CN109597315B - Method, equipment and system for identifying health degradation state of mechanical equipment - Google Patents
Method, equipment and system for identifying health degradation state of mechanical equipment Download PDFInfo
- Publication number
- CN109597315B CN109597315B CN201811281339.8A CN201811281339A CN109597315B CN 109597315 B CN109597315 B CN 109597315B CN 201811281339 A CN201811281339 A CN 201811281339A CN 109597315 B CN109597315 B CN 109597315B
- Authority
- CN
- China
- Prior art keywords
- training
- mechanical equipment
- degradation state
- monitoring signals
- gsom
- 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.)
- Active
Links
Images
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 method, equipment and a system for identifying the health degradation state of mechanical equipment, and belongs to the field of mechanical state monitoring and health degradation state identification. The method comprises the steps of firstly obtaining various monitoring signals of mechanical equipment, extracting time domain characteristics, power spectrum characteristics and intrinsic mode energy characteristics of the monitoring signals, then fusing the characteristics by utilizing a GSOM network to obtain fusion characteristics, then constructing a GenSVM model, training the GenSVM model by utilizing the fusion characteristic data to obtain a test model, finally collecting various monitoring signals of the mechanical equipment to be tested in real time, obtaining the signals, inputting the fusion characteristic data into the test model, and obtaining a health degradation state identification result. The method, the equipment and the system based on the method can accurately identify the health degradation state of the mechanical equipment in real time, realize the real-time monitoring of the state of the mechanical equipment and ensure the safe, stable and long-period operation of a numerical control machine.
Description
Technical Field
The invention belongs to the technical field of mechanical state monitoring and health degradation state identification, and particularly relates to a mechanical health degradation monitoring method, equipment and a system based on a growing self-organizing mapping (GSOM) network and a Generalized multi-class support vector machine (GenSVM).
Background
Mechanical equipment is used as core production equipment in manufacturing, metallurgy, chemical industry, environmental protection and other process production industries, the operation of the mechanical equipment directly affects the production efficiency, product quality and maintenance cost of enterprises, and especially sudden failures and safety accidents occurring in the operation of the equipment often cause interruption of the whole production process to cause major production loss. Therefore, real-time identification of health degradation conditions of mechanical equipment is critical in the manufacturing process.
The Support Vector Machine (SVM) algorithm is a common method for identifying the health degradation condition of mechanical equipment, the theory of which is proposed by Vapnik, and the SVM algorithm is widely applied to the fields of health state monitoring, fault diagnosis and service life prediction of mechanical equipment because the SVM algorithm is suitable for a small sample environment, is simple and understandable, has better robustness and higher generalization capability, and utilizes an inner product kernel function to replace high-dimensional nonlinear mapping. However, when the health degradation state of the mechanical equipment is identified, a single SVM model is only suitable for two degradation state identification scenes, and multiple degradation states cannot be identified. Although the multi-SVM combined model can be used for identifying various degradation states of mechanical equipment, a complex dual optimization problem exists.
Therefore, a solution that is simple and easy to implement and can identify the health degradation condition of the mechanical equipment in real time is needed.
Disclosure of Invention
Aiming at the defects or improvement requirements in the prior art, the invention provides a method for identifying the health degradation state of mechanical equipment, which aims to realize the online identification of the health degradation state of the mechanical equipment by acquiring various monitoring signals of the mechanical equipment to be detected in real time and fusing and training the signals respectively based on a GSOM (global system for mobile communications) and a GenSVM (support vector machine).
In order to achieve the above object, the present invention provides a method for identifying health degradation state of mechanical equipment, which comprises a training phase and a testing phase:
the training phase comprises the following steps:
step 1: acquiring various monitoring signals of mechanical equipment, and performing singular value removal processing and noise reduction processing on the various monitoring signals;
step 2: performing time domain analysis on each monitoring signal processed in the step 1, and respectively extracting time domain characteristics of each monitoring signal, wherein the time domain analysis comprises the following steps: an average value, a mean square deviation value, a square root amplitude value, a root mean square value, a maximum absolute value, a skewness index, a kurtosis index, a peak factor and a margin factor;
performing power spectrum analysis on each monitoring signal processed in the step 1, and respectively extracting power spectrum characteristics of each monitoring signal, wherein the power spectrum analysis comprises the following steps: frequency center, mean square frequency, root mean square frequency, frequency variance, and frequency root variance;
performing CEEMDAN decomposition on each monitoring signal processed in the step 1 to obtain each intrinsic modal component of each monitoring signal, and calculating the energy value of each modal component as the intrinsic modal energy characteristic of the health degradation state of the mechanical equipment;
and step 3: respectively fusing the time domain characteristics, the power spectrum characteristics and the intrinsic mode energy characteristics of the monitoring signals obtained in the step (2) by utilizing a GSOM (global system for mobile communications) network to obtain time domain fusion characteristics, power spectrum fusion characteristics and intrinsic mode energy fusion characteristics of the monitoring signals;
and 4, step 4: training the GenSVM model by using the fusion characteristics of the monitoring signals obtained in the step (3);
the testing phase comprises the following steps:
and 5: and (3) respectively extracting the characteristics of the various monitoring signals according to the step (2) and the step (3) and respectively performing characteristic fusion on the various monitoring signals, and inputting the various monitoring signals into the GenSVM model trained in the step (4) to obtain a health degradation state identification result.
Further, the process of fusing the time domain feature, the power spectrum feature and the eigen-modal energy feature by using the GSOM network in step 3 is as follows:
step 3.1: selecting one of the three characteristics of the multiple monitoring signal data under the normal state of the mechanical equipment part to establish a training data set Dn;
Step 3.2: constructing a GSOM network model, and initializing parameters;
step 3.3: inputting training sample D to GSOM network modelnAnd setting a growth threshold distance dmax;
Step 3.4. randomly and unrepeatedly selecting all time domain characteristics x (t) ∈ D of the mechanical equipment at a certain moment from the training data setn(ii) a Based on the principle that the neuron with the minimum Euclidean distance to x (t) is the winning neuron, i.e.Determining winning neurons in the GSOM network model;
step 3.5: judging whether new neurons need to grow or not; if oi(x)(x)>dmaxIf the current network structure is not enough to describe the characteristics of the input data, a new neuron needs to be added in the GSOM network model, and the step 3.6 is continued; otherwise, returning to the step 3.4;
step 3.6: setting the feedforward connection weight of the new neuron as x (t)TCalculating the distance from the new neuron to all other neurons in the specified distance range (a)1×dmax,a2×dmax) Searching adjacent neurons of the new neuron internally, and establishing an adjacent relation; wherein, a1<a2,a1And a2Is a preset range parameter;
step 3.7: repeating steps 3.4 to 3.6 until training DnAfter all the data are stored, a well-trained GSOM network model is obtained;
step 3.8: and (3) sequentially inputting various monitoring signal data to be fused into the GSOM network model trained in the step 3.7 according to the time sequence, and respectively obtaining the characteristic values of the fused detection signals.
Further, the training process of the GenSVM in step 4 is as follows:
step 4.1: establishing a performance decline characteristic matrix F by utilizing fusion characteristic data of various monitoring signals of mechanical equipment;
F=[F1,F2,F3,L]
wherein, F1、F2And F3Respectively representing a time domain fusion feature, a power spectrum fusion feature and an eigenmode energy fusion feature, and F1,F2,F3∈Rp×kP is the number of sampling points, k represents the number of monitoring signal types, Rp×kA real number matrix representing p × k dimension, L a mechanical equipment performance degradation state label, L ∈ Rp×1;
Step 4.2: randomly dividing the performance decline characteristic matrix F into training samples and check samples;
step 4.3: inputting the training sample into a GenSVM model, wherein a kernel function in the GenSVM model maps data in the input training sample to a high-dimensional modelSpace, denoted as xi→Φi;
Step 4.4: will beiExpressed in K-1 dimensional simplex space as:
s′=Ф′iW+T′
wherein W is a weight matrix, T biased transition vector; thereby constructing a classification boundary;
step 4.5: establish a loss function that combines all training samples:
where ρ isiIs an optional object weight, hpWeights, G, representing misclassification errors calculated based on the Huber hinge functionkThe representation is a data set belonging to the kth state, K is the health degradation state number of the mechanical equipment, lambda trW' W is a penalty term for avoiding overfitting, and lambda > 0 is a regularization parameter;
step 4.6: gradually minimizing the loss function, and finishing the training process when the loss function meets the following conditions:
Lt-Lt-1<∈Lt-1
wherein L istAnd Lt-1Representing the loss function values of L (W, T) at the current moment and the last moment, ∈ is a preset stopping parameter;
step 4.7: verifying the GenSVM model trained in the step 4.6 by using a verification sample, and if the precision meets the requirement, performing a step 5; and if the precision does not meet the requirement, re-executing the steps 4.3 to 4.6.
Further, in step 4.6, the loss function is minimized step by means of iterative optimization, and the steps are as follows:
step 4.6.1: definition V ═ T W']' and L ═ L (W, T), the following optimization equation was constructed
Wherein the content of the first and second substances,as a support point, VtIs the V value at the time t;
Step 4.6.4: if L (V)t+1)-L(Vt)<∈L(Vt) Then the optimum parameter V ═ Vt+1The optimization process is stopped; otherwise makeAnd returning to the step 4.6.2.
To achieve the above object, the present invention also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements any one of the methods described above.
In order to achieve the above object, the present invention also provides a mechanical device health degradation state identification device, including the computer readable storage medium as described above and a processor for calling and processing a computer program stored in the computer readable storage medium.
In order to achieve the above object, the present invention further provides a system for identifying health degradation state of mechanical equipment, including: the device comprises a processor, a training program module, a testing program module and a plurality of signal sensors; wherein the content of the first and second substances,
the processor is used for calling the training program module to execute the step of the training phase in any one method and calling the testing program module to execute the step of the testing phase in any one method;
the signal sensors are used for acquiring various preset monitoring signals according to the type of the equipment to be tested and uploading the monitoring signals to the processor for training and testing.
In general, compared with the prior art, the above technical solution contemplated by the present invention can obtain the following beneficial effects:
1. aiming at the problem of identification of the health degradation state of mechanical equipment, the invention establishes an online identification method of the health degradation state of the mechanical equipment based on GSOM and GenSVM, can accurately identify the health degradation state of the mechanical equipment in real time, realizes real-time monitoring of the state of the mechanical equipment, and ensures safe, stable and long-period operation of a numerical control machine.
2. The GSOM-based feature fusion method can adaptively adjust the network structure parameters according to the training data, and avoids the manual setting of the network structure parameters.
3. The algorithm adopted in the invention is simpler and easy to program, and can better solve the contradiction between the usability and the accuracy of the state identification algorithm.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a time domain plot of various monitor signals;
FIG. 3 is a power spectrum profile of various monitored signals;
FIG. 4 is a graph of the energy characteristics of the eigenmodes of various monitoring signals;
FIG. 5 shows a time domain fusion feature of principal axis AE signals.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Referring to fig. 1, a method for identifying health degradation state of mechanical equipment based on generalized multi-class support vector machine includes the following steps:
step 1: the method comprises the steps of obtaining various monitoring signals of mechanical equipment, and carrying out singular value removing processing and noise reduction processing on the various monitoring signals.
Step 2: performing time domain analysis on each monitoring signal processed in the step 1, and respectively extracting time domain characteristics of each monitoring signal, wherein the time domain analysis comprises the following steps: mean, mean square error, square root amplitude, root mean square value, maximum absolute value, skewness index, kurtosis index, peak factor, and margin factor.
Performing power spectrum analysis on each monitoring signal processed in the step 1, and respectively extracting power spectrum characteristics of each monitoring signal, wherein the power spectrum analysis comprises the following steps: center of frequency, mean square frequency, root mean square frequency, frequency variance, and frequency root variance.
And (3) performing CEEMDAN decomposition on each monitoring signal processed in the step (1) to obtain each intrinsic mode component of each monitoring signal, and calculating the energy value of each mode component as the intrinsic mode energy characteristic of the health degradation state of the mechanical equipment.
And step 3: and (3) fusing the time domain characteristics, the power spectrum characteristics and the intrinsic mode energy characteristics of the monitoring signals obtained in the step (2) by utilizing a GSOM (global system for mobile communications) network to obtain the time domain fusion characteristics, the power spectrum fusion characteristics and the intrinsic mode energy fusion characteristics of the monitoring signals.
In this step, the process of fusing the time domain feature, the power spectrum feature and the eigen-mode energy feature by using the GSOM network is similar, and taking the time domain feature as an example, the process of feature fusion is as follows:
step 3.1: selecting multiple monitoring signal data time domain characteristics under the normal state of mechanical equipment part to establish a training data set Dn。
Step 3.2: and constructing a GSOM network model and initializing parameters.
Step 3.3: input training sample DnAnd setting a growth threshold distance dmax。
Step 3.4: of machines randomly and non-repeatedly picking a moment from a training data setAll time-domain features x (t) ∈ Dn. Based on the principle that the neuron with the minimum Euclidean distance to x (t) is the winning neuron, i.e.A winning neuron is determined.
Step 3.5: and judging whether new neurons need to grow or not. If oi(x)(x)>dmaxIf the current network structure is not enough to describe the characteristics of the input data, a neuron needs to be added, and the step 3.6 is continued; otherwise, return to step 3.4.
Step 3.6: setting the feedforward connection weight of the new neuron as x (t)TCalculating the distance from the new neuron to all other neurons within a certain distance range (a)1×dmax,a2×dmax)(a1a2,a1And a2To set range parameters) to find the adjacent neurons of the new neuron and establish the adjacent relation.
Step 3.7: repeating steps 3.4 to 3.6 until training DnAnd obtaining a well-trained GSOM network model after all the data are acquired.
Step 3.8: and inputting various monitoring signal data of the mechanical equipment to be fused into the trained GSOM network model in sequence according to the time sequence to obtain the fused characteristic value.
And 4, step 4: based on the GenSVM model, the model is trained by utilizing fusion characteristic data of various monitoring signals of mechanical equipment.
Further, the training process of the GenSVM in step 4 is as follows:
step 4.1: and establishing a performance decline characteristic matrix F by utilizing fusion characteristic data of various monitoring signals of mechanical equipment.
F=[F1,F2,F3,L]
Wherein, F1、F2And F3Respectively representing a time domain fusion feature, a power spectrum fusion feature and an eigenmode energy fusion feature, and F1,F2,F3∈Rp×kP is the number of sampling points, k represents the number of monitoring signal types。L∈Rp×1And (4) a label representing the performance degradation state of the mechanical equipment.
Step 4.2: the performance degradation feature matrix F is randomly divided into training samples and check samples according to a certain ratio, which in a preferred embodiment may be set to 7: 3.
step 4.3: inputting the training sample into GenSVM model, and mapping the input data to high-dimensional space by kernel function in GenSVM model, wherein x is representedi→Φi。
Step 4.4: simplex coding algorithms for constructing classification boundaries, phiiExpressed as in K-1 dimensional simplex space
s′=Φ′iW+t′
Where W is the weight matrix, t is the offset translation vector.
Step 4.5: the weight of the misclassification error calculated based on the Huber hinge function is applied to the loss function, and finally the loss function combining all training samples is obtained:
where ρ isiIs an optional object weight, hpWeights, G, representing misclassification errors calculated based on the Huber hinge functionk={i:yiK represents the data set belonging to the kth state, K is the number of states of health degradation of the machine, λ trW' W is a penalty term to avoid overfitting, λ > 0 is a regularization parameter.
Step 4.6: an iterative optimization algorithm is used to gradually minimize the loss function. When the loss function satisfies the condition, the training process ends. The conditions are as follows:
Lt-Lt-1<∈Lt-1
wherein L istAnd Lt-1The loss function values representing the current time and the last time ∈ are the stopping parameters for the optimization algorithm.
The above-mentioned step of gradually minimizing the loss function by means of iterative optimization is as follows:
step 4.6.1: definition V = [ T W']' and L ═ L (W, T), the following optimization equation was constructed
Wherein the content of the first and second substances,as a support point, VtIs the V value at the time t;
Step 4.6.4: if L (V)t+1)-L(Vt)<∈L(Vt) Then the optimum parameter V x = Vt+1The optimization process is stopped; otherwise makeAnd returning to the step 4.6.2.
Step 4.7: and (5) verifying the trained GenSVM model by using a verification sample, and if the precision meets the requirement, performing the step 5. And if the precision does not meet the requirement, re-executing the steps 4.3 to 4.6.
And 5: in the testing stage, aiming at various monitoring signals of the mechanical equipment collected in real time, the characteristics of the various monitoring signals are respectively extracted according to the step 2 and the step 3, the characteristics are fused, and the signals are input into a trained GenSVM model to obtain a health degradation state identification result.
Next, the effectiveness of the present invention was verified using the Matsuura MC-510V NC machine tool multisensor monitor signal data of NASA.
In order to obtain various monitoring signals of mechanical equipment, two vibration sensors and two Acoustic Emission (AE) sensors are respectively installed on a main shaft and a workbench in the experiment, in addition, an Alternating Current (AC) sensor and a Direct Current (DC) sensor are also installed on the main shaft and used for monitoring the current of a motor, and the acquired signals comprise main shaft vibration signals, main shaft AE signals, main shaft AC signals and main shaft DC signals, and workbench vibration signals and workbench AE signals.
It should be noted that the selected object in the experiment is a device with a spindle, and in other embodiments, the test signal may be adjusted accordingly according to a difference of an actual test object, for example, some signals are reduced or other signals are added, such as a noise signal, a temperature signal, a pressure signal, and the like.
Collected multi-sensor signals (namely various monitoring signals) are amplified and filtered by an NI high-speed data collecting board and then input into a computer to identify the health degradation state of the cutter. In the experiments, a stainless steel workpiece having dimensions of 43mm × 178mm × 5mm was machined using a mechanical device. The spindle speed was set to 826 rpm. The experimental data had two conditions, labeled a and B, each comprising two experimental data sets a1 and a2 and B1 and B2, as shown in table 1.
TABLE 1 description of the operating conditions
The data sets A1 and B1 are used as training data for training a machine health degradation state identification model, and the data sets A2 and B2 are used as test data for verifying the validity of online degradation state identification of the trained model. Referring to international standards and actual machining processes, the entire life of a machining tool is divided into three states according to the wear width of the rear surface of the machining tool, namely: a healthy state, a degraded state, and a failed state.
Based on the above-mentioned set and obtained detection signals, the specific verification process of this experiment is as follows:
step 1: performing singular value removal processing and noise reduction processing on the training data;
step 2: and extracting time domain characteristics, power spectrum characteristics and intrinsic mode energy characteristics of various monitoring signals. The time domain characteristics of the various monitored signals in data set a1 are shown in fig. 2, the power spectrum characteristics are shown in fig. 3, and the eigenmode energy characteristics are shown in fig. 4.
And step 3: and then, respectively fusing the time domain characteristics, the power spectrum characteristics and the intrinsic mode energy characteristics, and acquiring the fusion characteristics of the monitoring signals according to the fusion mode of the same characteristics of the same monitoring signals. The principal axis AE signal time domain fusion feature is shown in FIG. 5.
And 4, step 4: constructing a GenSVM model, and training the GenSVM model by taking the fusion characteristics of various monitoring signals of a processing tool as training data in a training stage;
and 5: in the testing stage, time domain characteristics, power spectrum characteristics and intrinsic mode energy characteristics of data are extracted and fused respectively according to various monitoring signals of the mechanical equipment acquired in real time. And then inputting the fusion characteristic data into a trained generalized multi-classification support vector machine model to obtain a health degradation state identification result.
In order to highlight the advantages of the method, 4 methods including a support vector machine, a least square support vector machine, a BP neural network and an adaptive neural fuzzy system are used for comparison with the method. Table 2 shows the results of the present method and the other four comparison methods, and it can be seen from the table that the state identification accuracy of the GSOM + genSVM method of the present invention is integrally higher than the other four methods, and is also significantly better than the other four methods in the accuracy adaptability of different working conditions.
TABLE 2
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (6)
1. A method for identifying health degradation state of mechanical equipment is characterized by comprising a training stage and a testing stage:
the training phase comprises the following steps:
step 1: acquiring various monitoring signals of mechanical equipment, and performing singular value removal processing and noise reduction processing on the various monitoring signals;
step 2: performing time domain analysis on each monitoring signal processed in the step 1, and respectively extracting time domain characteristics of each monitoring signal, wherein the time domain analysis comprises the following steps: an average value, a mean square deviation value, a square root amplitude value, a root mean square value, a maximum absolute value, a skewness index, a kurtosis index, a peak factor and a margin factor;
performing power spectrum analysis on each monitoring signal processed in the step 1, and respectively extracting power spectrum characteristics of each monitoring signal, wherein the power spectrum analysis comprises the following steps: frequency center, mean square frequency, root mean square frequency, frequency variance, and frequency root variance;
performing CEEMDAN decomposition on each monitoring signal processed in the step 1 to obtain each intrinsic modal component of each monitoring signal, and calculating the energy value of each modal component as the intrinsic modal energy characteristic of the health degradation state of the mechanical equipment;
and step 3: respectively fusing the time domain characteristics, the power spectrum characteristics and the intrinsic mode energy characteristics of the monitoring signals obtained in the step (2) by utilizing a GSOM (global system for mobile communications) network to obtain time domain fusion characteristics, power spectrum fusion characteristics and intrinsic mode energy fusion characteristics of the monitoring signals;
the process of fusing the time domain characteristics, the power spectrum characteristics and the eigen-mode energy characteristics by using the GSOM network in the step 3 is as follows:
step 3.1: selecting one of the three characteristics of the multiple monitoring signal data under the normal state of the mechanical equipment part to establish a training data set Dn;
Step 3.2: constructing a GSOM network model, and initializing parameters;
step 3.3: inputting training data set D to GSOM network modelnAnd setting a growth threshold distance dmax;
Step 3.4. randomly and unrepeatedly selecting all time domain characteristics x (t) ∈ D of the mechanical equipment at a certain moment from the training data setn(ii) a Based on the principle that the neuron with the minimum Euclidean distance to x (t) is the winning neuron, i.e.Determining winning neurons in the GSOM network model;
step 3.5: judging whether new neurons need to grow or not; if ol(x)(x)>dmaxIf the current network structure is not enough to describe the characteristics of the input data, a new neuron needs to be added in the GSOM network model, and the step 3.6 is continued; otherwise, returning to the step 3.4;
step 3.6: setting the feedforward connection weight of the new neuron as x (t)TCalculating the distance from the new neuron to all other neurons in the specified distance range (a)1×dmax,a2×dmax) Searching adjacent neurons of the new neuron internally, and establishing an adjacent relation; wherein, a1<a2,a1And a2Is a preset range parameter;
step 3.7: repeating steps 3.4 to 3.6 until training DnAfter all the data are stored, a well-trained GSOM network model is obtained;
step 3.8: inputting the data of various monitoring signals to be fused into the GSOM network model trained in the step 3.7 in sequence according to the time sequence, and respectively obtaining the feature values of the fused detection signals;
and 4, step 4: training the GenSVM model by using the fusion characteristics of the monitoring signals obtained in the step (3);
the testing phase comprises the following steps:
and 5: and (3) respectively extracting the characteristics of the various monitoring signals according to the step (2) and the step (3) and respectively performing characteristic fusion on the various monitoring signals, and inputting the various monitoring signals into the GenSVM model trained in the step (4) to obtain a health degradation state identification result.
2. The method for identifying the health degradation state of the mechanical equipment as claimed in claim 1, wherein the training process of the GenSVM in the step 4 is as follows:
step 4.1: establishing a performance decline characteristic matrix F by utilizing fusion characteristic data of various monitoring signals of mechanical equipment;
F=[F1,F2,F3,L]
wherein, F1、F2And F3Respectively representing a time domain fusion feature, a power spectrum fusion feature and an eigenmode energy fusion feature, and F1,F2,F3∈Rp×mP is the number of sampling points, m represents the number of monitoring signal types, Rp×mA real number matrix representing p × m dimension, L a mechanical equipment performance degradation state label, L ∈ Rp×1;
Step 4.2: randomly dividing the performance decline characteristic matrix F into training samples and check samples;
step 4.3: inputting the training sample into a GenSVM model, and mapping data in the input training sample to a high-dimensional space by a kernel function in the GenSVM model, wherein the data are expressed as xi→Φi;
Step 4.4: will phiiExpressed in K-1 dimensional simplex space as:
s′=Φ′iW+T′
wherein W is a weight matrix, T biased transition vector; thereby constructing a classification boundary;
step 4.5: establish a loss function that combines all training samples:
where ρ isiIs an optional object weight, hpWeights, G, representing misclassification errors calculated based on the Huber hinge functionkThe representation is a data set belonging to the kth state, K is the health degradation state number of the mechanical equipment, lambda trW' W is a penalty term for avoiding overfitting, and lambda > 0 is a regularization parameter;
step 4.6: gradually minimizing the loss function, and finishing the training process when the loss function meets the following conditions:
Lt-Lt-1<∈Lt-1
wherein L istAnd Lt-1Representing the loss function values of L (W, T) at the current moment and the last moment, ∈ is a preset stopping parameter;
step 4.7: verifying the GenSVM model trained in the step 4.6 by using a verification sample, and if the precision meets the requirement, performing a step 5; and if the precision does not meet the requirement, re-executing the steps 4.3 to 4.6.
3. The method for identifying the health degradation state of mechanical equipment according to claim 2, wherein the loss function is gradually minimized in step 4.6 by means of iterative optimization, and the steps are as follows:
step 4.6.1: definition V ═ T W']' and L ═ L (W, T), the following optimization equation was constructed
Wherein the content of the first and second substances,as a support point, VtIs the V value at the time t;
4. A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, implements the method of any one of claims 1 to 3.
5. A mechanical device health degradation state recognition device, comprising the computer-readable storage medium of claim 4 and a processor for invoking and processing a computer program stored in the computer-readable storage medium.
6. A mechanical device health degradation state identification system, comprising: the device comprises a processor, a training program module, a testing program module and a plurality of signal sensors; wherein the content of the first and second substances,
the processor is used for calling the training program module to execute the training phase of any one of claims 1 to 3 and calling the testing program module to execute the testing phase of any one of claims 1 to 3;
the signal sensors are used for acquiring various preset monitoring signals according to the type of the equipment to be tested and uploading the monitoring signals to the processor for training and testing.
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 CN109597315A (en) | 2019-04-09 |
CN109597315B true 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) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020208783A1 (en) * | 2019-04-11 | 2020-10-15 | 三菱電機株式会社 | 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 |
CN110561193B (en) * | 2019-09-18 | 2020-09-29 | 杭州友机技术有限公司 | Cutter wear assessment and monitoring method and system based on feature fusion |
US20220350691A1 (en) * | 2019-12-30 | 2022-11-03 | Jiangsu Nangao Intelligent Equipment Innovation Center Co., Ltd. | Fault prediction system based on sensor data on numerical control machine tool and method therefor |
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 |
CN112034789B (en) * | 2020-08-25 | 2021-10-15 | 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 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6948381B1 (en) * | 2002-04-09 | 2005-09-27 | Rockwell Automation Technologies, Inc. | System and method for sensing torque on a rotating shaft |
CN103868690B (en) * | 2014-02-28 | 2017-02-01 | 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 |
CN107563414B (en) * | 2017-08-14 | 2018-05-29 | 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 |
CN108507787B (en) * | 2018-06-28 | 2024-03-15 | 山东大学 | Wind power gear speed increasing box fault diagnosis test platform and method based on multi-feature fusion |
-
2018
- 2018-10-31 CN CN201811281339.8A patent/CN109597315B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN109597315A (en) | 2019-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109597315B (en) | Method, equipment and system for identifying health degradation state of mechanical equipment | |
D’Addona et al. | Tool-wear prediction and pattern-recognition using artificial neural network and DNA-based computing | |
CN109918752B (en) | Mechanical fault diagnosis method, equipment and medium based on migration convolution neural network | |
Kan et al. | A review on prognostic techniques for non-stationary and non-linear rotating systems | |
CN110647943B (en) | Cutting tool wear monitoring method based on evolution data cluster analysis | |
CN110322048A (en) | A kind of production logistics conveying equipment failure method for early warning | |
Gong et al. | A fast anomaly diagnosis approach based on modified CNN and multisensor data fusion | |
Liang et al. | Data-driven anomaly diagnosis for machining processes | |
CN110795690A (en) | Wind power plant operation abnormal data detection method | |
CN112518425B (en) | Intelligent machining cutter wear prediction method based on multi-source sample migration reinforcement learning | |
CN108857577A (en) | Cutting-tool wear state monitoring method and equipment | |
CN112633098A (en) | Fault diagnosis method and system for rotary machine and storage medium | |
CN114273977A (en) | MES-based cutter wear detection method and system | |
CN112284440A (en) | Sensor data deviation self-adaptive correction method | |
CN111126255A (en) | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm | |
CN112207631B (en) | Method for generating tool detection model, method, system, device and medium for detecting tool detection model | |
WO2007067521A1 (en) | Use of sequential clustering for instance selection in machine condition monitoring | |
Hassan et al. | Benchmarking of pattern recognition techniques for online tool wear detection | |
Yan et al. | A hybrid method for on-line performance assessment and life prediction in drilling operations | |
CN116010900A (en) | Multi-scale feature fusion gearbox fault diagnosis method based on self-attention mechanism | |
CN111582406A (en) | Power equipment state monitoring data clustering method and system | |
CN111414943A (en) | Anomaly detection method based on mixed hidden naive Bayes model | |
CN114962390A (en) | Hydraulic system fault diagnosis method and system and working machine | |
CN112783138B (en) | Intelligent monitoring and abnormity diagnosis method and device for processing stability of production line equipment | |
CN117633688A (en) | Large-scale power data anomaly detection method based on ridge regression-k-means clustering-LOF-LSTM fusion 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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |