CN114266289A - Complex equipment health state assessment method - Google Patents
Complex equipment health state assessment method Download PDFInfo
- Publication number
- CN114266289A CN114266289A CN202111404374.6A CN202111404374A CN114266289A CN 114266289 A CN114266289 A CN 114266289A CN 202111404374 A CN202111404374 A CN 202111404374A CN 114266289 A CN114266289 A CN 114266289A
- Authority
- CN
- China
- Prior art keywords
- data
- state
- complex equipment
- normal state
- evaluated
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses a complex equipment health state evaluation method, which evaluates data to be evaluated and judges the health grade of complex equipment by establishing a complex equipment health state clustering center training model, wherein normal state data of the complex equipment are fused into a group of reference normal state data, and then characteristic parameters, namely cross correlation coefficients, agglomeration coefficients and spectral distance indexes, are calculated to serve as three-dimensional characteristic vector coordinates of each group of training data. And performing clustering analysis on the feature vector array to obtain a clustering center training model, calculating the membership degree of the data to be evaluated to the clustering center in a normal state according to the Euclidean distance from the data to be evaluated to the clustering center, evaluating the health grade, performing sampling mechanism analysis on the data to be evaluated, judging the accuracy of an evaluation result, and simultaneously adding the extracted data to be evaluated, which is evaluated accurately and is finished, to the training data of the complex equipment, so that the accuracy and the stability of subsequent evaluation work are improved.
Description
Technical Field
The invention relates to the technical field of equipment health management, in particular to a method for evaluating the health state of complex equipment.
Background
With the change of war forms, the requirements on the complexity of equipment are higher and higher, and the complex equipment refers to equipment with complex composition relationship, complex behavior, complex interaction relationship between subsystems of the system, complex interaction relationship between the system and the environment thereof and complex energy exchange, and has important strategic and tactical values due to the outstanding functional performance and the influence on the success or failure of tasks.
With the increasingly complex functional structure of the complex equipment, the experienced use profile becomes complex gradually, and if the multidimensional monitoring of the operation data of the complex equipment is realized in the use and maintenance process, the health state of the complex equipment is timely and accurately evaluated, the maintenance decision time can be greatly saved, and the actual combat level of the equipment is obviously improved.
In the present phase, most of methods for evaluating the health state of equipment perform time domain or frequency domain feature extraction based on equipment signals. However, the time-frequency domain signal processing technique requires preprocessing of the original signal and the calculation process is rather complicated. If the real-time diagnosis and analysis of the acquired signals are realized on site, the original signals are not preprocessed, and the health assessment is realized by directly utilizing the original signals through simple time domain, frequency domain or time-frequency domain parameters, which is the most ideal, but the method is difficult to realize in practice. The existing method is to establish the incidence relation between the equipment running state information and the reliability according to the mapping relation between the running state information and the running reliability and directly calculate the original signal through a mapping model. However, the calculation by using a single mapping function requires determining a function matched with specific equipment and actual conditions in advance, and the workload is large. For example, when the normal and fault states of the equipment are not distinguished obviously on the time domain signal, the effect of selecting the cross-correlation function for health assessment is not good. Moreover, single set of reference data selected during traditional reliability function calculation has contingency and cannot completely represent the health state of equipment, so that the accuracy of health assessment is reduced.
Disclosure of Invention
In view of the above, the invention provides a method for evaluating the health state of complex equipment, which can perform economic and efficient health state evaluation by using original data of the complex equipment aiming at different complex equipment, thereby realizing effective maintenance decision according to the situation.
The specific technical scheme of the invention is as follows:
a complex equipment health status assessment method, comprising:
step one, normal state data in training data of complex equipment are fused into a group of reference normal state data through data weighting fusion;
step two, calculating the cross correlation coefficient, the agglomeration coefficient and the spectral distance index of each group of training data according to the training data of the complex equipment and the reference normal state data;
step three, taking the cross correlation coefficient, the agglomeration coefficient and the spectral distance index as three-dimensional feature vector coordinates of each set of training data to obtain a feature vector array of the complex equipment; performing clustering analysis on the feature vector array to obtain a complex equipment health state clustering center training model which comprises a normal state clustering center and a fault state clustering center;
calculating the three-dimensional characteristic vector coordinates of each group of data to be evaluated according to the data to be evaluated of the complex equipment and the reference normal state data; and evaluating the health state grade of the complex equipment according to Euclidean distances from the three-dimensional characteristic vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center.
Further, in the step one, the step of fusing the normal state data in the training data of the complex equipment into a set of reference normal state data through data weighted fusion is as follows:
based on cross-correlation function, carrying out pairwise cross-correlation operation on the normal state data to obtain cross-correlation coefficient R of any two groups of normal state dataab：
Wherein x isa、xbRespectively represents the normal state data of the a-th group and the b-th group, and a is equal to [1, r ∈],b∈[1,r]A is not equal to b, and r is a positive integer and represents the total number of normal state data;
indicating normal state data xaThe standard deviation of (a) is determined,
indicating normal state data xbThe standard deviation of (a) is determined,
indicating normal state data xaAnd normal state data xbA positive covariance function of;
total correlation energy E of the a group data and other group dataaExpressed as:
weighted value PaProportional to the total correlation energy of each set of data, i.e. P1:P2:…:Pr＝E1:E2:…:ErDue to P1+P2+…+P r1, the set of baseline normal state data obtained is:
X＝P1x1+P2x2+…+Prxr。
further, in the second step, the spectral distance index is calculated in the following manner:
h represents a spectral distance index, alpha is a sensitivity coefficient and is not less than 0, and when the complex equipment fails, the more serious the damage degree is, the smaller the alpha value is; the more slight the damage degree is, the larger the value of alpha is; j. the design is a squarex,yIs the J divergence between the baseline normal state data x (t) and the training data y (t);
wherein S isx(k) And Sy(k) Self-power spectra of data x (t) and y (t), respectively, t representing a time series; n is the number of power spectral lines, and k belongs to N.
Further, in the third step, a fuzzy C-means clustering analysis method is adopted to perform clustering analysis on the feature vector array;
after the normal state clustering center and the fault state clustering center are obtained, evaluating the effectiveness of clustering analysis through a partition coefficient, if the clustering analysis effect is good, continuing to evaluate the data to be evaluated, if the clustering analysis effect is not good, acquiring the training data again, and performing clustering analysis again; the partition coefficient is expressed as:
wherein M is the total number of the sample arrays, c is the number of the cluster types, i.e. the cluster centers, muimThe membership degree of the mth sample to the ith class is set as i belongs to c, and M belongs to M; the closer the partition coefficient is to 1, the better the cluster analysis effect is.
Further, in the fourth step, the judging that the health state grade of the complex equipment is as follows according to the euclidean distance from the three-dimensional feature vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center: respectively calculating Euclidean distances from the three-dimensional characteristic vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center, and calculating the membership degree of the data to be evaluated to the normal state clustering center according to the Euclidean distances:
wherein mu represents membership degree, and the value range of mu is [0, 1%]，D1Euclidean distance D from three-dimensional feature vector coordinates representing data to be evaluated to normal state cluster center2Representing Euclidean distance from three-dimensional characteristic vector coordinates of data to be evaluated to a fault state clustering center, wherein i represents the ith clustering center;
judging health status grades according to the numerical values of the membership degrees, wherein the health status grades comprise five grades, and the five grades are respectively as follows: fault state, pseudo fault state, usable state, better state and good state; wherein the fault state is when the membership value is [0,0.2), the pseudo-fault state is when the membership value is [0.2, 0.4), the usable state is when the membership value is [0.4,0.6), the better state is when the membership value is [0.6, 0.8), and the better state is when the membership value is [0.8,1 ].
Further, after the health state grade of the complex equipment is evaluated, sampling mechanism analysis is carried out on the data to be evaluated according to the evaluation result, the accuracy of the evaluation result is judged, the extracted data to be evaluated, which is evaluated and accurate, is added into the training data of the complex equipment, and the steps from the first step to the third step are repeated, so that an optimized complex equipment health state clustering center training model is obtained for the next evaluation.
Further, the optimized complex equipment health state clustering center training model comprises a normal state clustering center, a fault state clustering center, a usable state clustering center and a better state clustering center according to different health state grades of the extracted data to be evaluated.
Further, in the step one, the training data is obtained by performing data washing and label processing on historical data of the complex equipment.
Has the advantages that:
(1) according to the method for evaluating the health state of the complex equipment, the health grade of the complex equipment is judged by evaluating data to be evaluated by establishing a complex equipment health state clustering center training model, wherein normal state data of the complex equipment are fused into a group of reference normal state data, so that characteristic parameters, namely cross correlation coefficient, agglomeration coefficient and spectral distance index, are calculated, the accidental property of the traditional method of selecting single group of data as reference data is reduced, the health state of the complex equipment can be represented better, and the accuracy of health evaluation is improved; the three-dimensional characteristic vector coordinates of each group of training data are constructed, the characteristic vector array is subjected to cluster analysis, the multi-dimensional evaluation of the health state of complex equipment can be better performed, the raw data can be used as the data to be evaluated to calculate the cross-correlation coefficient, the agglomeration coefficient and the spectral distance index, the preprocessing is not needed, the evaluation efficiency of the health state is improved while the calculated amount of the evaluation process is reduced, when a certain characteristic parameter fails due to the type of equipment or some actual environmental conditions, the traditional evaluation method adopting a single mapping function is easy to fail, and the three characteristic parameters can be used for improving the application range of the evaluation method and enhancing the stability of the evaluation process.
(2) When the spectral distance index is calculated through the spectral distance index function, the sensitivity coefficient is set according to the performance degradation degree of the complex equipment, the value range of the spectral distance index can be finely adjusted according to signals of different types of equipment, the uniformity of the value ranges of the cross-correlation coefficient, the agglomeration coefficient and the spectral distance index is kept, normalization processing is not needed, the complex calculation process is reduced, the cluster analysis effect is improved, and the accuracy of health grade evaluation is further improved.
(3) The effectiveness of the cluster analysis is evaluated through the partition coefficient, whether the next evaluation is carried out or the cluster analysis is carried out again is determined, and the accuracy of the health grade evaluation can be ensured.
(4) The health status levels are divided into five levels, which are: compared with the normal grade and the fault grade, the equipment can be determined whether to continue to be used according to the actual health state grade and the use requirement of the equipment, so that the equipment in the fault simulating state, the available state and the good state is still possible to continue to be used, the service life of the equipment is prolonged to the maximum extent, meanwhile, a more specific optional maintenance decision is determined according to different grades, and the maintenance efficiency is also improved.
(5) After the evaluation is completed, the data to be evaluated is subjected to sampling mechanism analysis, the accuracy of the evaluation result is judged, the test failure or equipment cost loss caused by evaluation errors can be prevented, and meanwhile, the extracted data to be evaluated, which is subjected to evaluation and is accurate in evaluation, is added into the training data of the complex equipment, so that the accuracy and the stability of the subsequent evaluation work can be improved.
(6) When the accurate data to be evaluated is added into the training data according to the health state grades, the accurate data to be evaluated is not directly added into the normal state data or the fault state data, new classification of the training data is established according to the specific health state grades, new clustering analysis is further carried out, clustering centers with more health state grades are obtained, the health state grades of the data to be evaluated can be determined more quickly and better in the subsequent evaluation process, and maintenance decisions can be made more efficiently.
Drawings
FIG. 1 is a flow chart of a complex equipment health state cluster center training model before optimization in the present invention.
FIG. 2 is a flow chart of the health status evaluation of the complex equipment according to the present invention.
Detailed Description
The invention provides a complex equipment health state evaluation method, which evaluates data to be evaluated and judges the health grade of complex equipment by establishing a complex equipment health state clustering center training model, wherein normal state data of the complex equipment are fused into a group of reference normal state data, and characteristic parameters, namely cross correlation coefficient, agglomeration coefficient and spectral distance index, are calculated. And then, three-dimensional feature vector coordinates of each group of training data are constructed, the feature vector arrays are subjected to clustering analysis, a normal state clustering center and a fault state clustering center are obtained, and the health state of the complex equipment is subjected to multi-dimensional evaluation. And the original data can be used as the data to be evaluated to calculate the cross-correlation coefficient, the agglomeration coefficient and the spectral distance index without preprocessing. After the evaluation is completed, sampling mechanism analysis is carried out on the data to be evaluated, the accuracy of the evaluation result is judged, the extracted data to be evaluated, which is evaluated and accurate, is added into the training data of the complex equipment, a new clustering center can be generated through a complex equipment health state clustering center training model, and the accuracy and the stability of the subsequent evaluation work are improved.
The training data of the complex equipment often refers to data that vibration signals, temperature or voltage, etc. of the complex equipment can be used to assess the health status of the complex equipment.
The invention is described in detail below by way of example with reference to the accompanying drawings.
As shown in fig. 1, a flow chart of a complex equipment health state clustering center training model before optimization is a flow for establishing a complex equipment health state clustering center training model before first evaluation, and the specific process includes the following steps:
step one, normal state data in training data of complex equipment are fused into a group of reference normal state data through data weighting fusion.
The training data of the complex equipment has two types, namely normal state data and fault state data, and the normal state data of the complex equipment is fused into a group of reference normal state data through data weighting fusion.
Wherein the training data is obtained by performing data washing and label processing on historical data of the complex equipment.
Collecting current mastered historical data of the complex equipment, and obtaining more ideal data by data cleaning means such as abnormal data elimination and missing value processing, wherein the abnormal data elimination method can utilize Lauda rule, Shouyuler criterion, Dixon criterion and the like; in the missing value processing method, when the data lengths are different, the data lengths may be made uniform by signal reconstruction or by thinning-out, and when the data are missing, the missing value may be made up by omitting the data of the missing item, by specializing the missing value, or by interpolation.
And constructing a training database by using ideal training data obtained after data cleaning, wherein the training data in the database needs to be subjected to label processing, and the data is divided into normal state data and fault state data.
And then, normal state data of the complex equipment are fused into a group of reference normal state data through data weighted fusion. The data fusion method based on the cross-correlation function adjusts the weighted value according to the correlation function between the data signals, and performs correlation calculation on any data signal and other signals, wherein the greater the correlation degree is, the greater the weighted value is given. Generally, the more accurately information reflecting the equipment state, the higher the corresponding weight value.
Firstly, based on cross-correlation function, carrying out pairwise cross-correlation operation on normal state data to obtain cross-correlation coefficient R of any two groups of normal state dataab：
Wherein x isa、xbRespectively represents the normal state data of the a-th group and the b-th group, and a is equal to [1, r ∈],b∈[1,r]A is not equal to b, and r is a positive integer and represents the total number of normal state data;
indicating normal state data xaThe standard deviation of (a) is determined,
indicating normal state data xbThe standard deviation of (a) is determined,
indicating normal state data xaAnd normal state data xbA positive covariance function of;
total correlation energy E of the a group data and other group dataaExpressed as:
weighted value PaProportional to the total correlation energy of each set of data, i.e. P1:P2:…:Pr＝E1:E2:…:ErDue to P1+P2+…+P r1, the set of baseline normal state data obtained is:
X＝P1x1+P2x2+…+Prxr。
and step two, calculating the cross correlation coefficient, the agglomeration coefficient and the spectral distance index of each group of training data according to the training data of the complex equipment and the reference normal state data.
And performing cross-correlation function calculation, aggregation function calculation and spectral distance index function calculation on the training data of the complex equipment and the reference normal state data to obtain a cross-correlation coefficient, an aggregation coefficient and a spectral distance index of each group of training data.
(1) Cross correlation function
For the same equipment, the subsequent operation state of the equipment is judged by taking the datum normal state data as a reference, the reliability of the operation state is represented by adopting a cross-correlation coefficient, and the process from normal to fault is represented by the cross-correlation coefficient from large to small:
wherein x represents reference normal state data; y represents training data; t represents a time series, Cx,y(τ) is a positive covariance function between the baseline normal state data x (t) and the training data y (t); deltaxAnd deltayThe standard deviations of the data x (t) and y (t), respectively.
(2) Agglomeration function
For the same equipment, the subsequent running state of the equipment is judged by taking the datum normal state data as a reference, and the reliability of the running state is represented by adopting an agglomeration function:
wherein x represents reference normal state data; y represents training data; sx,y(f) Is the cross-power spectrum between the reference normal state data x (t) and the training data y (t); sx(f) And Sy(f) Are the self-power spectra of data x (t) and y (t), respectively.
(3) Spectral distance index function
For the same equipment, the subsequent running state of the equipment is judged by taking the datum normal state data as a reference, and the reliability of the running state is represented by adopting a spectral distance index-J divergence method:
h represents a spectral distance index, alpha is a sensitivity coefficient and is more than or equal to 0, and when the complex equipment completely fails, the more serious the damage degree is, the smaller the value of alpha is; the more slight the damage degree is, the larger the value of alpha is; j. the design is a squarex,yIs the J divergence between the baseline normal state data x (t) and the training data y (t);
wherein S isx(k) And Sy(k) Self-power spectra of data x (t) and y (t), respectively, t representing a time series; n is the number of power spectral lines, and k belongs to N.
In general, when the reliability function is calculated, the reference data selected as the reference has unicity, and the quality of the reference data often affects the result of feature extraction, so that the uncertainty of single group of reference normal state data is eliminated by introducing a data fusion method in the step one, and in addition, because of the setting of sensitivity coefficients, the value ranges of three types of feature parameters obtained by calculation are all [0, 1], and normalization processing is not needed.
Step three, taking the cross correlation coefficient, the agglomeration coefficient and the spectral distance index as three-dimensional feature vector coordinates of each set of training data to obtain a feature vector array of the complex equipment; and carrying out clustering analysis on the feature vector array to obtain a complex equipment health state clustering center training model, which comprises a normal state clustering center and a fault state clustering center.
Constructing a three-dimensional characteristic vector coordinate according to the characteristic parameters of the cross correlation coefficient, the agglomeration coefficient and the spectral distance index, and mainly considering two reasons: firstly, redundancy caused by excessive characteristic parameters can be avoided, and clustering results can be visually displayed in a three-dimensional rectangular coordinate system; secondly, the change of the health state of the complex equipment sometimes cannot be completely reflected on the change of the single characteristic parameter, for example, when the normal state and the fault state of the equipment are different in the time domain signal characteristic parameter value, the value ranges may be completely overlapped in the frequency domain. Therefore, three types of characteristic parameters are selected to construct three-dimensional characteristic vector coordinates, and when one type or two types of characteristic parameters fail to reflect the health state, the rest characteristic parameters can still support subsequent clustering analysis, so that the robustness of the model is improved.
And carrying out clustering analysis on a feature vector array constructed according to the feature parameters of the normal state data and the fault state data of the complex equipment by using a fuzzy C mean value clustering analysis method.
In the fuzzy C-means clustering analysis algorithm, normal state and fault state data are mixed into a sample set X ═ X1,x2,…,xM]Each sample being a three-dimensional vector, i.e. xm＝[xm1,xm2,xm3]TWherein x ism1、xm2、xm3The method is characterized in that three types of characteristic parameters of the mth group of data, namely cross-correlation coefficients, agglomeration coefficients and spectral distance indexes, M is the total number of the sample array, and M belongs to M.
The sample array is divided into classes c. Defining a partition matrix U and a clustering center matrix V as follows:
U＝[μim](1≤i≤c,1≤m≤M)
V＝[v1,v2,…,vc](νi∈Rn)
wherein: v isiThe cluster center vector of the ith class; mu.simThe membership degree of the mth sample to the ith sample;
the membership degree needs to satisfy constraint conditions:
in one embodiment, the mixed sample data is divided into two classes (normal and fault), and the iteration termination threshold epsilon is 10-4The initial iteration count L is 1 and the maximum iteration count L is 500, and the matrix U is initially divided(0)Until U | | | U(l)-U(l-1)Stopping iterative computation when | | < epsilon or L equals to L, and training to obtain two clustering center coordinates which are respectively a normal state clustering center and a fault state clustering center. Here, iteration number setting is obtained, and the termination threshold setting is only used for illustrating some parameter settings in the process of cluster analysis, and is not used for limiting specific values.
After the normal state clustering center and the fault state clustering center are obtained, evaluating the effectiveness of clustering analysis through a partition coefficient, if the clustering analysis effect is good, continuing to evaluate the data to be evaluated, if the clustering analysis effect is not good, acquiring the training data again, and performing clustering analysis again; the partition coefficient is expressed as:
wherein M is the total number of the sample arrays, c is the number of the cluster types, i.e. the cluster centers, muimThe membership degree of the mth sample to the ith class is set as i belongs to c, and M belongs to M; the closer the partition coefficient is to 1, the better the cluster analysis effect is.
And obtaining a preliminary complex equipment health state clustering center training model through the first step to the third step, and performing evaluation work on data to be evaluated, wherein the data to be evaluated refers to measured data of the complex equipment and original data which are not subjected to preprocessing, such as filtering noise reduction, Fourier transform and the like.
The evaluation procedure was as follows:
calculating the three-dimensional characteristic vector coordinates of each group of data to be evaluated according to the data to be evaluated of the complex equipment and the reference normal state data; and evaluating the health state grade of the complex equipment according to Euclidean distances from the three-dimensional characteristic vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center.
And performing membership calculation on the data to be evaluated based on the normal state clustering center to obtain an evaluation result, namely the health state grade of the complex equipment.
As shown in fig. 2, a flow diagram of the complex equipment health state assessment before the complex equipment health state cluster center training model is optimized.
And taking the non-tag data monitored by the complex equipment in real time as data to be evaluated as input to evaluate the health state of the complex equipment. And (3) calculating a cross-correlation function, an aggregation function and a spectral distance index function of the data to be evaluated on the reference normal state data in the step one, extracting to obtain three types of characteristic parameters, and constructing a three-dimensional characteristic vector coordinate.
Respectively calculating Euclidean distances from the three-dimensional characteristic vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center, and calculating the membership degree of the data to be evaluated according to the Euclidean distances:
wherein mu represents membership degree, and the value range of mu is [0, 1%]，D1Euclidean distance D from three-dimensional feature vector coordinates representing data to be evaluated to normal state cluster center2And expressing Euclidean distance from the three-dimensional characteristic vector coordinates of the data to be evaluated to the fault state clustering center, wherein i represents the ith clustering center.
The health status levels include five levels, which are: fault state, pseudo fault state, usable state, better state and good state; wherein the fault state is when the membership value is [0,0.2), the pseudo-fault state is when the membership value is [0.2, 0.4), the usable state is when the membership value is [0.4,0.6), the better state is when the membership value is [0.6, 0.8), and the better state is when the membership value is [0.8,1 ].
After the health state grade of the complex equipment is evaluated, sampling mechanism analysis is carried out on data to be evaluated according to the evaluation result, the accuracy of the evaluation result is judged, the extracted data to be evaluated, which is evaluated accurately and is finished, is added into training data of the complex equipment, and the steps from the first step to the third step are repeated to obtain an optimized complex equipment health state clustering center training model for the next evaluation.
The optimized complex equipment health state clustering center training model comprises a normal state clustering center, a fault state clustering center, a usable state clustering center and a better state clustering center according to different health state grades of extracted data to be evaluated.
The above sampling mechanism analysis of the data to be evaluated means that, in the evaluation process, for the data evaluated in the same state, the data evaluated in the same state is also in the same state with a high probability of the actual situation, that is, the state of the data to be evaluated, that is, the actually measured data, judged in the same health state is fixed. Therefore, several groups or a plurality of groups of the data to be evaluated which are judged to be in the same health state can be extracted to verify whether the evaluation result is accurate, namely, the mechanism analysis is carried out, for the complex equipment health state clustering center training model, whether the evaluation result of the model is accurate can be known by using the result of the mechanism analysis, but in order to make an optional maintenance decision, namely, a maintenance decision determined according to the actual situation, the reason that the equipment is in the state needs to be analyzed, and therefore the whole process of the mechanism analysis is completed. And when the evaluation result of the data to be evaluated is in accordance with the actual state of the equipment, the evaluation result is considered to be accurate, at the moment, the extracted data to be evaluated, which is evaluated and is accurate in evaluation, is added into the training data of the complex equipment, and the steps from the first step to the third step are repeated to obtain an optimized complex equipment health state clustering center training model for the next evaluation. If the result of the evaluation of the data to be evaluated is the usable state, the extracted data is also the data of the usable state, therefore, the extracted data to be evaluated in the usable state, namely the measured data, is supplemented into the training data, the labels of the data are marked as the data of the usable state, a new optimized complex equipment health state cluster center training model can be obtained by repeating the steps from the first step to the third step, the training result of the complex equipment health state cluster center training model comprises three cluster centers, namely a normal state cluster center, a usable state cluster center and a fault state cluster center, so that the subsequent evaluation process is carried out, the efficiency is higher, the specific evaluation process is the same as the evaluation process comprising the two cluster centers, and the Euclidean distance from the data to be evaluated to each cluster center is calculated, and then calculating the membership degree, which will not be described in detail herein. When the obtained measured data is better data and the evaluation result is accurate, the optimized training model of the complex equipment health state clustering center comprises a normal state clustering center, a better state clustering center and a fault state clustering center. Similarly, through the processes of continuous evaluation, training and evaluation, the final complex equipment health state clustering center training model comprises a normal state clustering center, a fault state clustering center, a simulated fault state clustering center, a usable state clustering center and a better state clustering center. And if the data to be evaluated is in a good state, the data to be evaluated belongs to a normal state clustering center.
The method for evaluating the health state of the complex equipment is provided based on the current situation that the evaluation of the health state of the complex equipment lacks a fine quantification method, and has important significance for promoting the fault prediction and health management work of the complex equipment and promoting the development of the intelligent guarantee work of the complex equipment. The invention is based on limited complex equipment historical data samples, increases the ideal data sample amount by a data cleaning method, obtains datum normal state data by a data fusion method based on a correlation function, and utilizes three types of reliability functions: the data characteristic parameters are extracted through a cross-correlation function in a time domain, a condensation function in a frequency domain and a J divergence spectrum distance index function, the equipment health state is comprehensively considered from different dimensions to reflect the characteristic change on the data signals, and compared with the method for carrying out health assessment through a single parameter, the method is higher in robustness. The method is closer to engineering practice and has good application prospect.
The above embodiments only describe the design principle of the present invention, and the shapes and names of the components in the description may be different without limitation. Therefore, a person skilled in the art of the present invention can modify or substitute the technical solutions described in the foregoing embodiments; such modifications and substitutions do not depart from the spirit and scope of the present invention.
Claims (8)
1. A method for evaluating the health status of complex equipment is characterized by comprising the following steps:
step one, normal state data in training data of complex equipment are fused into a group of reference normal state data through data weighting fusion;
step two, calculating the cross correlation coefficient, the agglomeration coefficient and the spectral distance index of each group of training data according to the training data of the complex equipment and the reference normal state data;
step three, taking the cross correlation coefficient, the agglomeration coefficient and the spectral distance index as three-dimensional feature vector coordinates of each set of training data to obtain a feature vector array of the complex equipment; performing clustering analysis on the feature vector array to obtain a complex equipment health state clustering center training model which comprises a normal state clustering center and a fault state clustering center;
calculating the three-dimensional characteristic vector coordinates of each group of data to be evaluated according to the data to be evaluated of the complex equipment and the reference normal state data; and evaluating the health state grade of the complex equipment according to Euclidean distances from the three-dimensional characteristic vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center.
2. The method for evaluating the health status of complex equipment according to claim 1, wherein in the first step, the normal status data in the training data of the complex equipment are fused into a set of reference normal status data through data weighted fusion, and the set of reference normal status data are as follows:
based on cross-correlation function, carrying out pairwise cross-correlation operation on the normal state data to obtain cross-correlation coefficient R of any two groups of normal state dataab：
Wherein x isa、xbRespectively represents the normal state data of the a-th group and the b-th group, and a is equal to [1, r ∈],b∈[1,r]A is not equal to b, and r is a positive integer and represents the total number of normal state data;
indicating normal state data xaThe standard deviation of (a) is determined,
indicating normal state data xbThe standard deviation of (a) is determined,
indicating normal state data xaAnd normal state data xbA positive covariance function of;
total correlation energy E of the a group data and other group dataaExpressed as:
weighted value PaProportional to the total correlation energy of each set of data, i.e. P1:P2:…:Pr＝E1:E2:…:ErDue to P1+P2+…+Pr1, the set of baseline normal state data obtained is:
X＝P1x1+P2x2+…+Prxr。
3. the method for evaluating the health status of complex equipment according to claim 1, wherein in the second step, the spectral distance index is calculated by:
h represents a spectral distance index, alpha is a sensitivity coefficient and is not less than 0, and when the complex equipment fails, the more serious the damage degree is, the smaller the alpha value is; the more slight the damage degree is, the larger the value of alpha is; j. the design is a squarex,yIs the J divergence between the baseline normal state data x (t) and the training data y (t);
wherein S isx(k) And Sy(k) Self-power spectra of data x (t) and y (t), respectively, t representing a time series; n is the number of power spectral lines, and k belongs to N.
4. The method for evaluating the health status of complex equipment according to claim 1, wherein in step three, the feature vector array is subjected to cluster analysis by using a fuzzy C-means cluster analysis method;
after the normal state clustering center and the fault state clustering center are obtained, evaluating the effectiveness of clustering analysis through a partition coefficient, if the clustering analysis effect is good, continuing to evaluate the data to be evaluated, if the clustering analysis effect is not good, acquiring the training data again, and performing clustering analysis again; the partition coefficient is expressed as:
wherein M is the total number of the sample arrays, c is the number of the cluster types, i.e. the cluster centers, muimThe membership degree of the mth sample to the ith class is set as i belongs to c, and M belongs to M; the closer the partition coefficient is to 1, the better the cluster analysis effect is.
5. The method for evaluating the health status of the complex equipment according to claim 1, wherein in step four, the health status grade of the complex equipment is determined according to the euclidean distance from the three-dimensional feature vector coordinates of the data to be evaluated to the normal status cluster center and the fault status cluster center: respectively calculating Euclidean distances from the three-dimensional characteristic vector coordinates of the data to be evaluated to the normal state clustering center and the fault state clustering center, and calculating the membership degree of the data to be evaluated to the normal state clustering center according to the Euclidean distances:
wherein mu represents membership degree, and the value range of mu is [0, 1%]，D1Euclidean distance D from three-dimensional feature vector coordinates representing data to be evaluated to normal state cluster center2Representing Euclidean distance from three-dimensional characteristic vector coordinates of data to be evaluated to a fault state clustering center, wherein i represents the ith clustering center;
judging health status grades according to the numerical values of the membership degrees, wherein the health status grades comprise five grades, and the five grades are respectively as follows: fault state, pseudo fault state, usable state, better state and good state; wherein the fault state is when the membership value is [0,0.2), the pseudo-fault state is when the membership value is [0.2, 0.4), the usable state is when the membership value is [0.4,0.6), the better state is when the membership value is [0.6, 0.8), and the better state is when the membership value is [0.8,1 ].
6. The method for evaluating the health status of the complex equipment according to claim 1, wherein after the evaluation of the health status grade of the complex equipment, the data to be evaluated is subjected to sampling mechanism analysis in combination with the evaluation result, the accuracy of the evaluation result is judged, the extracted data to be evaluated, which is evaluated completely and accurately, is added to the training data of the complex equipment, and the steps one to three are repeated to obtain an optimized complex equipment health status clustering center training model for the next evaluation.
7. The complex equipment state of health assessment method of claim 6, wherein the optimized complex equipment state of health cluster center training model further comprises a failure state simulation cluster center, a usable state cluster center and a better state cluster center, while including a normal state cluster center and a failure state cluster center, according to the difference of the extracted state of health levels of the data to be assessed.
8. The method for evaluating the health status of complex equipment according to claim 1, wherein in the first step, the training data is obtained by performing data washing and labeling on historical data of complex equipment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111404374.6A CN114266289A (en) | 2021-11-24 | 2021-11-24 | Complex equipment health state assessment method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111404374.6A CN114266289A (en) | 2021-11-24 | 2021-11-24 | Complex equipment health state assessment method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114266289A true CN114266289A (en) | 2022-04-01 |
Family
ID=80825422
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111404374.6A Pending CN114266289A (en) | 2021-11-24 | 2021-11-24 | Complex equipment health state assessment method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114266289A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114800036A (en) * | 2022-06-24 | 2022-07-29 | 成都飞机工业（集团）有限责任公司 | Equipment health state assessment method |
| CN115049299A (en) * | 2022-07-01 | 2022-09-13 | 湖北博华自动化系统工程有限公司 | Fault diagnosis and health assessment method based on alarm time sequence |
| CN115659188A (en) * | 2022-12-29 | 2023-01-31 | 四川观想科技股份有限公司 | Equipment health management abnormity positioning method based on event correlation |
| CN117435939A (en) * | 2023-12-14 | 2024-01-23 | 广东力宏微电子有限公司 | IGBT health state evaluation method based on big data |
-
2021
- 2021-11-24 CN CN202111404374.6A patent/CN114266289A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114800036A (en) * | 2022-06-24 | 2022-07-29 | 成都飞机工业（集团）有限责任公司 | Equipment health state assessment method |
| CN114800036B (en) * | 2022-06-24 | 2022-10-25 | 成都飞机工业（集团）有限责任公司 | Equipment health state assessment method |
| CN115049299A (en) * | 2022-07-01 | 2022-09-13 | 湖北博华自动化系统工程有限公司 | Fault diagnosis and health assessment method based on alarm time sequence |
| CN115659188A (en) * | 2022-12-29 | 2023-01-31 | 四川观想科技股份有限公司 | Equipment health management abnormity positioning method based on event correlation |
| CN115659188B (en) * | 2022-12-29 | 2023-06-23 | 四川观想科技股份有限公司 | Event correlation-based equipment health management abnormality positioning method |
| CN117435939A (en) * | 2023-12-14 | 2024-01-23 | 广东力宏微电子有限公司 | IGBT health state evaluation method based on big data |
| CN117435939B (en) * | 2023-12-14 | 2024-03-08 | 广东力宏微电子有限公司 | IGBT health state evaluation method based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114266289A (en) | Complex equipment health state assessment method | |
| CN103033362B (en) | Gear fault diagnosis method based on improving multivariable predictive models | |
| CN107454105B (en) | Multidimensional network security assessment method based on AHP and grey correlation | |
| CN108520357A (en) | A kind of method of discrimination, device and the server of line loss abnormal cause | |
| CN113011530B (en) | Intelligent ammeter fault prediction method based on multi-classifier fusion | |
| CN112039903B (en) | Network security situation assessment method based on deep self-coding neural network model | |
| CN116450399B (en) | Fault diagnosis and root cause positioning method for micro service system | |
| CN107679734A (en) | It is a kind of to be used for the method and system without label data classification prediction | |
| CN111538311B (en) | Flexible multi-state self-adaptive early warning method and device for mechanical equipment based on data mining | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN108919059A (en) | A kind of electric network failure diagnosis method, apparatus, equipment and readable storage medium storing program for executing | |
| CN115828140A (en) | Neighborhood mutual information and random forest fusion fault detection method, system and application | |
| CN110706213A (en) | Bridge cluster structure damage judgment method based on strain response cumulative distribution function difference | |
| CN112700324A (en) | User loan default prediction method based on combination of Catboost and restricted Boltzmann machine | |
| CN103577676A (en) | Grey weighting method for sewage treatment process comprehensive evaluation | |
| CN115438726A (en) | Device life and fault type prediction method and system based on digital twin technology | |
| CN111639882A (en) | Deep learning-based power utilization risk judgment method | |
| CN114819777A (en) | Enterprise sales business analysis and management system based on digital twin technology | |
| CN114611372A (en) | Industrial equipment health prediction method based on Internet of things edge calculation | |
| CN114330486A (en) | Power system bad data identification method based on improved Wasserstein GAN | |
| CN115407753A (en) | Industrial fault diagnosis method for multivariate weighted ensemble learning | |
| CN115935187B (en) | Nuclear sensitivity alignment network-based mechanical fault diagnosis method under variable working conditions | |
| CN116776209A (en) | Method, system, equipment and medium for identifying operation state of gateway metering device | |
| CN108053093A (en) | A kind of k- neighbour's method for diagnosing faults based on the conversion of average influence Value Data | |
| CN114626594A (en) | Medium-and-long-term electric quantity prediction method based on cluster analysis and deep learning |
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 |