CN109597315A - A kind of mechanical equipment health degenerate state discrimination method, equipment and system - Google Patents
A kind of mechanical equipment health degenerate state discrimination method, equipment and system Download PDFInfo
- Publication number
- CN109597315A CN109597315A CN201811281339.8A CN201811281339A CN109597315A CN 109597315 A CN109597315 A CN 109597315A CN 201811281339 A CN201811281339 A CN 201811281339A CN 109597315 A CN109597315 A CN 109597315A
- Authority
- CN
- China
- Prior art keywords
- mechanical equipment
- monitoring signals
- training
- variety
- degenerate state
- 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
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B17/00—Systems involving the use of models or simulators of said systems
- G05B17/02—Systems involving the use of models or simulators of said systems electric
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a kind of mechanical equipment health degenerate state discrimination method, equipment and system, belongs to machine performance monitoring and healthy degenerate state recognizes field.This method first obtains a variety of monitoring signals of mechanical equipment, extract its temporal signatures, power spectrum characteristic and intrinsic mode energy feature, then features described above is merged using GSOM network, obtain its fusion feature, GenSVM model is constructed again, GenSVM model is trained using above-mentioned fusion feature data, obtain test model, a variety of monitoring signals of mechanical equipment to be measured are finally acquired in real time, above-mentioned signal fusion feature data are input in test model is obtained, healthy degenerate state identification result is obtained.The above method of the invention and the equipment based on the above method and system can carry out accurate identification in real time to the healthy degenerate state of mechanical equipment, realize the status real time monitor of mechanical equipment, ensure safe and stable, the long-term operation of numerically-controlled machine tool.
Description
Technical field
The invention belongs to machine performance monitorings and healthy degenerate state identification technique field, and in particular to one kind is based on growth
Self-organizing Maps (growing self-organizing maps, GSOM) network and broad sense multi-category support vector machines
The mechanical health degeneration monitoring method of (Generalized multiclass support vector machine, GenSVM),
Equipment and system.
Background technique
Mechanical equipment produces the core production equipment in industry as many processes such as manufacture, metallurgy, chemical industry, environmental protection, it
Operation quality directly affect occur in the production efficiency, product quality, maintenance cost of enterprise, especially equipment operation it is prominent
Hair property failure and safety accident, often lead to the interruption of entire production procedure and cause great production loss.Therefore, it was manufacturing
It is most important that real-time identification is carried out to the healthy degraded condition of mechanical equipment in journey.
Support vector machines (support vector machine, SVM) algorithm is with a kind of health of common mechanical equipment
Degraded condition discrimination method, theory are put forward by Vapnik, because it is suitble to small sample environment, theory to be easily understood, have
Preferable robustness is had the advantages that higher generalization ability, is mapped using inner product kernel function instead of high dimensional nonlinear, is widely applied
In the health status monitoring of mechanized equipment, fault diagnosis and life prediction field.However it is distinguished in mechanical equipment health degenerate state
When knowledge, single SVM model be only applicable to two kinds of degenerate states identification scene, can not a variety of degenerate states of progressive identification.
Although the identification that mechanical equipment carries out a variety of degenerate states may be implemented using more SVM built-up patterns, there is complicated antithesis
Optimization problem.
Therefore, the scheme of a kind of simple and easy to do and energy real-time identification mechanical equipment healthy degraded condition is needed.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of mechanical equipment health degenerate states
Discrimination method, it is intended that passing through a variety of monitoring signals for obtaining mechanical equipment to be measured in real time and based on GSOM and GenSVM
Above-mentioned signal is merged and trained respectively, and then realizes the on-line identification of mechanical equipment health degenerate state.
To achieve the goals above, the present invention provides a kind of mechanical equipment health degenerate state discrimination methods, including instruction
Practice stage and test phase:
Training stage includes the following steps:
Step 1: obtaining a variety of monitoring signals of mechanical equipment, singular value processing and noise reduction are carried out to a variety of monitoring signals
Processing；
Step 2: to step 1, treated that each monitoring signals carry out time-domain analysis, extracts the time domain of each monitoring signals respectively
Feature, comprising: average value, mean square deviation, root amplitude, root-mean-square value, maximum value, flexure index, kurtosis index, peak value
The factor and the nargin factor；
To step 1, treated that each monitoring signals carry out power spectrumanalysis, and the power spectrum for extracting each monitoring signals respectively is special
Sign, comprising: center frequency, square frequency, root mean square frequency, frequency variance and frequency root variance；
To step 1, treated that each monitoring signals carry out CEEMDAN decomposition, obtains each intrinsic mode of each monitoring signals
Component calculates intrinsic mode energy feature of the energy value as mechanical equipment health degenerate state of each modal components；
Step 3: utilize GSOM network, the temporal signatures of each monitoring signals that step 2 is obtained respectively, power spectrum characteristic and
Intrinsic mode energy feature is merged, and Time Domain Fusion feature, power spectrum fusion feature and the eigen mode of each monitoring signals are obtained
State energy fusion feature；
Step 4: each fusion feature of each monitoring signals obtained using step 3 is trained GenSVM model；
Test phase includes the following steps:
Step 5: for a variety of monitoring signals of mechanical equipment acquired in real time, being extracted respectively according to step 2 and step 3 a variety of
After the feature of monitoring signals and respectively progress Fusion Features, it is input in the trained GenSVM model of step 4, secures good health and move back
Change state identification result.
Further, in step 3 using GSOM network to temporal signatures, power spectrum characteristic and intrinsic mode energy feature into
The process of row fusion is as follows:
Step 3.1: choose mechanical equipment part normal condition under a variety of monitoring signals data above-mentioned three kinds of features it
One sets up training dataset Dn；
Step 3.2: building GSOM network model, initiation parameter；
Step 3.3: inputting training sample D to GSOM network modeln, and set growing threshold distance dmax；
Step 3.4: random all temporal signatures x without the mechanical equipment for repeatedly choosing a certain moment from training dataset
(t)∈Dn；According to being triumph neuron principle with the neuron of x (t) minimum euclidean distance, i.e.,Determine the triumph neuron in GSOM network model；
Step 3.5: judging whether to need to grow new neuron；If oi(x)(x) > dmax, then Exist Network Structure is insufficient
To describe the feature of input data, needs to increase a new neuron in GSOM network model, continue step 3.6；Otherwise, it returns
Return step 3.4；
Step 3.6: setting the feedforward connection weight of new neuron as x (t)T, new neuron is calculated to other all nerves
The distance of member, in distance to a declared goal range (a1×dmax, a2×dmax) in find the adjoining neuron of new neuron, establish adjacent close
System；Wherein, a1< a2, a1And a2For preset range parameter；
Step 3.7: step 3.4 to 3.6 is repeated, until the complete D of trainingnIn after all data, obtain trained GSOM net
Network model；
Step 3.8: it is trained that a variety of monitoring signals data to be fused being sequentially input into step 3.7 sequentially in time
Characteristic value in GSOM network model, after respectively obtaining each detection signal fused.
Further, the training process of GenSVM is as follows in step 4:
Step 4.1: setting up performance degradation eigenmatrix F using the fusion feature data of a variety of monitoring signals of mechanical equipment；
F=[F1, F2, F3, L]
Wherein, F1、F2And F3It is special to respectively indicate Time Domain Fusion feature, power spectrum fusion feature and the fusion of intrinsic mode energy
Sign, and F1, F2, F3∈Rp×k, p is sampling number, and k indicates monitoring signals species number, Rp×kIndicate the real number matrix of p × k dimension；L
Indicate capability of engineering equipment decay state label, L ∈ Rp×1；
Step 4.2: performance degradation eigenmatrix F is randomly divided into training sample and verification sample；
Step 4.3: training sample being inputted into GenSVM model, the kernel function in GenSVM model is by the training sample of input
In data be mapped to higher dimensional space, be expressed as xi→Φi；
Step 4.4: by ФiIt is indicated in K-1 dimension simplex space are as follows:
S '=Ф 'iW+T′
Wherein, W is weight matrix, the converting vector of T biasing；To construct classification boundaries；
Step 4.5: establish the loss function for combining all training samples:
Wherein, ρiIt is optional object weight, hpIndicate the weight of the misclassification error calculated based on Huber hinge function,
GkExpression is the data acquisition system for belonging to k-th of state, and K is mechanical equipment health degenerate state number, and λ trW ' W is to avoid over-fitting
Penalty term, λ > 0 is regularization parameter；
Step 4.6: loss function is gradually minimized, when loss function meets following condition, training process terminates:
Lt-Lt-1< ∈ Lt-1
Wherein, LtAnd Lt-1Indicate L (W, T) in the loss function value at current time and last moment；∈ is preset stopping
Parameter；
Step 4.7: being verified using verification sample GenSVM model trained to step 4.6, if precision satisfaction is wanted
It asks, then carries out step 5；If precision is unsatisfactory for requiring, then 4.3 to 4.6 are re-execute the steps.
Further, loss function is gradually minimized by way of iteration optimization in step 4.6, steps are as follows:
Step 4.6.1: V=[T W '] ' and L=L (W, T) are defined, following optimization method is constructed
Wherein,For supporting point, VtFor the V value of t moment；
Step 4.6.2: it enablesV0For random initial point；
Step 4.6.3: order is acquiredTake minimum VtIt is worth and is named as Vt+1；
Step 4.6.4: if L (Vt+1)-L(Vt) < ∈ L (Vt), then optimized parameter V*=Vt+1, optimization process stopping；It is no
Then enableReturn step 4.6.2.
To achieve the goals above, the present invention also provides a kind of computer readable storage medium, this is computer-readable to be deposited
It is stored with computer program on storage media, any one side as described above is realized when which is executed by processor
Method.
To achieve the goals above, the present invention also provides a kind of mechanical equipment health degenerate state identification apparatus, including
Computer readable storage medium and processor as described above, processor is for calling and handling computer readable storage medium
The computer program of middle storage.
To achieve the goals above, the present invention also provides a kind of mechanical equipment health degenerate state identification systems, comprising:
Processor, training program module, test program module and multiple signal transducers；Wherein,
The step of processor is for calling training program module to execute the training stage in any one method as described above,
And the step of calling test program module to execute test phase in any one method as described above；
Multiple signal transducers are uploaded to place for obtaining according to Devices to test type preset a variety of monitoring signals
Device is managed so that training and test use.
In general, the above technical scheme conceived by the present invention compared with prior art, can obtain following beneficial to effect
Fruit:
1, the present invention is directed to mechanical equipment health degenerate state identification problem, establishes a kind of machine based on GSOM and GenSVM
Tool equipment health degenerate state on-line identification method can carry out accurate identification in real time to the healthy degenerate state of mechanical equipment,
The status real time monitor for realizing mechanical equipment, ensures safe and stable, the long-term operation of numerically-controlled machine tool.
2, the Feature fusion of the present invention based on GSOM, can be according to training data automatic adjusument network
Structural parameters avoid artificial settings network architecture parameters.
3, the algorithm used in the present invention is relatively simple, is easily programmed, it is easy-to-use can preferably to solve state identification algorithm
The contradictory problems of property and accuracy.
Detailed description of the invention
The flow chart of Fig. 1 the method for the present invention；
The a variety of monitoring signals temporal signatures figures of Fig. 2；
The a variety of monitoring signals power spectrum characteristic figures of Fig. 3；
The intrinsic mode energy feature figure of a variety of monitoring signals of Fig. 4；
Fig. 5 principal axis A E signal Time Domain Fusion feature.
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.
Such as Fig. 1, a kind of mechanical equipment health degenerate state discrimination method based on broad sense multi-category support vector machines includes
Following steps:
Step 1: obtaining a variety of monitoring signals of mechanical equipment, singular value processing and noise reduction are carried out to a variety of monitoring signals
Processing.
Step 2: to step 1, treated that each monitoring signals carry out time-domain analysis, extracts the time domain of each monitoring signals respectively
Feature, comprising: average value, mean square deviation, root amplitude, root-mean-square value, maximum value, flexure index, kurtosis index, peak value
The factor and the nargin factor.
To step 1, treated that each monitoring signals carry out power spectrumanalysis, and the power spectrum for extracting each monitoring signals respectively is special
Sign, comprising: center frequency, square frequency, root mean square frequency, frequency variance and frequency root variance.
To step 1, treated that each monitoring signals carry out CEEMDAN decomposition, obtains each intrinsic mode of each monitoring signals
Component calculates intrinsic mode energy feature of the energy value as mechanical equipment health degenerate state of each modal components.
Step 3: utilize GSOM network, the temporal signatures of each monitoring signals that step 2 is obtained respectively, power spectrum characteristic and
Intrinsic mode energy feature is merged, and Time Domain Fusion feature, power spectrum fusion feature and the eigen mode of each monitoring signals are obtained
State energy fusion feature.
Temporal signatures, power spectrum characteristic and intrinsic mode energy feature are merged using GSOM network in this step
Process is similar, and by taking temporal signatures as an example, the process of Fusion Features is as follows:
Step 3.1: a variety of monitoring signals data temporal signatures chosen under the normal condition of mechanical equipment part set up training
Data set Dn。
Step 3.2: building GSOM network model, initiation parameter.
Step 3.3: input training sample Dn, and set growing threshold distance dmax。
Step 3.4: random all temporal signatures x without the mechanical equipment for repeatedly choosing a moment from training dataset
(t)∈Dn.According to being triumph neuron principle with the neuron of x (t) minimum euclidean distance, i.e.,Determine triumph neuron.
Step 3.5: judging whether to need to grow new neuron.If oi(x)(x) > dmax, then Exist Network Structure is insufficient
It to describe the feature of input data, needs to increase a neuron, continues step 3.6；Otherwise, return step 3.4.
Step 3.6: setting the feedforward connection weight of new neuron as x (t)T, new neuron is calculated to other all nerves
The distance of the member, (a within the scope of certain distance1×dmax, a2×dmax)(a1a2, a1And a2For setting range parameter) find new nerve
The adjoining neuron of member, establishes syntople.
Step 3.7: step 3.4 to 3.6 is repeated, until the complete D of trainingnIn after all data, obtain trained GSOM net
Network model.
Step 3.8: a variety of monitoring signals data of mechanical equipment to be fused are sequentially input into training sequentially in time
In good GSOM network model, fused characteristic value is obtained.
Step 4: being based on GenSVM model, model is carried out using the fusion feature data of a variety of monitoring signals of mechanical equipment
Training.
Further, the training process of GenSVM is as follows in step 4:
Step 4.1: setting up performance degradation eigenmatrix F using the fusion feature data of a variety of monitoring signals of mechanical equipment.
F=[F1, F2, F3, L]
Wherein, F1、F2And F3It is special to respectively indicate Time Domain Fusion feature, power spectrum fusion feature and the fusion of intrinsic mode energy
Sign, and F1, F2, F3∈Rp×k, p is sampling number, and k indicates monitoring signals species number.L∈Rp×1Indicate capability of engineering equipment decline
State tag.
Step 4.2: performance degradation eigenmatrix F is randomly divided into training sample and verification sample according to a certain percentage,
In a preferred embodiment, which can be set to 7:3.
Step 4.3: training sample being inputted into GenSVM model, input data is mapped to by the kernel function in GenSVM model
Higher dimensional space is expressed as xi→Φi。
Step 4.4: simplex encryption algorithm is used for structural classification boundary, ΦiIt is expressed as in K-1 dimension simplex space
S '=Φ 'iW+t′
Wherein W is weight matrix, the converting vector of t biasing.
Step 4.5: the weight of the misclassification error based on the calculating of Huber hinge function is applied in loss function, most
The loss function in conjunction with all training samples is obtained eventually:
Wherein, ρiIt is optional object weight, hpIndicate the weight of the misclassification error calculated based on Huber hinge function,
Gk={ i:yi=k } indicate it is the data acquisition system for belonging to k-th of state, K is mechanical equipment health degenerate state number, and λ trW ' W is
The penalty term of over-fitting is avoided, λ > 0 is regularization parameter.
Step 4.6: iteration optimization algorithms are for gradually minimizing loss function.When loss function meets condition, training
Process terminates.Condition are as follows:
Lt-Lt-1< ∈ Lt-1
Wherein, LtAnd Lt-1Indicate the loss function value at current time and last moment.∈ is that the stopping of optimization algorithm is joined
Number.
The step of gradually minimizing loss function above by the mode of iteration optimization is as follows:
Step 4.6.1: V=[T W '] ' and L=L (W, T) are defined, following optimization method is constructed
Wherein,For supporting point, VtFor the V value of t moment；
Step 4.6.2: it enablesV0For random initial point；
Step 4.6.3: order is acquiredTake minimum VtIt is worth and is named as Vt+1；
Step 4.6.4: if L (Vt+1)-L(Vt) < ∈ L (Vt), then optimized parameter V*=Vt+1, optimization process stopping；Otherwise
It enablesReturn step 4.6.2.
Step 4.7: trained GenSVM model is verified using verification sample, if precision is met the requirements, into
Row step 5.If precision is unsatisfactory for requiring, then 4.3 to 4.6 are re-execute the steps.
Step 5: in test phase, for a variety of monitoring signals of mechanical equipment acquired in real time, according to step 2 and step 3
The feature of a variety of monitoring signals is extracted respectively and carries out Fusion Features, is input in trained GenSVM model, is obtained health
Degenerate state identification result.
In the following, using the MatsuuraMC -510V numerically-controlled machine tool multi-sensor monitoring signal data of NASA to of the invention
Validity is verified.
In order to obtain a variety of monitoring signals of mechanical equipment, it is mounted with two vibrations respectively on main shaft and workbench in this experiment
Sensor and two sound emission (AE) sensors, in addition, alternating current (AC) sensor and direct current (DC) biography are also mounted on main shaft
For sensor for monitoring current of electric, the signal of acquisition includes spindle vibration signal, principal axis A E signal, principal axis A C signal and main shaft DC
Signal and workbench vibration signal and workbench AE signal.
It should be noted that the object that this experiment is chosen is the equipment with main shaft, in other embodiments, according to reality
The difference of test object, above-mentioned test signal can also be correspondingly adjusted, such as reduced some signals or increased other letters
Number, such as increase noise signal, temperature signal and pressure signal.
Collected multiple sensor signals (i.e. a variety of monitoring signals) are amplified through NI High Speed Data Acquisition, after filtering, defeated
Enter computer and carries out the identification of cutter health degenerate state.In an experiment, use mechanical equipment processing dimension for 43mm × 178mm ×
The stainless steel work-piece of 5mm.The speed of mainshaft is set as 826 revs/min.There are two types of operating conditions for experimental data, are respectively denoted as A and B, often
Kind operating condition includes two experimental data set A1 and A2 and B1 and B2, as shown in table 1.
1 operating condition explanation of table
Data set A1 and B1 is as training data for training mechanical equipment health degenerate state to recognize model, data set A2
It is used to verify the validity of the online degenerate state identification of training pattern as test data with B2.Referring to international standard and reality
The entire service life of machining tool is divided into three states according to the abrasion width of machining tool rear surface by border process,
That is: health status, degenerate state and failure state.
Detection signal based on above-mentioned setting and acquisition, the specific verification process of this experiment are as follows:
Step 1: singular value processing and noise reduction process are carried out to training data；
Step 2: extracting the temporal signatures, power spectrum characteristic and intrinsic mode energy feature of a variety of monitoring signals.Data set
In A1 a variety of monitoring signals temporal signatures as shown in Fig. 2, power spectrum characteristic as shown in figure 3, intrinsic mode energy feature such as Fig. 4
It is shown.
Step 3: and then temporal signatures, power spectrum characteristic and intrinsic mode energy feature are merged respectively, to above-mentioned
For feature in such a way that the feature of the same race of monitoring signals of the same race is merged, the fusion for obtaining each feature of each monitoring signals is special
Sign.Principal axis A E signal Time Domain Fusion feature is as shown in Figure 5.
Step 4: building GenSVM model, in the training stage, using the fusion feature of a variety of monitoring signals of machining tool as
Training data is trained GenSVM model；
Step 5: in test phase, for a variety of monitoring signals of mechanical equipment acquired in real time, the time domain for extracting data is special
Sign, power spectrum characteristic and intrinsic mode energy feature are simultaneously merged respectively.Then fusion feature data are input to and are trained
Broad sense multi-category support vector machines model in, obtain healthy degenerate state identification result.
In order to protrude the advantage of this method, support vector machines, least square method supporting vector machine, BP neural network and adaptive
Totally 4 kinds of methods are used to compare with this method neuro fuzzy systems.Table 2 illustrates this method and other four kinds of control methods
As a result, it can be seen that the state identification precision of method of GSOM+genSVM of the present invention is higher than other four kinds on the whole from table
Method, and other four kinds of methods are also significantly better than in the accuracy adaptability of different operating conditions.
Table 2
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 (7)
1. a kind of mechanical equipment health degenerate state discrimination method, which is characterized in that including training stage and test phase:
Training stage includes the following steps:
Step 1: obtaining a variety of monitoring signals of mechanical equipment, a variety of monitoring signals are carried out at singular value processing and noise reduction
Reason；
Step 2: to step 1, treated that each monitoring signals carry out time-domain analysis, extracts the temporal signatures of each monitoring signals respectively,
It include: average value, mean square deviation, root amplitude, root-mean-square value, maximum value, flexure index, kurtosis index, peak factor
And the nargin factor；
To step 1, treated that each monitoring signals carry out power spectrumanalysis, extracts the power spectrum characteristic of each monitoring signals respectively, wraps
It includes: center frequency, square frequency, root mean square frequency, frequency variance and frequency root variance；
To step 1, treated that each monitoring signals carry out CEEMDAN decomposition, obtains each intrinsic mode point of each monitoring signals
Amount, calculates intrinsic mode energy feature of the energy value as mechanical equipment health degenerate state of each modal components；
Step 3: utilizing GSOM network, the temporal signatures of each monitoring signals obtained respectively to step 2, power spectrum characteristic and intrinsic
Mode energy feature is merged, and Time Domain Fusion feature, power spectrum fusion feature and the intrinsic mode energy of each monitoring signals are obtained
Measure fusion feature；
Step 4: each fusion feature of each monitoring signals obtained using step 3 is trained GenSVM model；
Test phase includes the following steps:
Step 5: for a variety of monitoring signals of mechanical equipment acquired in real time, extracting a variety of monitorings respectively according to step 2 and step 3
After the feature of signal and respectively progress Fusion Features, it is input in the trained GenSVM model of step 4, secure good health shape of degenerating
State identification result.
2. a kind of mechanical equipment health degenerate state discrimination method as described in claim 1, which is characterized in that sharp in step 3
The process merged with GSOM network to temporal signatures, power spectrum characteristic and intrinsic mode energy feature is as follows:
Step 3.1: choosing one of above-mentioned three kinds of features of a variety of monitoring signals data under the normal condition of mechanical equipment part group
Build training dataset Dn；
Step 3.2: building GSOM network model, initiation parameter；
Step 3.3: inputting training sample D to GSOM network modeln, and set growing threshold distance dmax；
Step 3.4: random all temporal signatures x (t) without the mechanical equipment for repeatedly choosing a certain moment from training dataset
∈Dn；According to being triumph neuron principle with the neuron of x (t) minimum euclidean distance, i.e.,Determine the triumph neuron in GSOM network model；
Step 3.5: judging whether to need to grow new neuron；If Oi(x)(x) > dmax, then Exist Network Structure is not enough to describe
The feature of input data needs to increase a new neuron in GSOM network model, continues step 3.6；Otherwise, return step
3.4；
Step 3.6: setting the feedforward connection weight of new neuron as x (t)T, calculate new neuron to other all neurons away from
From in distance to a declared goal range (a1×dmax, a2×dmax) in find the adjoining neuron of new neuron, establish syntople；Its
In, a1< a2, a1And a2For preset range parameter；
Step 3.7: step 3.4 to 3.6 is repeated, until the complete D of trainingnIn after all data, obtain trained GSOM network mould
Type；
Step 3.8: a variety of monitoring signals data to be fused are sequentially input into the trained GSOM of step 3.7 sequentially in time
Characteristic value in network model, after respectively obtaining each detection signal fused.
3. a kind of mechanical equipment health degenerate state discrimination method as claimed in claim 1 or 2, which is characterized in that in step 4
The training process of GenSVM is as follows:
Step 4.1: setting up performance degradation eigenmatrix F using the fusion feature data of a variety of monitoring signals of mechanical equipment；
F=[F1,F2,F3, L]
Wherein, F1、F2And F3Time Domain Fusion feature, power spectrum fusion feature and intrinsic mode energy fusion feature are respectively indicated, and
F1,F2, F3∈Rp×k, p is sampling number, and k indicates monitoring signals species number, Rp×kIndicate the real number matrix of p × k dimension；L indicates machine
Tool device performance decay state tag, L ∈ Rp×1；
Step 4.2: performance degradation eigenmatrix F is randomly divided into training sample and verification sample；
Step 4.3: training sample being inputted into GenSVM model, the kernel function in GenSVM model will be in the training sample of input
Data are mapped to higher dimensional space, are expressed as xi→Φi；
Step 4.4: by ФiIt is indicated in K-1 dimension simplex space are as follows:
S '=Φ 'iW+T′
Wherein, W is weight matrix, the converting vector of T biasing；To construct classification boundaries；
Step 4.5: establish the loss function for combining all training samples:
Wherein, ρiIt is optional object weight, hpIndicate the weight of the misclassification error calculated based on Huber hinge function, GkTable
Show the set for belonging to the sample of k-th of state, K is mechanical equipment health degenerate state number, and λ trW ' W is to avoid punishing for over-fitting
Item is penalized, λ > 0 is regularization parameter；
Step 4.6: loss function is gradually minimized, when loss function meets following condition, training process terminates:
Lt-Lt-1< ∈ Lt-1
Wherein, LtAnd Lt-1Indicate L (W, T) in the loss function value at current time and last moment；∈ is preset stopping parameter；
Step 4.7: it is verified using verification sample GenSVM model trained to step 4.6, if precision is met the requirements,
Carry out step 5；If precision is unsatisfactory for requiring, then 4.3 to 4.6 are re-execute the steps.
4. a kind of mechanical equipment health degenerate state discrimination method as claimed in claim 3, which is characterized in that in step 4.6
Loss function is gradually minimized by way of iteration optimization, steps are as follows:
Step 4.6.1: V=[T W '] ' and L=L (W, T) are defined, following optimization method is constructed
Wherein,For supporting point, VtFor the V value of t moment；
Step 4.6.2: it enablesV0For random initial point；
Step 4.6.3: order is acquiredTake minimum VtIt is worth and is named as Vt+1；
Step 4.6.4: if L (Vt+1)-L(Vt) < ∈ L (Vt), then optimized parameter V*=Vt+1, optimization process stopping；Otherwise it enablesReturn step 4.6.2.
5. a kind of computer readable storage medium, which is characterized in that be stored with computer journey on the computer readable storage medium
Sequence realizes such as Claims 1 to 4 described in any item methods when the computer program is executed by processor.
6. a kind of mechanical equipment health degenerate state identification apparatus, which is characterized in that including computer as claimed in claim 5
Readable storage medium storing program for executing and processor, processor is for calling and handling the computer journey stored in computer readable storage medium
Sequence.
7. a kind of mechanical equipment health degenerate state identification system characterized by comprising processor, is surveyed training program module
Try program module and multiple signal transducers；Wherein,
The step of training stage that processor is used to that training program module to be called to execute as described in Claims 1 to 4 any one,
And the step of calling test program module to execute the test phase as described in Claims 1 to 4 any one；
Multiple signal transducers are uploaded to processor for obtaining according to Devices to test type preset a variety of monitoring signals
So that training and test use.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811281339.8A CN109597315B (en) | 2018-10-31 | 2018-10-31 | Method, equipment and system for identifying health degradation state of mechanical equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811281339.8A CN109597315B (en) | 2018-10-31 | 2018-10-31 | Method, equipment and system for identifying health degradation state of mechanical equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109597315A true CN109597315A (en) | 2019-04-09 |
| CN109597315B CN109597315B (en) | 2020-08-18 |
Family
ID=65957121
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811281339.8A Active CN109597315B (en) | 2018-10-31 | 2018-10-31 | Method, equipment and system for identifying health degradation state of mechanical equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109597315B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110320033A (en) * | 2019-08-08 | 2019-10-11 | 广东省智能机器人研究院 | A kind of rolling bearing health degenerate state discrimination method |
| CN110561193A (en) * | 2019-09-18 | 2019-12-13 | 杭州友机技术有限公司 | Cutter wear assessment and monitoring method and system based on feature fusion |
| CN111504385A (en) * | 2020-05-13 | 2020-08-07 | 兰州工业学院 | Multi-parameter monitoring device and method suitable for abnormal state of mechanical equipment |
| CN112034789A (en) * | 2020-08-25 | 2020-12-04 | 国家机床质量监督检验中心 | Health assessment method, system and assessment terminal for key parts and complete machine of numerical control machine tool |
| CN112350878A (en) * | 2019-08-07 | 2021-02-09 | 阿里巴巴集团控股有限公司 | Pressure test system |
| WO2021134253A1 (en) * | 2019-12-30 | 2021-07-08 | 江苏南高智能装备创新中心有限公司 | Fault prediction system based on sensor data on numerical control machine tool and method therefor |
| CN113711143A (en) * | 2019-04-11 | 2021-11-26 | 三菱电机株式会社 | Numerical control device |
| CN115687984A (en) * | 2023-01-05 | 2023-02-03 | 绍兴市特种设备检测院 | Method for monitoring health state of stirring kettle |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6981423B1 (en) * | 2002-04-09 | 2006-01-03 | Rockwell Automation Technologies, Inc. | System and method for sensing torque on a rotating shaft |
| CN103868690A (en) * | 2014-02-28 | 2014-06-18 | 中国人民解放军63680部队 | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics |
| CN107563414A (en) * | 2017-08-14 | 2018-01-09 | 北京航空航天大学 | A kind of complex device degenerate state recognition methods based on Kohonen SVM |
| CN108332970A (en) * | 2017-11-17 | 2018-07-27 | 中国铁路总公司 | A kind of Method for Bearing Fault Diagnosis based on LS-SVM and D-S evidence theory |
| CN108507787A (en) * | 2018-06-28 | 2018-09-07 | 山东大学 | Wind power gear speed increase box fault diagnostic test platform based on multi-feature fusion and method |
-
2018
- 2018-10-31 CN CN201811281339.8A patent/CN109597315B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6981423B1 (en) * | 2002-04-09 | 2006-01-03 | Rockwell Automation Technologies, Inc. | System and method for sensing torque on a rotating shaft |
| CN103868690A (en) * | 2014-02-28 | 2014-06-18 | 中国人民解放军63680部队 | Rolling bearing state automatic early warning method based on extraction and selection of multiple characteristics |
| CN107563414A (en) * | 2017-08-14 | 2018-01-09 | 北京航空航天大学 | A kind of complex device degenerate state recognition methods based on Kohonen SVM |
| CN108332970A (en) * | 2017-11-17 | 2018-07-27 | 中国铁路总公司 | A kind of Method for Bearing Fault Diagnosis based on LS-SVM and D-S evidence theory |
| CN108507787A (en) * | 2018-06-28 | 2018-09-07 | 山东大学 | Wind power gear speed increase box fault diagnostic test platform based on multi-feature fusion and method |
Non-Patent Citations (1)
| Title |
|---|
| 董绍江: "基于优化支持向量机的空间滚动轴承寿命预测方法研究", 《中国博士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113711143A (en) * | 2019-04-11 | 2021-11-26 | 三菱电机株式会社 | Numerical control device |
| CN112350878A (en) * | 2019-08-07 | 2021-02-09 | 阿里巴巴集团控股有限公司 | Pressure test system |
| CN110320033A (en) * | 2019-08-08 | 2019-10-11 | 广东省智能机器人研究院 | A kind of rolling bearing health degenerate state discrimination method |
| CN110561193A (en) * | 2019-09-18 | 2019-12-13 | 杭州友机技术有限公司 | Cutter wear assessment and monitoring method and system based on feature fusion |
| WO2021134253A1 (en) * | 2019-12-30 | 2021-07-08 | 江苏南高智能装备创新中心有限公司 | Fault prediction system based on sensor data on numerical control machine tool and method therefor |
| CN111504385A (en) * | 2020-05-13 | 2020-08-07 | 兰州工业学院 | Multi-parameter monitoring device and method suitable for abnormal state of mechanical equipment |
| CN111504385B (en) * | 2020-05-13 | 2022-04-29 | 兰州工业学院 | Multi-parameter monitoring device and method suitable for abnormal state of mechanical equipment |
| CN112034789A (en) * | 2020-08-25 | 2020-12-04 | 国家机床质量监督检验中心 | Health assessment method, system and assessment terminal for key parts and complete machine of numerical control machine tool |
| CN115687984A (en) * | 2023-01-05 | 2023-02-03 | 绍兴市特种设备检测院 | Method for monitoring health state of stirring kettle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109597315B (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109597315A (en) | A kind of mechanical equipment health degenerate state discrimination method, equipment and system | |
| Huang et al. | An enhanced deep learning-based fusion prognostic method for RUL prediction | |
| Yang et al. | Remaining useful life prediction based on a double-convolutional neural network architecture | |
| Lee et al. | Application of deep neural network and generative adversarial network to industrial maintenance: A case study of induction motor fault detection | |
| CN106682814B (en) | Wind turbine generator fault intelligent diagnosis method based on fault knowledge base | |
| Xiang et al. | Fault diagnosis of rolling bearing under fluctuating speed and variable load based on TCO spectrum and stacking auto-encoder | |
| Samanta et al. | Artificial neural networks and genetic algorithm for bearing fault detection | |
| CN113255848B (en) | Water turbine cavitation sound signal identification method based on big data learning | |
| CN108073158A (en) | Based on PCA and KNN density algorithm Wind turbines Method for Bearing Fault Diagnosis | |
| CN112304613A (en) | Wind turbine generator bearing early warning method based on feature fusion | |
| CN110738360A (en) | equipment residual life prediction method and system | |
| Gong et al. | A fast anomaly diagnosis approach based on modified CNN and multisensor data fusion | |
| Thoppil et al. | Deep learning algorithms for machinery health prognostics using time-series data: A review | |
| Li et al. | Gear pitting fault diagnosis using raw acoustic emission signal based on deep learning | |
| CN117320236B (en) | Lighting control method and system of unmanned aerial vehicle | |
| CN115438714A (en) | Clustering federal learning driven mechanical fault diagnosis method, device and medium | |
| Naveen Venkatesh et al. | Transfer Learning‐Based Condition Monitoring of Single Point Cutting Tool | |
| CN115994308B (en) | Numerical control machine tool fault identification method, system, equipment and medium based on meta learning | |
| Delgado et al. | A novel condition monitoring scheme for bearing faults based on curvilinear component analysis and hierarchical neural networks | |
| Kuo et al. | Dense-block structured convolutional neural network-based analytical prediction system of cutting tool wear | |
| Kim et al. | AnoGAN-based anomaly filtering for intelligent edge device in smart factory | |
| CN110956331A (en) | Method, system and device for predicting operation state of digital factory | |
| CN113486926B (en) | Automatic change pier equipment anomaly detection system | |
| Xu et al. | Bearing Fault Diagnosis in the Mixed Domain Based on Crossover‐Mutation Chaotic Particle Swarm | |
| US11687772B2 (en) | Optimization of cyber-physical systems |
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 |