ten classes tube expansion failure as shown in Table 1.Together
When, two parts are divided the data into, randomly select 83.3% as training set and complete the feature extraction and network struction of DNN, are remained
Remaining 16.7% sample the correctness of DNN failure modes model is calculated as test set and assess data analysis quality.
Sample label under table 1 critical component, ten kinds of malfunctions
2. tube expansion pressure data pre-processes
Firstly, carrying out missing values cleaning, deleted signal 16 in sample are deleted.Then, it is clustered using based on K-Means
The point methods that peel off, the threshold value of outlier is set as 100, only removes the gross error of data in data detection process.
Finally, the signal after being denoised using the method for Wavelet Denoising Method.
3. the training of the deep neural network fault diagnosis model based on sDSAE
As a result, using select 800 sample points as an input data of model, and according to expanding rod fault data before
Data analysis is carried out, expanding rod fault category is always divided into 10 classes, while fault category being set as to the output target of DNN;Further according to
Many experiments training experience, it is 400,200,100 that the number of nodes of DNN middle layer, which is separately designed, is extracted needed for as every layer of DAE
Characteristic vector dimension, so, the DNN failure modes model of building is the five layer network knots of " 800-400-200-100-10 "
Structure, specific parameter setting are as shown in table 2:
2 deep neural network parameter setting of table
4. deep neural network fault diagnosis model carries out fault diagnosis
After the training of deep neural network fault diagnosis model, the fault-signal data of test set are imported into depth mind
Through network fault diagnosis model to carry out fault diagnosis classification, and the accuracy of statistical forecast, as shown in figure 8, sDSAE depth mind
Rate of correct diagnosis through network fault diagnosis model and other failure modes models counts.
As can be seen from Figure 8, SVM and BP neural network construct its correlation map ability it is not strong enough, can not be at these
Too strong correlation is found in data, or easily falls into local optimum, there are problems that gradient disappearance, so point of test set
Class accuracy is only between 70% to 88%;And the classification accuracy rate of sDSAE deep neural network fault diagnosis model can be steady
Fixed arrival 97% or so, significantly larger than shallow Model, and it is also seen that sDSAE deep neural network fault diagnosis from figure
The sparse autocoder of the more traditional stack of model (sSAE) also has performance boost to a certain extent.
The above result shows that: in the processing problem of production scene of facing the reality a large amount of real time fail data collected,
Deep neural network fault diagnosis model based on sDSAE can extract fault signature and malfunction classification is incorporated in one
It rises, without separating the two, has saved manpower and time;Moreover, sDSAE deep neural network fault diagnosis model is to complexity
During data processing, solve the problems, such as the gradient disappearance and fall into part most that shallow Model (such as BP neural network) easily occurs
Excellent problem ensure that the accuracy of Data Management Analysis.Meanwhile sDSAE deep neural network fault diagnosis model is deep layer
Structure can give full play to information included in complicated field data, to complete the failure modes of high-accuracy.
Intelligent failure diagnosis method provided by the invention suitable for pipe expanding equipment, the intelligent failure diagnosis method is with reality
When the pressure signal of pipe expanding equipment that acquires based on, mentioned using the adaptive failure of depth network failure model realization data
It takes, gets rid of classical signal processing and a large amount of Engineering experience is necessarily required the drawbacks of carrying out signal processing, to have
Stronger skid resistance and accuracy.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (8)
1. a kind of intelligent failure diagnosis method suitable for pipe expanding equipment, which is characterized in that method includes the following steps:
(1) signature analysis is carried out with clear failure type to pipe expanding equipment, and determines the critical component for causing failure, meanwhile, it is real
When acquire the pressure data of the critical component;
(2) missing values cleaning is carried out to collected original pressure data, and original is found out using the outlier detection based on cluster
The error dot of beginning pressure data bias data race and discarding, with the signal data after being arranged;
(3) high-frequency noise in the signal data is washed, by the way of wavelet analysis to obtain smooth working signal number
According to;
(4) using 01 standardized modes by the smooth working signal hough transformation in (0,1) section, to optimize data
Structure, and obtained signal data is divided into training set and test set;
(5) sparse autocoder, building deep neural network event are denoised based on the improved stack of Leacky line rectification function
Hinder diagnostic model, the activation letter using Softmax function as the BP classifier of the deep neural network fault diagnosis model
Number;The deep neural network fault diagnosis model is trained using the training set again, and the test set is inputted
Into the trained deep neural network fault diagnosis model, the deep neural network fault diagnosis model is according to described in
Test set carries out diagnostic classification to predict fault type, thus completes the fault diagnosis to pipe expanding equipment.
