CN109002847A - A kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index - Google Patents
A kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index Download PDFInfo
- Publication number
- CN109002847A CN109002847A CN201810725071.6A CN201810725071A CN109002847A CN 109002847 A CN109002847 A CN 109002847A CN 201810725071 A CN201810725071 A CN 201810725071A CN 109002847 A CN109002847 A CN 109002847A
- Authority
- CN
- China
- Prior art keywords
- index
- fault
- frequency
- plunger pump
- domain
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The deepness belief network axial plunger pump Multiple faults diagnosis approach based on index that the present invention relates to a kind of.The fault mode of all kinds of known fault types is predefined first, and acquires the original vibration signal of each fault type；Secondly the energy spectrometer that the analysis of Time-domain Statistics amount, the analysis of frequency domain statistic and time-frequency domain are done to original vibration signal, obtains time domain index, frequency-domain index and time-frequency domain index, forms the achievement data collection of each fault mode；Then, the achievement data collection for constructing all known fault modes is input to the automatic study of implementation pattern feature in deepness belief network as training sample；Finally using the achievement data collection of unknown failure mode as test sample, it is input to trained deepness belief network model, realizes the pattern-recognition of the unknown failure type monitored to needs.Side of the present invention can efficiently solve the multiple faults identification problem of axial plunger pump, provide a kind of new solution for the fault diagnosis of axial plunger pump.
Description
Technical field
The invention belongs to mechanical fault diagnosis field, in particular to a kind of axis of the deepness belief network based on index
To plunger pump Multiple faults diagnosis approach.
Background technique
Hydraulic Power Transmission System occupies an important position in modern industry, is a kind of indispensable important dynamic of modern comfort
Power device, structure type are complicated, various.In recent decades, Hydraulic State Monitoring System also receives more and more attention.Axis
It is common hydraulic pump to plunger pump, according to statistics, in the failure that the mechanical equipment of Hydraulic Power Transmission System occurs, Hydraulic pump fault is most
To be common, its defect will lead to equipment downtime, result in significant economic losses even injures and deaths.Therefore, it accurately and efficiently detects
The failure for haveing hydraulic pump has become a urgent task for guaranteeing hydraulic system safe and reliable operation.
However, the working environment of axial plunger pump is complicated, fault message is easily covered by various strong background noises.Separately
Outside, research shows that in axial plunger pump the characteristic frequency of many different faults be it is identical, in this case, regardless of using more
Advanced signal processing technology, characteristic matching can not all succeed.In addition, axial plunger pump is as the crucial portion in mechanical electrical and hydraulic system
The failure mechanism of part, most of failures is indefinite, to pass through signal processing means for the fault signature of extraction and theoretical event
Barrier feature match almost impossible.
Deepness belief network (Deep Belief Networks, DBNs), is substantially a kind of artificial nerve network model,
It expresses signal with probability distribution, by constructing two layers of neural network model step by step, forms the restricted Boltzmann of stacking
Machine (Restricted Boltzmann Machine, RBM) is finally finely tuned by BP (Back Propogation) network, fine tuning
Entire hierarchical structure, to realize feature self study and pattern-recognition.However, increasing since original signal data dimension itself is big
The difficulty of deepness belief network feature self study is added.
Summary of the invention
The purpose of the invention is to overcome shortcoming and defect of the existing technology, and provide a kind of depth based on index
The axial plunger pump Multiple faults diagnosis approach for spending belief network, greatly improves the feature of deepness belief network by this method
Learning ability, and be used successfully to solve the multi-fault Diagnosis of axial plunger pump.
To achieve the above object, the technical scheme is that including:
S1: the fault mode of all kinds of axial plunger pump known fault types is predefined, and acquires the original of each fault type
Beginning vibration signal；
S2: doing the analysis of Time-domain Statistics amount, the analysis of frequency domain statistic and time-frequency domain energy spectrometer to original vibration signal,
Establish the achievement data collection of each corresponding fault mode；
S3: the fault mode of all known fault types is constructed using time domain index, frequency-domain index and time-frequency domain index
Training sample is input to the automatic study of implementation pattern feature in deepness belief network, obtains the diagnosis mould of known fault mode
Type；
S4: the achievement data collection of time domain, frequency domain and time-frequency domain is established to the fault-signal that needs monitor, constructs test specimens
This, inputs known diagnosis model, finally determines fault type.
Further setting is that the step S2 is specifically included:
(1) nine time domain indexes are calculated, it is as follows to obtain time domain index data set:
T=[I1 I2 I3 I4 I5 I6 I7 I8 I9] (1)
Wherein, I1For standard deviation requirement, I2For peak index, I3For degree of skewness index, I4For kurtosis index, I5For root mean square
It is worth index, I6For peak index, I7For margin index, I8For waveform index, I9For pulse index；
Achievement data collection in frequency domain are as follows:
F=[I10 I11 I12 I13] (2)
Wherein, I10For mean frequency value index, I11For center frequency index, I12For frequency root mean square index, I13For frequency mark
Quasi- poor index；
(2) service band wavelet package transforms calculate I14–I21To obtain energy indexes data set:
W=[I14 I15 I16 I17 I18 I19 I20 I21] (3)
Wavelet package transforms basic function are as follows: Daubieches small echo, the DB20 small echo that small echo vanishing moment is 20；I14–I21It is right
Answer the wavelet-packet energy index of the Decomposition order of different frequency bands wavelet package transforms；
(3) I is calculated using set empirical mode decomposition22–I27Obtain the energy indexes data set of 6 mode functions,
E=[I22 I23 I24 I25 I26 I27] (4)
Wherein I22–I27For the EEMD energy indexes of corresponding different data sequence；
(4) by T, W, F, E group is combined into input vector:
I=[T F W E] (5)
According to formula (5), time domain index, frequency-domain index and time-frequency domain index are obtained, the index of each fault mode is formed
Data set.
Further setting is in the step S3: the deepness belief network includes restricted Bohr of three stackings
Hereby graceful machine and an output layer, each restricted Boltzmann machine are the artificial neural networks containing two layers of neuron, the
One layer is known as visible layer, and it is mutual between each neuron of each restricted Boltzmann machine of layer that the second layer, which is known as hidden layer,
It is independent, and be then connected with each other between the neuron of interlayer.
Further setting is in the step S3:
The process of pattern feature learnt automatically is divided into two stages: the first stage is successively instructed using unsupervised mode
Practice each restricted Boltzmann machine；Second stage is carried out using back-propagation algorithm to whole network in a manner of supervision
Fine tuning.
It is an advantage of the invention that the advantages of invention is: on the one hand the method for the present invention utilizes time domain, frequency domain and time-frequency domain statistics
Analysis, establishes the achievement data collection of each fault mode, is 27 data points by analysis signal compression, and in energy stick signal
Key message；On the other hand, ability is characterized using the complicated function of deepness belief network, learns the spy of each quasi-mode automatically
Sign, and successfully identify each fault type.It is specifically shown in the embodiment of the present application.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention, for those of ordinary skill in the art, without any creative labor, according to
These attached drawings obtain other attached drawings and still fall within scope of the invention.
Flow chart Fig. 1 of the invention；
Deepness belief network axial plunger pump multi-fault Diagnosis schematic diagram of the Fig. 2 based on index；
Fig. 3 deepness belief network structure chart of the present invention；
The structure chart of the restricted Boltzmann machine of Fig. 4；
The axial plunger pump fault-signal waveform diagram of Fig. 5 embodiment of the present invention case one；
The more classification confusion matrix figures of the axial plunger failure of pump of Fig. 6 embodiment of the present invention case one；
The axial plunger pump fault-signal waveform diagram of Fig. 7 embodiment of the present invention case two；
The more classification confusion matrixes of axial plunger failure of pump of Fig. 8 embodiment of the present invention case two.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step ground detailed description.
In the embodiment of the present invention as shown in Figs. 1-2, comprising:
S1: the fault mode of all kinds of axial plunger pump known fault types is predefined, and acquires the original of each fault type
Beginning vibration signal；
S2: doing the analysis of Time-domain Statistics amount, the analysis of frequency domain statistic and time-frequency domain energy spectrometer to original vibration signal,
Establish the achievement data collection of each corresponding fault mode；
S3: the fault mode of all known fault types is constructed using time domain index, frequency-domain index and time-frequency domain index
Training sample is input to the automatic study of implementation pattern feature in deepness belief network, obtains the diagnosis mould of known fault mode
Type；
S4: the achievement data collection of time domain, frequency domain and time-frequency domain is established to the fault-signal that needs monitor, constructs test specimens
This, inputs known diagnosis model, finally determines fault type.
The technical scheme comprises the following steps for this method:
1, fault mode is carried out to all kinds of known fault types to predefine.
It is predefined that fault modes are carried out to all kinds of known fault types: establish cover one of various fault types it is complete
Event group (the unknown failure type for needing to monitor belongs to the self-contained mode) and failure definition mode.For this reason, it may be necessary to collect each
The original vibration signal of class fault type.
2, the achievement data collection of time domain, frequency domain and time-frequency domain is established to original signal.
When an error occurs, vibration signal is different from working normally the signal of pump.These variations are generally reflected at time domain, frequency
On domain and time-frequency domain.Generally, the amplitude of vibration signal and energy carry original signal information abundant.For detection axis
To the failure of plunger pump, the energy of the analysis of Time-domain Statistics amount, the analysis of frequency domain statistic and time-frequency domain is done to original vibration signal
Analysis.The index extracted from initial data is shown in Table 1.
The index and calculation formula of 1 time domain of table, frequency domain and time-frequency domain
Annotation: x (n) is a data sequence, and N is the number of data point；With σ respectively indicate data sequence mean value and
Standard deviation；S (k) is a spectrum, and K is the number of spectral line.F (k) is the frequency values of k-th of spectral line；J is that service band wavelet packet becomes
Change (WPT) i-th of decomposition coefficient sequence (i=0,1 ..., 2j-1, j are WPT Decomposition orders)；IMFiIt (n) is i-th of set warp
Data sequence after testing Mode Decomposition；NI is to utilize EEMD Decomposition order.
(1) nine time domain indexes are calculated, it is as follows to obtain time domain index data set:
T=[I1 I2 I3 I4 I5 I6 I7 I8 I9] (1)
Equally, the achievement data collection in frequency domain are as follows:
F=[I10 I11 I12 I13] (2)
(2) using WPT (basic function are as follows: Daubieches small echo, the DB20 small echo that small echo vanishing moment is 20, selection are decomposed
Number of plies j=3, subsignal number 2j=8) calculate I14-I21To obtain energy indexes data set:
W=[I14 I15 I16 I17 I18 I19 I20 I21] (3)
(3) I is calculated22-I27Obtain the energy indexes data set of 6 mode functions,
E=[I22 I23 I24 I25 I26 I27] (4)
(4) by T, W, F, E group is combined into input vector:
I=[T F W E] (5)
According to formula (5), time domain index, frequency-domain index and time-frequency domain index are obtained, the index of each fault mode is formed
Data set significantly compresses the data volume of original signal, by 27 data points of original signal boil down to, and in energy stick signal
Key message.
3, the deepness belief network intelligent diagnostics model of known fault mode is established.
Deepness belief network is a kind of neural network with multiple hidden layers.It is deep because having multiple hidden layers
Degree belief network can learn complicated function, complete data conversion and abstract by continuous learning process.Depth letter
The primary structure for reading network includes the restricted Boltzmann machine combination of a stacking.Every two layers of neural network constitutes a limitation
Property Boltzmann machine.Herein, deepness belief network is the restricted Boltzmann machine and an output layer structure stacked by three
At.
As shown in figure 3, a deepness belief network learning process is divided into two stages: the first stage is using unsupervised
Mode successively trains each restricted Boltzmann machine；Second stage is using back-propagation algorithm in a manner of supervision to whole
A network is finely adjusted.Initial data is input to input layer and removes first restricted Boltzmann machine of training.By limiting before
The feature learnt in property Boltzmann machine, remaining restricted Boltzmann machine of lasting training.Positive training process is until institute
The restricted Boltzmann machine study having is completed to terminate.The purpose of reversed adjustment process is optimize the parameter of each layer to improve entirely
The performance of deepness belief network.Restricted Boltzmann machine as a special probabilistic model in restricted Boltzmann machine,
It is the important composition unit of deepness belief network.
As shown in figure 4, each restricted Boltzmann machine is the artificial neural network containing two layers of neuron, first
Layer is known as visible layer (Visible layer), and the second layer is known as hidden layer (Hidden layer).Each restricted Bohr of layer is hereby
It is independent from each other between each neuron of graceful machine, and is then connected with each other between the neuron of interlayer.Each restricted Bohr
The hereby visible layer input vector v=[v of graceful machine1 v2 v3 … vi … vn-2 vn-1 vn] (its element is random binary amount, n
For this layer of neuron number) with the feature vector that learns(its element is random binary amount, and m is the layer
Neuron number) it is connected with each other.Pass through symmetrical weight matrix W between visible layer input vector v and the vector h of hidden layern×mMutually
It is connected.The energy function of visible layer and the joint configuration (v, h) of hidden layer neuron is as follows:
In formula (6), viIt is i-th of element in visible layer input vector v, biIt is the biasing of the visible layer；hjFor hidden layer
J-th of element of h, cjFor the biasing of the hidden layer.wijIt is weight matrix Wn×mIn element, for weight connection viAnd hj.It can
See that the joint probability distribution P (v, h | θ) of layer and hidden layer neuron is defined as:
In formula (7), Z (θ)=∑v,hExp (- E (v, h | θ)) it is partition function；θ={ bi,wij,cjIt is restricted Bohr
The hereby parameter of graceful machine probabilistic model.Since the neuron of same layer is mutually indepedent, so the conditional probability of hidden layer neuron are as follows:
Likewise, the probability of visible neuronal are as follows:
In forward direction training process, the restricted Boltzmann machine of training in a manner of greedy.In order to train each individually
Restricted Boltzmann machine, Hinton et al. proposes a kind of fast learning algorithm " to sdpecific dispersion " (Contrastive
Divergence,CD).Firstly, initial data is used to construct visible vector V as training data；Then it is calculated and is hidden with formula (8)
The conditional probability of layer neuron；Secondly according to probability is calculated, a step gibbs sampler is carried out to hidden layer neuron.Visible layer mind
It is updated through first state with formula (9), i.e. reconstruction visible layer neuron state (reconstruct data).Finally, using stochastic gradient descent
Method updates the parameter θ of restricted Boltzmann machine:
Δwij=ε (< vihj>data-<vihj>recon) (10)
Δbi=ε (< vi>data-<vi>recon) (11)
Δcj=ε (< hj>data-〈hj>recon) (12)
In formula, ε ∈ (0,1) is learning rate,<>dataIndicate the desired value of training data, < >reconIt indicates to rebuild number
According to desired value.
After carrying out pre-training to each restricted Boltzmann machine, using BP algorithm to entire depth model into
Row optimization fine tuning.In this stage, the weight of whole network is with biasing as error back propagation is constantly adjusted.
Finally, according to constructing the achievement data collection of all known fault modes, as shown in formula (5), as training sample,
By above step, it is input to the automatic study of implementation pattern feature in deepness belief network, obtains trained depth conviction net
Network model.
4, axial plunger pump multi-fault Diagnosis is realized in the pattern-recognition of unknown failure type.
To the vibration signal of the axial plunger pump of unknown failure mode, the achievement data of time domain, frequency domain and time-frequency domain is established
Collection is input to trained deepness belief network model as test sample, carries out the unknown failure type monitored to needs
Pattern-recognition finally determines fault type, realizes axial plunger pump multi-fault Diagnosis.
Case one:
Take certain Fluid Drive Experiment platform axial plunger pump fault-signal, plunger pump model are as follows: A4VSO180DR, nominal pressure
For 35Mpa, nominal discharge capacity is 180ml/r, and the theoretical flow that working shaft turns when frequency is 1500rpm is 270L/min.In present case
6 kinds of axial plunger pump fault datas are taken, fault type is respectively as follows: (1) normal operating conditions, is predefined as fault mode 1；
(2) bolt looseness is assembled, fault mode 2 is predefined as；(3) trace is ground between the pump housing and the bearing surface of valve plate, is predefined as failure
Mode 3；(4) bearing outer ring damage fault is predefined as fault mode 4；(5) bearing inner race damage fault is predefined as failure
Mode 5；(6) shaft shoulder wear-out failure is predefined as fault mode 6.Sample frequency 48k Hz.
Original signal waveform is as shown in Figure 5: fault mode 1 (a)；(b) fault mode 2；(c) fault mode 3；(d) failure
Mode 4；(e) fault mode 5；(f) fault mode 6.As seen from the figure, every a kind of fault-signal is in addition to having some differences in terms of amplitude
It is different, it is difficult to observe fault signature.Therefore, it is necessary to using the parameter data set I proposed in the present invention, to signal into one
Dimensionality reduction is walked, to learn the feature of every a kind of fault mode.
In order to further extract fault characteristic information, index is extracted to the original signal for being characterized as matrix form, is reduced
Initial data dimension.Select 360000 points as pretreated initial data, 400 data of every section of interception in sampled point
Point,.Every kind of fault mode training set chooses 240000 (600 × 400) a data points, and test set chooses 120000 (300 × 400)
A data point, i.e. every kind of fault mode have 400 training samples and 100 test samples respectively.The time domain given according to table 1,
Data compression is 27 data points by the index extraction of frequency domain and time-frequency domain, then initial data is expressed as 500 × 27 matrix
Form.Index is input to the feature that deepness belief network network learns every a kind of fault mode automatically, obtains axial plunger pump
Multistream heat exchanger result.
Fig. 6 show the classification confusion matrix of axial plunger pump multiple faults in case one.Confusion matrix gives all events
The classification results of barrier mode.The longitudinal axis indicates all physical fault types, and horizontal axis indicates to utilize the depth conviction net based on index
The fault mode result obtained after network classification.The classification of the every a kind of fault mode of element representation in confusion matrix on diagonal line is quasi-
True rate, remaining element outside diagonal entry then indicate that a certain fault mode is judged to the probability of another mode.Fig. 6 (a),
(b) classification accuracy of training sample and test sample in case one is set forth, wherein the failure in training sample is average
Classification accuracy is 100%, test sample 97.17%.As shown in Fig. 6 (a), the classification accuracy of every one kind fault mode is equal
It is 100%, the average classification accuracy of failure is also 100%.As shown in Fig. 6 (b), the classification accuracy of fault mode 1 is
94%, it is 4% and 2% that fault mode 1, which is mistaken for fault mode 2 and the probability of fault mode 5,；The classification of fault mode 2 is quasi-
True rate is 94%, and the probability that fault mode 2 is mistaken for fault mode 5 is 6%；The classification accuracy of fault mode 3 is 99%,
The probability that fault mode 3 is mistaken for fault mode 4 is 1%；The classification accuracy of fault mode 4 is 99%, 4 quilt of fault mode
The probability for being mistaken for fault mode 5 is 1%；The classification accuracy of fault mode 5 is 100%；The classification accuracy of fault mode 6
It is 97%, the probability that fault mode 6 is mistaken for fault mode 2 is 3%；The average classification accuracy of failure is 97.17%.
Case two:
Take certain Fluid Drive Experiment platform axial plunger pump fault-signal, plunger pump model are as follows: A4VSG250LR, nominal pressure
For 35Mpa, nominal discharge capacity is 250ml/r, and the theoretical flow that working shaft turns when frequency is 1500rpm is 375L/min.Axis is used in experiment
Six kinds of fault types are respectively provided with to plunger pump, is respectively as follows: (1) normal operating conditions, is predefined as fault mode 1；(2) plunger
Wear-out failure is predefined as fault mode 2；(3) cylinder body static pressure supports hole plug, is predefined as fault mode 3；(4) shaft shoulder is ground
Failure is damaged, fault mode 4 is predefined as；(5) Slipper coupling looseness fault is predefined as fault mode 5；(6) bearing outer ring is worn
Failure is predefined as fault mode 6.Sample frequency 48kHz.
Original signal waveform is as shown in Figure 7: fault mode 1 (a)；(b) fault mode 2；(c) fault mode 3；(d) failure
Mode 4；(e) fault mode 5；(f) fault mode 6.As seen from the figure, every a kind of fault-signal is in addition to having some differences in terms of amplitude
It is different, it is difficult to observe fault signature.Therefore, it is necessary to using the parameter data set I proposed in the present invention, to signal into one
Dimensionality reduction is walked, to learn the feature of every a kind of fault mode.
In order to further extract fault characteristic information, index is extracted to the original signal for being characterized as matrix form, is reduced
Initial data dimension.Select 200000 points as pretreated initial data, 400 data of every section of interception in sampled point
Point.Every kind of fault mode training set chooses 160000 (400 × 400) a data points, and it is a that test set chooses 40000 (100 × 400)
Data point, i.e. every kind of fault mode have 400 training samples and 100 test samples respectively.The time domain given according to table 1, frequency
Data compression is 27 data points by the index extraction in domain and time-frequency domain, then initial data is expressed as 500 × 27 rectangular
Formula.Index is input to the feature that deepness belief network network learns every a kind of fault mode automatically, it is more to obtain axial plunger pump
Failure modes result.
Fig. 8 show the classification confusion matrix of axial plunger pump multiple faults in case two.Fig. 8 (a), (b) are set forth
The classification accuracy of training sample and test sample in case two, wherein the classification accuracy that is averaged of the failure in training sample be
100%, test sample 97.5%.As shown in Fig. 8 (a), the classification accuracy of every one kind fault mode is 100%, failure
Average classification accuracy be also 100%.As shown in Fig. 8 (b), the classification accuracy of fault mode 1 is 90%, fault mode 1
The probability for being mistaken for fault mode 2 is 10%；The classification accuracy of fault mode 2 is 100%；The classification of fault mode 3 is quasi-
True rate is 95%, and the probability that fault mode 3 is mistaken for fault mode 4 is 5%；The classification accuracy of fault mode 4 is
100%；The classification accuracy of fault mode 5 is 100%；The classification accuracy of fault mode 6 is 100%；The average mark of failure
Class accuracy rate is 97.5%.
Those of ordinary skill in the art will appreciate that implement the method for the above embodiments be can be with
Relevant hardware is instructed to complete by program, the program can be stored in a computer readable storage medium,
The storage medium, such as ROM/RAM, disk, CD.
The above disclosure is only the preferred embodiments of the present invention, cannot limit the right model of the present invention with this certainly
It encloses, therefore equivalent changes made in accordance with the claims of the present invention, is still within the scope of the present invention.
Claims (4)
1. a kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index, characterized by comprising:
S1: predefining the fault mode of all kinds of axial plunger pump known fault types, and acquires the original vibration of each fault type
Dynamic signal；
S2: the analysis of Time-domain Statistics amount, the analysis of frequency domain statistic and time-frequency domain energy spectrometer are done to original vibration signal, established
The achievement data collection of each corresponding fault mode；
S3: the training of the fault mode of all known fault types is constructed using time domain index, frequency-domain index and time-frequency domain index
Sample is input to the automatic study of implementation pattern feature in deepness belief network, obtains the diagnostic model of known fault mode；
S4: establishing the achievement data collection of time domain, frequency domain and time-frequency domain to the fault-signal that needs monitor, and constructs test sample, defeated
Enter known diagnosis model, finally determines fault type.
2. a kind of axial plunger pump multi-fault Diagnosis side of deepness belief network based on index according to claim 1
Method, it is characterised in that the step S2 is specifically included:
(1) nine time domain indexes are calculated, it is as follows to obtain time domain index data set:
T=[I1 I2 I3 I4 I5 I6 I7 I8 I9] (1)
Wherein, I1For standard deviation requirement, I2For peak index, I3For degree of skewness index, I4For kurtosis index, I5Refer to for root-mean-square value
Mark, I6For peak index, I7For margin index, I8For waveform index, I9For pulse index；
Achievement data collection in frequency domain are as follows:
F=[I10 I11 I12 I13] (2)
Wherein, I10For mean frequency value index, I11For center frequency index, I12For frequency root mean square index, I13It is poor for frequency standard
Index；
(2) service band wavelet package transforms calculate I14–I21To obtain energy indexes data set:
W=[I14 I15 I16 I17 I18 I19 I20 I21] (3)
Wavelet package transforms basic function are as follows: Daubieches small echo, the DB20 small echo that small echo vanishing moment is 20；I14–I21To correspond to not
With the wavelet-packet energy index of the Decomposition order of frequency band wavelet package transforms；
(3) I is calculated using set empirical mode decomposition22–I27Obtain the energy indexes data set of 6 mode functions,
E=[I22 I23 I24 I25 I26 I27] (4)
Wherein I22–I27For the EEMD energy indexes of corresponding different data sequence；
(4) by T, W, F, E group is combined into input vector:
I=[T F W E] (5)
According to formula (5), time domain index, frequency-domain index and time-frequency domain index are obtained, the achievement data of each fault mode is formed
Collection.
3. a kind of axial plunger pump multi-fault Diagnosis side of deepness belief network based on index according to claim 2
Method, it is characterised in that in the step S3: the deepness belief network includes the restricted Boltzmann machine of three stackings
With an output layer, each restricted Boltzmann machine is the artificial neural network containing two layers of neuron, and first layer claims
For visible layer, the second layer is known as hidden layer, is independent from each other between each neuron of each restricted Boltzmann machine of layer,
And it is then connected with each other between the neuron of interlayer.
4. a kind of axial plunger pump multi-fault Diagnosis side of deepness belief network based on index according to claim 3
Method, it is characterised in that in the step S3:
The process of pattern feature learnt automatically is divided into two stages: the first stage is that successively training is every using unsupervised mode
One restricted Boltzmann machine；Second stage is micro- to whole network progress using back-propagation algorithm in a manner of supervision
It adjusts.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810725071.6A CN109002847A (en) | 2018-07-04 | 2018-07-04 | A kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810725071.6A CN109002847A (en) | 2018-07-04 | 2018-07-04 | A kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109002847A true CN109002847A (en) | 2018-12-14 |
Family
ID=64598789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810725071.6A Withdrawn CN109002847A (en) | 2018-07-04 | 2018-07-04 | A kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109002847A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109519340A (en) * | 2018-12-27 | 2019-03-26 | 中国船舶重工集团海装风电股份有限公司 | A kind of wind turbine generator drive system method for diagnosing faults |
| CN109946075A (en) * | 2018-12-25 | 2019-06-28 | 东北大学 | A kind of bearing condition monitoring and method for diagnosing faults |
| CN111912611A (en) * | 2020-07-10 | 2020-11-10 | 王亮 | Method and device for predicting fault state based on improved neural network |
| CN112884027A (en) * | 2021-02-02 | 2021-06-01 | 北京航空航天大学 | Cutting process real-time state monitoring method and device based on pattern recognition |
| CN112901472A (en) * | 2021-01-27 | 2021-06-04 | 赛腾机电科技（常州）有限公司 | Diagnosis method for automatically identifying plunger pump fault based on signal characteristic frequency |
| CN113361372A (en) * | 2021-06-02 | 2021-09-07 | 长江大学 | Main reducer multi-fault intelligent diagnosis method based on multi-channel data deep mining |
| CN114198295A (en) * | 2021-12-15 | 2022-03-18 | 中国石油天然气股份有限公司 | Compressor unit whole-system vibration monitoring method and device and electronic equipment thereof |
| CN114893390A (en) * | 2022-07-15 | 2022-08-12 | 安徽云磬科技产业发展有限公司 | Pump equipment fault detection method based on attention and integrated learning mechanism |
| CN115562133A (en) * | 2022-11-10 | 2023-01-03 | 浙江大学 | Intelligent gateway of axial plunger pump |
| CN115898850A (en) * | 2022-11-10 | 2023-04-04 | 浙江大学 | Axial plunger pump edge calculation processor |
| WO2023115486A1 (en) * | 2021-12-23 | 2023-06-29 | 烟台杰瑞石油服务集团股份有限公司 | Fault early-warning method and apparatus for plunger pump device and plunger pump device system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105626502A (en) * | 2016-02-01 | 2016-06-01 | 上海交通大学 | Plunger pump health assessment method based on wavelet packet and Laplacian Eigenmap |
| CN107036816A (en) * | 2016-11-17 | 2017-08-11 | 重庆工商大学 | A kind of Aero-engine Bearing method for diagnosing faults |
| CN107229269A (en) * | 2017-05-26 | 2017-10-03 | 重庆工商大学 | A kind of wind-driven generator wheel-box method for diagnosing faults of depth belief network |
| CN107917805A (en) * | 2017-10-16 | 2018-04-17 | 铜仁职业技术学院 | Fault Diagnosis of Roller Bearings based on depth belief network and support vector machines |
-
2018
- 2018-07-04 CN CN201810725071.6A patent/CN109002847A/en not_active Withdrawn
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105626502A (en) * | 2016-02-01 | 2016-06-01 | 上海交通大学 | Plunger pump health assessment method based on wavelet packet and Laplacian Eigenmap |
| CN107036816A (en) * | 2016-11-17 | 2017-08-11 | 重庆工商大学 | A kind of Aero-engine Bearing method for diagnosing faults |
| CN107229269A (en) * | 2017-05-26 | 2017-10-03 | 重庆工商大学 | A kind of wind-driven generator wheel-box method for diagnosing faults of depth belief network |
| CN107917805A (en) * | 2017-10-16 | 2018-04-17 | 铜仁职业技术学院 | Fault Diagnosis of Roller Bearings based on depth belief network and support vector machines |
Non-Patent Citations (1)
| Title |
|---|
| SHUHUI WANG ET AL.: "A data indicator-based deep belief networks to detect multiple faults in axial piston pumps", 《MECHANICAL SYSTEMS AND SIGNAL PROCESSING》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109946075A (en) * | 2018-12-25 | 2019-06-28 | 东北大学 | A kind of bearing condition monitoring and method for diagnosing faults |
| CN109519340A (en) * | 2018-12-27 | 2019-03-26 | 中国船舶重工集团海装风电股份有限公司 | A kind of wind turbine generator drive system method for diagnosing faults |
| CN111912611A (en) * | 2020-07-10 | 2020-11-10 | 王亮 | Method and device for predicting fault state based on improved neural network |
| CN112901472A (en) * | 2021-01-27 | 2021-06-04 | 赛腾机电科技（常州）有限公司 | Diagnosis method for automatically identifying plunger pump fault based on signal characteristic frequency |
| CN112884027A (en) * | 2021-02-02 | 2021-06-01 | 北京航空航天大学 | Cutting process real-time state monitoring method and device based on pattern recognition |
| CN113361372A (en) * | 2021-06-02 | 2021-09-07 | 长江大学 | Main reducer multi-fault intelligent diagnosis method based on multi-channel data deep mining |
| CN114198295A (en) * | 2021-12-15 | 2022-03-18 | 中国石油天然气股份有限公司 | Compressor unit whole-system vibration monitoring method and device and electronic equipment thereof |
| WO2023115486A1 (en) * | 2021-12-23 | 2023-06-29 | 烟台杰瑞石油服务集团股份有限公司 | Fault early-warning method and apparatus for plunger pump device and plunger pump device system |
| CN114893390A (en) * | 2022-07-15 | 2022-08-12 | 安徽云磬科技产业发展有限公司 | Pump equipment fault detection method based on attention and integrated learning mechanism |
| CN114893390B (en) * | 2022-07-15 | 2023-08-04 | 安徽云磬科技产业发展有限公司 | Pump equipment fault detection method based on attention and integrated learning mechanism |
| CN115562133A (en) * | 2022-11-10 | 2023-01-03 | 浙江大学 | Intelligent gateway of axial plunger pump |
| CN115898850A (en) * | 2022-11-10 | 2023-04-04 | 浙江大学 | Axial plunger pump edge calculation processor |
| CN115898850B (en) * | 2022-11-10 | 2024-01-26 | 浙江大学 | Edge calculation processor of axial plunger pump |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109002847A (en) | A kind of axial plunger pump Multiple faults diagnosis approach of the deepness belief network based on index | |
| CN106980822B (en) | A kind of rotary machinery fault diagnosis method based on selective ensemble study | |
| Zhao et al. | Intelligent fault diagnosis of multichannel motor–rotor system based on multimanifold deep extreme learning machine | |
| CN105550700B (en) | A kind of time series data cleaning method based on association analysis and principal component analysis | |
| Zhao et al. | A new local-global deep neural network and its application in rotating machinery fault diagnosis | |
| Dau et al. | Anomaly detection using replicator neural networks trained on examples of one class | |
| CN107122790A (en) | Non-intrusion type load recognizer based on hybrid neural networks and integrated study | |
| Zhao et al. | Multiple-order graphical deep extreme learning machine for unsupervised fault diagnosis of rolling bearing | |
| CN109186973A (en) | A kind of mechanical failure diagnostic method of unsupervised deep learning network | |
| CN109765333A (en) | A kind of Diagnosis Method of Transformer Faults based on GoogleNet model | |
| Shi et al. | Image manipulation detection and localization based on the dual-domain convolutional neural networks | |
| CN101587155A (en) | Oil soaked transformer fault diagnosis method | |
| CN109829916A (en) | A kind of Diagnosis Method of Transformer Faults based on CNN | |
| CN110399854B (en) | Rolling bearing fault classification method based on mixed feature extraction | |
| CN107067182A (en) | Towards the product design scheme appraisal procedure of multidimensional image | |
| Ma et al. | Cross-domain meta learning fault diagnosis based on multi-scale dilated convolution and adaptive relation module | |
| Wu et al. | A transformer-based approach for novel fault detection and fault classification/diagnosis in manufacturing: A rotary system application | |
| US20200393329A1 (en) | Diagnosing method of engine condition and diagnostic modeling method thereof | |
| CN106779066A (en) | A kind of radar circuit plate method for diagnosing faults | |
| Qingjun et al. | Bearing performance degradation assessment based on information-theoretic metric learning and fuzzy c-means clustering | |
| Gohar et al. | Terrorist group prediction using data classification | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| CN110318731A (en) | A kind of oil well fault diagnostic method based on GAN | |
| Jiao et al. | Partly interpretable transformer through binary arborescent filter for intelligent bearing fault diagnosis | |
| Behnam et al. | Singular Lorenz Measures Method for seizure detection using KNN-Scatter Search optimization algorithm |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20181214 |
|
| WW01 | Invention patent application withdrawn after publication |