2. being suitable for the intelligent failure diagnosis method of pipe expanding equipment as described in claim 1, it is characterised in that: the stack is gone
Sparse autocoder make an uproar in the training process, first to initial data XnAdd and make an uproar, then allows contaminated dataReproduction is original
Data Xn, so that the intermediate features vector extracted has robustness to the feature extraction of data under noise circumstance.
3. being suitable for the intelligent failure diagnosis method of pipe expanding equipment as described in claim 1, it is characterised in that: step (5) is adopted
Plus function of making an uproar is that Gauss adds function of making an uproar, expression formula are as follows:
In formula, σ2I is the covariance matrix of the noise, and σ takes 0.05;XnFor original input data.
4. being suitable for the intelligent failure diagnosis method of pipe expanding equipment as described in claim 1, it is characterised in that: step (5)
Weight activation primitive used by using the layer-by-layer extracting mode of feature, feature successively to extract in training process is Leaky-ReLU letter
Number.
5. being suitable for the intelligent failure diagnosis method of pipe expanding equipment as claimed in claim 4, it is characterised in that: Leaky-ReLU
The expression formula of function are as follows:
In formula, α is constant as defined in artificial experience.
6. the intelligent failure diagnosis method as described in any one in claim 1-5 suitable for pipe expanding equipment, it is characterised in that: step
Suddenly in (4), standardization processing is carried out to pressure signal data using zero-mean normalization.
7. being suitable for the intelligent failure diagnosis method of pipe expanding equipment as claimed in claim 6, it is characterised in that: the mean value of data
It is 0, standard deviation is the normalized conversion formula of 1, Zero-Mean are as follows:
In formula,It is the mean value of initial data;σ is the standard deviation of initial data.
8. the intelligent failure diagnosis method as described in any one in claim 1-5 suitable for pipe expanding equipment, it is characterised in that: adopt
With the mode of wavelet analysis to the denoising of the signal data the following steps are included:
(a) original pressure data is decomposed;(b) the characteristics of judging pressure data is to determine high frequency threshold value;(c) step (b) is obtained
Signal be reconstructed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811589368.0A CN109858345B (en) | 2018-12-25 | 2018-12-25 | Intelligent fault diagnosis method suitable for pipe expansion equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811589368.0A CN109858345B (en) | 2018-12-25 | 2018-12-25 | Intelligent fault diagnosis method suitable for pipe expansion equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109858345A true CN109858345A (en) | 2019-06-07 |
CN109858345B CN109858345B (en) | 2021-06-11 |
Family
ID=66892247
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811589368.0A Active CN109858345B (en) | 2018-12-25 | 2018-12-25 | Intelligent fault diagnosis method suitable for pipe expansion equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109858345B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110262417A (en) * | 2019-06-10 | 2019-09-20 | 中云开源数据技术(上海)有限公司 | A kind of production equipment failure predication diagnostic method |
CN110673556A (en) * | 2019-09-23 | 2020-01-10 | 珠海格力电器股份有限公司 | Quality control method and system for pipe expanding process |
CN111310830A (en) * | 2020-02-17 | 2020-06-19 | 湖北工业大学 | Combine harvester blocking fault diagnosis system and method |
CN112697435A (en) * | 2021-01-26 | 2021-04-23 | 山西三友和智慧信息技术股份有限公司 | Rolling bearing fault diagnosis method based on improved SELD-TCN network |
CN113848417A (en) * | 2021-08-09 | 2021-12-28 | 上海应用技术大学 | Rail transit power supply equipment fault prediction method and device |
CN116628564A (en) * | 2023-04-20 | 2023-08-22 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106443447A (en) * | 2016-09-26 | 2017-02-22 | 南京航空航天大学 | An aero-generator fault feature extraction method based on iSDAE |
CN106682688A (en) * | 2016-12-16 | 2017-05-17 | 华南理工大学 | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization |
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 |
CN108363382A (en) * | 2018-02-09 | 2018-08-03 | 哈尔滨工业大学(威海) | A kind of complex equipment fault diagnosis method and system |
CN108830127A (en) * | 2018-03-22 | 2018-11-16 | 南京航空航天大学 | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure |
-
2018
- 2018-12-25 CN CN201811589368.0A patent/CN109858345B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106443447A (en) * | 2016-09-26 | 2017-02-22 | 南京航空航天大学 | An aero-generator fault feature extraction method based on iSDAE |
CN106682688A (en) * | 2016-12-16 | 2017-05-17 | 华南理工大学 | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization |
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 |
CN108363382A (en) * | 2018-02-09 | 2018-08-03 | 哈尔滨工业大学(威海) | A kind of complex equipment fault diagnosis method and system |
CN108830127A (en) * | 2018-03-22 | 2018-11-16 | 南京航空航天大学 | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure |
Non-Patent Citations (3)
Title |
---|
代杰杰 等: "基于栈式降噪自编码器的输变电设备状态数据清洗方法", 《电力***自动化》 * |
张庆磊 等: "基于SDAE-SVM的压力机轴承故障诊断方法", 《数字制造科学》 * |
袁兆辉 等: "液压胀管机的压力测控及故障诊断", 《江苏石油化工学院学报》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110262417A (en) * | 2019-06-10 | 2019-09-20 | 中云开源数据技术(上海)有限公司 | A kind of production equipment failure predication diagnostic method |
CN110262417B (en) * | 2019-06-10 | 2020-09-01 | 中云开源数据技术(上海)有限公司 | Production equipment fault prediction and diagnosis method |
CN110673556A (en) * | 2019-09-23 | 2020-01-10 | 珠海格力电器股份有限公司 | Quality control method and system for pipe expanding process |
CN111310830A (en) * | 2020-02-17 | 2020-06-19 | 湖北工业大学 | Combine harvester blocking fault diagnosis system and method |
CN112697435A (en) * | 2021-01-26 | 2021-04-23 | 山西三友和智慧信息技术股份有限公司 | Rolling bearing fault diagnosis method based on improved SELD-TCN network |
CN112697435B (en) * | 2021-01-26 | 2022-09-09 | 山西三友和智慧信息技术股份有限公司 | Rolling bearing fault diagnosis method based on improved SELD-TCN network |
CN113848417A (en) * | 2021-08-09 | 2021-12-28 | 上海应用技术大学 | Rail transit power supply equipment fault prediction method and device |
CN116628564A (en) * | 2023-04-20 | 2023-08-22 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
CN116628564B (en) * | 2023-04-20 | 2024-03-12 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
Also Published As
Publication number | Publication date |
---|---|
CN109858345B (en) | 2021-06-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109858345A (en) | A kind of intelligent failure diagnosis method suitable for pipe expanding equipment | |
CN110262463B (en) | Rail transit platform door fault diagnosis system based on deep learning | |
US20190285517A1 (en) | Method for evaluating health status of mechanical equipment | |
CN106895975A (en) | Method for Bearing Fault Diagnosis based on Stacked SAE deep neural networks | |
CN109102005A (en) | Small sample deep learning method based on shallow Model knowledge migration | |
CN110647830B (en) | Bearing fault diagnosis method based on convolutional neural network and Gaussian mixture model | |
CN105738109A (en) | Bearing fault classification diagnosis method based on sparse representation and ensemble learning | |
CN112949715A (en) | SVM (support vector machine) -based rail transit fault diagnosis method | |
CN104832418B (en) | A kind of based on local mean value conversion and the Fault Diagnosis of Hydraulic Pump method of Softmax | |
CN111397902B (en) | Rolling bearing fault diagnosis method based on feature alignment convolutional neural network | |
CN111126819B (en) | Intelligent analysis method for urban driving condition | |
CN106271881A (en) | A kind of Condition Monitoring of Tool Breakage method based on SAEs and K means | |
CN110737976A (en) | mechanical equipment health assessment method based on multi-dimensional information fusion | |
CN112418175A (en) | Rolling bearing fault diagnosis method and system based on domain migration and storage medium | |
CN111325410A (en) | General fault early warning system based on sample distribution and early warning method thereof | |
CN114495983A (en) | Equipment failure voiceprint monitoring system based on cloud edge collaboration | |
CN113705405B (en) | Nuclear pipeline fault diagnosis method | |
CN116842460A (en) | Cough-related disease identification method and system based on attention mechanism and residual neural network | |
CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
CN112146880A (en) | Intelligent diagnosis method for internal structure faults of rolling bearing at different rotating speeds | |
CN116543154A (en) | Medical image segmentation method based on multi-level semantic features | |
CN111783941A (en) | Mechanical equipment diagnosis and classification method based on probability confidence degree convolutional neural network | |
CN114997749B (en) | Intelligent scheduling method and system for power personnel | |
CN116340812A (en) | Transformer partial discharge fault mode identification method and system | |
CN116704609A (en) | Online hand hygiene assessment method and system based on time sequence attention |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |