CN118035815A - Compressor health state degradation identification method, device and storage medium - Google Patents
Compressor health state degradation identification method, device and storage medium Download PDFInfo
- Publication number
- CN118035815A CN118035815A CN202410025382.7A CN202410025382A CN118035815A CN 118035815 A CN118035815 A CN 118035815A CN 202410025382 A CN202410025382 A CN 202410025382A CN 118035815 A CN118035815 A CN 118035815A
- Authority
- CN
- China
- Prior art keywords
- compressor
- health state
- feature
- health
- characteristic
- 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
- 230000036541 health Effects 0.000 title claims abstract description 135
- 238000000034 method Methods 0.000 title claims abstract description 63
- 230000015556 catabolic process Effects 0.000 title claims abstract description 33
- 238000006731 degradation reaction Methods 0.000 title claims abstract description 33
- 238000003860 storage Methods 0.000 title claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000011156 evaluation Methods 0.000 claims abstract description 21
- 238000005457 optimization Methods 0.000 claims abstract description 17
- 230000006866 deterioration Effects 0.000 claims abstract description 8
- 238000013528 artificial neural network Methods 0.000 claims description 65
- 230000003862 health status Effects 0.000 claims description 54
- 238000004422 calculation algorithm Methods 0.000 claims description 23
- 230000009467 reduction Effects 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 4
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 238000004590 computer program Methods 0.000 claims 1
- 230000008569 process Effects 0.000 description 15
- 210000002569 neuron Anatomy 0.000 description 13
- 238000004458 analytical method Methods 0.000 description 11
- 238000004364 calculation method Methods 0.000 description 9
- 238000012545 processing Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 238000009826 distribution Methods 0.000 description 7
- 238000000513 principal component analysis Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000012544 monitoring process Methods 0.000 description 5
- 230000003595 spectral effect Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000012423 maintenance Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000002860 competitive effect Effects 0.000 description 3
- 230000005484 gravity Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 230000035772 mutation Effects 0.000 description 2
- 238000005057 refrigeration Methods 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004378 air conditioning Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 239000000049 pigment Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 239000003507 refrigerant Substances 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a method and a device for identifying deterioration of a health state of a compressor and a storage medium. The method comprises the following steps: extracting various characteristic parameters of a plurality of domains of the working state of the compressor according to the full life cycle vibration data of the compressor; performing feature optimization on various feature parameters based on a preset evaluation index to obtain an optimized feature parameter set; based on a preset fusion clustering mode, carrying out health state division on the optimized characteristic parameter set to obtain a health state characteristic data set; constructing a training health state identification model based on the health state characteristic data set; and carrying out degradation identification on the health state of the compressor by using the trained health state identification model. The invention can greatly improve the accuracy of the identification of the health state of the compressor.
Description
Technical Field
The present invention relates to the field of compressor technologies, and in particular, to a method and apparatus for identifying degradation of a health state of a compressor, and a storage medium.
Background
Compressors are a common industrial device for compressing a gas or liquid into a high density state. It increases the density of the gas or liquid by decreasing the volume or increasing the pressure, thereby achieving the purpose of storage, transportation or use. The compressor is widely applied to various industries including refrigeration and air conditioning, petrochemical industry, energy production, manufacturing industry and the like.
In industrial production, compressors play a critical role, such as providing compressed air for power transmission, driving pneumatic tools and equipment, or for refrigerant compression in refrigeration cycles. The performance and reliability of the compressor are critical to production efficiency and product quality, and therefore also to maintenance and monitoring of the compressor. When the compressor is in a sub-health state and is in a sub-health state, the change of the health state of the compressor is known earlier, so that the compressor can be maintained in time, the influence on production can be reduced to the minimum extent, and the health state identification of the compressor plays a vital role.
However, the existing compressor health state identification mode has the problems of poor universality, high operation difficulty, large operation amount, low identification accuracy, high cost and the like. Therefore, how to accurately and efficiently identify the health status of the compressor is a technical problem to be solved.
Disclosure of Invention
The invention aims to solve the technical problems existing in the prior art and provides a method and a device for identifying the deterioration of the health state of a compressor and a storage medium.
In order to solve the above technical problems, an embodiment of the present invention provides a method for identifying degradation of health status of a compressor, including:
extracting various characteristic parameters of a plurality of domains of the working state of the compressor according to the full life cycle vibration data of the compressor;
performing feature optimization on the multiple feature parameters based on a preset evaluation index to obtain an optimized feature parameter set;
dividing the health state of the preferred characteristic parameter set based on a preset fusion clustering mode to obtain a health state characteristic data set;
Constructing a training SOM-BP series neural network based on the health state characteristic data set to obtain a health state identification model;
and carrying out degradation identification on the health state of the compressor by using the trained health state identification model.
In order to solve the above technical problem, an embodiment of the present invention further provides a device for identifying degradation of health status of a compressor, including:
The parameter extraction module is used for extracting various characteristic parameters of a plurality of domains of the working state of the compressor according to the full life cycle vibration data of the compressor;
The feature optimization module is used for performing feature optimization on the various feature parameters based on a preset evaluation index to obtain an optimized feature parameter set;
The state dividing module is used for dividing the health state of the preferred characteristic parameter set based on a preset fusion clustering mode to obtain a health state characteristic data set;
The model training module is used for constructing a training SOM-BP series neural network based on the health state characteristic data set to obtain a health state identification model;
and the state recognition module is used for recognizing the deterioration of the health state of the compressor by using the trained health state recognition model.
In order to solve the above technical problem, an embodiment of the present invention further provides a device for identifying degradation of health status of a compressor, including: the compressor health state degradation identification method provided by the technical scheme is realized when the processor executes the program.
To solve the above technical problem, an embodiment of the present invention further provides a computer readable storage medium, including instructions, which when executed on a computer, cause the computer to execute the method for identifying degradation of health status of a compressor according to the above technical solution.
The beneficial effects of the application are as follows: in order to comprehensively describe the health state of the compressor, based on analysis of the structure and the health state process of the compressor, the application comprehensively considers a plurality of domain characteristic parameters of the vibration signal, extracts the characteristic parameters reflecting the health state, and optimizes the characteristic parameters so as to more accurately evaluate the health state degree of the compressor; fusion clustering is carried out on various characteristic parameters, so that heterogeneous characteristic fusion and health state division are realized; the application is based on SOM-BP series neural network as health status recognition model, the average recognition rate can reach 98.1%, thus greatly improving the accuracy of compressor health status recognition.
Additional aspects of the invention and advantages thereof will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is a flowchart of a method for identifying deterioration of health status of a compressor according to an embodiment of the present invention;
FIG. 2 is a diagram of a compressor health process vibration signal frequency domain power spectral entropy signature;
FIG. 3 is a diagram of a dimension reduction fusion result of PCA of the health status characteristics of the compressor according to the embodiment of the invention;
FIG. 4 is a contour diagram of health status division results based on GG clusters according to an embodiment of the present invention;
FIG. 5 is a contour plot of health status partitioning results based on FCM clustering;
FIG. 6 is a contour plot of health status partitioning results based on GK clustering;
FIG. 7 is a block diagram of a BP neural network;
FIG. 8 is a flow chart of a compressor health status recognition model based on BP neural network;
FIG. 9 is a block diagram of an SOM neural network;
FIG. 10 is a flow chart for identifying the health status of a compressor based on an SOM neural network;
FIG. 11 is a schematic diagram of a SOM-BP series neural network according to an embodiment of the present invention;
FIG. 12 is a flowchart for identifying the health status of a compressor based on a SOM-BP neural network according to an embodiment of the present invention;
FIG. 13 is a graph of compressor health status recognition results for three neural networks;
Fig. 14 is a block diagram of a compressor health status degradation recognition device according to an embodiment of the present invention.
Detailed Description
Other advantages and effects of the present disclosure will become readily apparent to those skilled in the art from the following disclosure, which describes embodiments of the present disclosure by way of specific examples. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present disclosure. The disclosure may be embodied or practiced in other different specific embodiments, and details within the subject specification may be modified or changed from various points of view and applications without departing from the spirit of the disclosure. It should be noted that the following embodiments and features in the embodiments may be combined with each other without conflict. All other embodiments, which can be made by one of ordinary skill in the art without inventive effort, based on the embodiments in this disclosure are intended to be within the scope of this disclosure.
It is noted that various aspects of the embodiments are described below within the scope of the following claims. It should be apparent that the aspects described herein may be embodied in a wide variety of forms and that any specific structure and/or function described herein is merely illustrative. Based on the present disclosure, one skilled in the art will appreciate that one aspect described herein may be implemented independently of any other aspect, and that two or more of these aspects may be combined in various ways. For example, an apparatus may be implemented and/or a method practiced using any number of the aspects set forth herein. In addition, such apparatus may be implemented and/or such methods practiced using other structure and/or functionality in addition to one or more of the aspects set forth herein.
The compressor health status identification means that the health status or faults of the compressor are accurately identified through monitoring and analyzing the running status of the compressor, and corresponding maintenance measures are timely taken to ensure the normal running of the compressor and prolong the service life.
The existing compressor health status recognition scheme has the following problems:
1. Diversity and complexity: compressors of different types and scales have different operating principles and structures, and thus health status characteristics are also different. Formulating a universal health status recognition method faces challenges and requires custom processing for different compressors.
2. Data interpretation and analysis: the amount of data acquired from the sensors is enormous, and how to accurately interpret and analyze such data is a difficult problem. Advanced algorithms and models need to be developed to distinguish between health status and degradation status signals and to accurately interpret the data.
3. Signal interference and noise: the compressor operating environment is complex and there are disturbances in vibration, temperature and pressure and noise. These disturbances may mask the true signal of health status, increasing the difficulty of recognition.
4. Real-time monitoring and early warning: real-time monitoring of health status and timely early warning is critical, but this requires high performance sensors, data acquisition systems and fast algorithm processing capabilities. At the same time, monitoring coverage and real-time assurance is also a challenge for large-scale industrial compressor systems.
5. Maintenance cost and feasibility: deployment and maintenance of health status identification techniques requires significant investment in capital and human resources. For some small and medium-sized enterprises or resource-limited scenarios, there may be cost and feasibility issues to implement these techniques.
That is, the existing compressor health status recognition method has the problems of poor universality, high operation difficulty, large operation amount, low recognition accuracy, high cost and the like.
In order to solve the above problems, embodiments of the present application provide a method, apparatus and computer readable storage medium for identifying degradation of health status of a compressor. These embodiments will be described in detail below.
Referring first to fig. 1, fig. 1 is a flow chart illustrating compressor health degradation identification according to an exemplary embodiment of the present application. The method may be specifically executed by a server, where the server may be an independent server, or may be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms, which are not limited herein.
As shown in fig. 1, in an exemplary embodiment, the method for identifying degradation of health status of a compressor may include steps S101 to S107, which are described in detail as follows:
S101, extracting various characteristic parameters of a plurality of domains of the working state of the compressor according to the full life cycle vibration data of the compressor.
The embodiment comprehensively considers various characteristic parameters of a plurality of domains of the working state of the compressor based on analysis of the structure and the health state process of the compressor so as to more accurately evaluate the health state degree of the compressor.
And S102, carrying out feature optimization on the various feature parameters based on a preset evaluation index to obtain an optimized feature parameter set.
According to the embodiment, the characteristic optimization is carried out by presetting various characteristic parameters of the evaluation index, the preferable characteristic parameter set is obtained, and the quality and applicability of the health state characteristic parameters can be comprehensively evaluated.
And S103, dividing the health state of the preferred characteristic parameter set based on a preset fusion clustering mode to obtain a health state characteristic data set.
In the embodiment, heterogeneous feature fusion and health state division are realized by fusion and clustering of various feature parameters.
S104, constructing a training SOM-BP series neural network based on the health state characteristic data set to obtain a health state identification model.
According to the embodiment, the SOM-BP series neural network is used as the health state identification model, so that the health state identification model has higher accuracy and higher convergence rate, and the high-speed and accurate identification of the model on the health state is achieved.
S105, performing compressor health state degradation identification by using the trained health state identification model.
According to the embodiment of the application, the characteristic parameters of a plurality of domains of the vibration signal are comprehensively considered, the characteristic parameters reflecting the health state are extracted, and the characteristic parameters are optimized, so that the health state degree of the compressor is estimated more accurately; fusion clustering is carried out on various characteristic parameters, so that heterogeneous characteristic fusion and health state division are realized; the application is based on SOM-BP series neural network as health status recognition model, the average recognition rate can reach 98.1%, thus greatly improving the accuracy of compressor health status recognition.
Optionally, in some embodiments, the plurality of characteristic parameters includes statistical features and entropy features of at least two of a time domain, a frequency domain, and a time-frequency domain. Specifically, the time domain statistical features in the various feature parameters comprise dimensional features and dimensionless features, and the time domain entropy features in the various feature parameters comprise information entropy and permutation entropy.
Time domain analysis is a process of studying the variation over time of the digital and distribution characteristics of the vibration signal. The extraction of signal features by time domain analysis is a fundamental part of signal processing, and is widely used in state identification research. The method extracts time domain features including time domain statistical features and time domain entropy features through time domain analysis.
The time domain statistical features are generally classified into two types, namely a dimensionless feature and a dimensionless feature, and the expression of each statistical feature is shown in table 1, assuming that the discrete vibration signal is { x i } (i=1, 2..the number of samples is N, N).
Table 1 time domain statistics calculation formula table
The time domain statistical characteristics selected by the embodiment of the invention comprise peak value, peak-peak value, average amplitude value, root mean square value, square root amplitude value, variance, standard deviation, peak value factor, pulse factor, margin factor, waveform factor, kurtosis factor and skewness.
The time domain analysis has the advantage of simplicity and rapidness, but the time domain features can only reflect the overall change trend of the running state of the compressor. Some of these features are sensitive to early health conditions, but they are often not stable enough and are susceptible to noise interference. Therefore, the application of only the time domain feature cannot obtain an accurate operation state recognition result. In order to realize accurate identification, the embodiment of the invention combines the frequency domain characteristics, the time-frequency domain characteristics, the entropy and other characteristics to carry out comprehensive analysis. Such a comprehensive feature extraction method can more fully describe the operation state of the compressor.
In order to describe the health state rule of the compressor from more angles, the embodiment of the invention introduces parameter entropy capable of representing the inherent confusion of the system. Normal operation of the compressor is a regularized motion, which breaks down when the compressor deteriorates, resulting in disorder of the compressor vibration signal, and this disorder can be measured by entropy. Information entropy and permutation entropy are commonly used in time domain signals.
When the compressor fails, the internal components of the machine body are periodically bumped, so that a periodic pulse signal is generated. During the health state, these signals are relatively more ordered. Thus, the entropy value will decrease relatively during the health state. This suggests that the information entropy and permutation entropy can be used to describe the health process of the compressor. Compared with the information entropy, the permutation entropy is more obvious in the state mutation part and has better robustness.
Frequency domain analysis is a classical approach to study mechanical vibration signals. Since the signal may be represented as a combination of sinusoidal signals of different frequencies, the frequency structure of the vibration signal may change when the compressor fails. To characterize this variation and analyze the health of the compressor, frequency domain statistics and entropy features may be extracted. These features may reveal variations in the spectrum. Meanwhile, through spectrum analysis, the characteristic frequency and the frequency multiplication of the fault part of the compressor can be intuitively found, so that whether the compressor has faults or not is determined.
There are also some frequency domain statistical characteristic parameters in the frequency domain analysis relative to the time domain statistical characteristic parameters in the time domain analysis. The general calculation formula of the frequency domain statistical characteristics is shown in the following table:
Table 2 frequency domain statistics calculation formula table
The frequency domain statistical characteristics selected by the embodiment of the invention comprise mean amplitude frequency, gravity center frequency, variance frequency, standard deviation pigment rate, root mean square frequency, skewness frequency and kurtosis frequency.
Fourier transforming each sample data to obtain a discrete representation of the spectral signal. Then, the frequency domain statistics of each sample are calculated according to the formula in table 2. And obtaining a frequency domain statistical feature vector describing the whole health state process of the compressor by arranging each obtained feature parameter value according to the sample sequence.
The mean amplitude frequency, the standard deviation frequency and the variance frequency reflect the distribution condition of the vibration signal frequency spectrum in the health state process. When the operating state of the compressor changes, the amplitude of the spectral line changes with frequency. As the compressor health increases, the amplitude of the high frequency portion increases. These three eigenvalues increase significantly in the middle and late stages, consistent with the health status. However, these parameters do not change significantly in the early stages of health.
The center of gravity frequency and the root mean square frequency both have relative changes in different health phases, which reflect the health course of the compressor to some extent, but there are large fluctuations in the overall.
In contrast, the skewness frequency and kurtosis frequency only change during the severe phases of the compressor health, with little change during the longer time periods before. Therefore, they are insufficiently characterized for a healthy state process.
The complexity of the structural distribution of the vibration signal frequency is usually characterized by power spectral entropy in the frequency domain, and the power spectral entropy of the sample data is extracted, as shown in fig. 2 of the drawings.
According to the embodiment of the invention, on the basis of analyzing the basic structure and the health state process of the compressor, aiming at the problem that the health state of the compressor cannot be comprehensively reflected by a single characteristic, according to the full life cycle vibration data of the compressor, various statistical characteristics and entropy characteristics in the time domain, the frequency domain and the time-frequency domain of the working state of the compressor are extracted, and the characterization condition of each characteristic parameter on the health state of the compressor is analyzed.
Optionally, in some embodiments, the performing feature optimization on the multiple feature parameters based on a preset evaluation index to obtain a preferred feature parameter set includes:
S201, respectively calculating correlation indexes, robustness indexes and information utility indexes of various characteristic parameters.
The operation state of the compressor is a continuous process which is subject to environmental influences, so that it is required that the health state characteristic parameter must be able to maintain a certain correlation with the time series of the operation process, and at the same time, maintain a certain robustness to noise interference signals. The smaller the information entropy of a feature is, the greater the degree of variation is, and the larger the amount of information is, based on the concept of information entropy, and therefore the feature information entropy is considered as an information utility index of the health state feature. Therefore, the embodiment of the invention evaluates the health state characteristics by using three indexes of relevance, robustness and information utility.
Because the health state of the compressor is a random process when the compressor is in operation, in order to better study the correlation and the robustness, the embodiment of the invention carries out the moving average processing on each obtained characteristic sequence, and the deterministic component and the random component of each sequence are separated, namely:
F(ti)＝FT(ti)+FR(ti) (1)
Wherein:
T i -the ith time point in time series T;
F (t i) -a signature sequence at time t i;
F T(ti) -deterministic component of the signature sequence at time t i;
F R(ti) -the randomness component of the signature sequence at time t i.
The specific calculation principle of the three indexes is as follows:
(1) Correlation index
The correlation index of the characteristic sequence F and the time sequence T is expressed by Corr (F, T), and is obtained by calculating the Pearson correlation coefficient of the deterministic component of the characteristic sequence and the time sequence:
Wherein:
-means of the time series;
-mean value of stationary trend term.
From the formula (2), corr (F, T) is shown to be within [0,1], and the larger the correlation coefficient is, the stronger the correlation between the feature and the time sequence is shown.
(2) Robustness index
The robustness of the feature sequence F is denoted Rob (F), defined as follows:
The robustness index takes a value within 0 and 1, and is mainly used for representing the tolerance capability of the degradation characteristic to abnormal values or noise interference, and the larger the robustness value is, the better the stability of the characteristic is.
(3) Information utility index
For the information utility index, the information entropy of each feature quantity is calculated, and then the information entropy is inverted to obtain the information utility value of the feature quantity, and the information utility value is calculated as follows:
1) The feature value is subjected to maximum and minimum normalization processing, for example, the j-th class feature at the time t i is f j(ti, and the normalization processing is as follows:
Wherein:
Max (f j)——fj(ti);
min (f j)——fj(ti).
2) Calculating information entropy of various features, and calculating information entropy of the j-th feature according to definition of the information entropy as follows:
Wherein:
H j, namely, various characteristic quantity information entropy, wherein H j is more than or equal to 0 and less than or equal to 1;
m-feature class number;
The specific gravity of the j-th class feature at the time of P ij——ti is calculated as follows:
3) Calculating information utility values of various features:
Ij＝1-Hj (7)
The value of the characteristic information utility index is also within 0 and 1, and from the concept of information entropy, the larger the information utility value of a certain type of degradation characteristic is, the larger the variation degree of the characteristic is, the more information is carried, and the larger the effect in classification is.
S202, carrying out weighted linear combination on the basis of preset weights of three performance indexes and each performance index value of the characteristic parameters to obtain a comprehensive evaluation index value.
Each performance evaluation index is used for evaluating various degradation characteristics from a single angle, the performance evaluation index has one-sided performance, and in order to integrate three indexes to achieve the purpose of comprehensively evaluating the degradation characteristics, the embodiment of the invention constructs a weighted linear combination of the three indexes, wherein the three indexes are positively correlated with the feature performance. The expression is as follows:
C＝ω1Corr(F,T)+ω2I(F)+ω3Rob(F) (8)
Wherein:
C, comprehensive evaluation index;
corr (F, T) -an index of correlation of the characteristic sequence F with the time sequence T;
I (F) -information utility values of the feature sequences F;
Rob (F) -robustness of the feature sequence F;
omega-importance weight of three indexes, the determination of the weight needs to satisfy two conditions, as follows:
For the value of omega i, the selection is needed according to specific requirements, and after the calculation of different characteristic index values of each characteristic quantity is completed, the embodiment of the invention normalizes each index to the [0,1] interval by a maximum and minimum value method, so that the order-of-magnitude difference of each index factor value in the weighting calculation process is prevented from being submerged.
And S203, performing feature optimization based on the sequence of the comprehensive evaluation index values of the feature parameters to obtain an optimized feature parameter set.
At present, most of researches on compressors directly apply a plurality of features extracted from a single domain or multiple domains to perform fault diagnosis or residual life prediction, but the suitability of each feature parameter to research contents is rarely considered. The quantity of the extracted characteristic parameters is huge, some characteristics possibly sensitive to health state information of a certain stage can effectively describe the health state process of the compressor, and the compressor is fully utilized. However, there are also features that are not sensitive enough to the health of the compressor, do not represent the health well, and use of these parameters can reduce the computational efficiency of the health identification, even leading to inaccurate results.
Therefore, before model training is carried out, the embodiment of the invention adopts a proper feature selection method to screen out the feature parameters which can effectively represent the health state, and eliminates redundant feature quantities, so as to ensure the accuracy of the state identification model, reduce the operation time of the algorithm and improve the operation efficiency. Therefore, the embodiment of the invention provides a feature optimization method based on three performance evaluation indexes (correlation index, robustness index and information utility index), and feature selection is realized by evaluating the feature performance of the health state of the compressor.
The embodiment of the invention provides a comprehensive evaluation criterion, which integrates three performance evaluation indexes and is used for evaluating the performance of health state characteristics. According to the criterion, useless characteristic parameters are eliminated, and a characteristic vector set capable of comprehensively and accurately representing the health state of the compressor is constructed.
Optionally, in some embodiments, the classifying the health state of the preferred feature parameter set based on a preset fusion clustering manner to obtain a health state feature data set includes:
S301, mapping the various characteristic parameters to a preset dimension based on a linear dimension reduction fusion algorithm, wherein the mapped variables in each dimension are linear combinations of original variables.
Specifically, the linear dimension reduction fusion algorithm in this embodiment may employ a PCA feature fusion algorithm. PCA is a common linear dimension-reduction fusion algorithm, and the main idea is that n-dimensional characteristic variables are mapped to k dimensions, and each dimension of the mapped variables is a linear combination of original variables; wherein n and k are positive integers, and n is greater than k. In this way, the purpose of heterogeneous feature dimension reduction fusion is achieved, and in the accompanying drawings, a PCA dimension reduction fusion result of the health state features of the compressor is represented in FIG. 3.
S302, fuzzy clustering is carried out on the multidimensional characteristic parameters after linear dimension reduction fusion based on a fuzzy clustering algorithm, and health state division is carried out on the multidimensional characteristic parameters based on the distance measure of fuzzy maximum likelihood estimation, so that a health state characteristic data set is obtained.
In particular, the fuzzy clustering algorithm in this embodiment may employ a GG fuzzy clustering algorithm. Fuzzy clustering is one of typical clustering methods, and common Fuzzy clustering methods include Fuzzy C-means (Fuzzy CENTER MEANS, FCM) clustering, GK (Gustafaon Kessel, GK) clustering, GG clustering, and the like. The FCM and GK clustering method has certain application limitation, and is suitable for clustering research of data distribution in a sphere-like shape. The GG clustering introduces a distance measure based on fuzzy maximum likelihood estimation, so that the applicability of the algorithm is higher, and the main trend and the state mutation point of the irregularly distributed data can be better identified.
Fig. 4, fig. 5 and fig. 6 of the drawings represent the health status classification result based on GG cluster, the health status classification result based on FCM cluster and the health status classification result based on GK cluster, respectively. By comparison, the condition characteristic dataset and GG clustering algorithm after screening are adopted to divide the health condition of the compressor, so that the accuracy of the result is ensured.
According to the embodiment of the invention, the first Principal Component Analysis (PCA) based dimension reduction fusion is utilized to carry out the comparative analysis of a plurality of clustering methods. The quantitative analysis of the classification coefficient proves the superiority of the Gath Geva (GG) clustering method in performance. In addition, the first principal component of the unfiltered health status feature set is clustered by using the GG algorithm, so that the necessity and feasibility of health status feature optimization by using the comprehensive performance evaluation index are further proved. Finally, the health of the compressor is partitioned according to the GG algorithm and the preferred feature set.
Optionally, in some embodiments, the SOM-BP series neural network is a contention layer added before an underlying layer of the BP neural network structure, and an output of the SOM neural network is taken as an input of the BP neural network.
BP neural network is proposed by D.E. Rumelhart et al, B is a common feedforward artificial neural network, is trained by a back propagation algorithm, has a plurality of hidden layers and output layers, can be used for tasks such as pattern recognition, function approximation and the like, and has stronger nonlinear fitting capability.
A typical structure of a BP neural network is shown in fig. 7 of the drawings, comprising an input layer, an hidden layer and an output layer. The compressor health status recognition model operation flow based on the BP neural network is shown in figure 8 of the accompanying drawings. The specific implementation flow is as follows:
1. Performing basic network setting, and determining the number M of input layer nodes, the number q of hidden layer nodes, the number L of output layer nodes, a transfer function g (), the maximum iteration number T, the learning efficiency eta and the error tolerance epsilon;
2. Initializing a network;
3. Inputting degradation characteristic sample data to be identified and expected output sample data into a network, normalizing, and performing forward propagation calculation to actually output;
4. Calculating total error E S between actual output and expected output of the S samples after the calculation of the BP neural network;
5. And (3) judging operation stop: if E S is less than or equal to epsilon or t=T (T is the iteration number), stopping the network training, otherwise, continuing to run the algorithm;
6. error back propagation calculation: the updated weight is carried into the step 3 for continuous operation by updating the network weight;
7. Repeating the steps 3-6 until the error tolerance or iteration number requirement is met, and stopping training;
8. model test: and inputting the degradation state test sample into the trained model, and checking a state division result.
The specific results of the BP neural network for identifying each health state of the compressor are shown in table 3.
TABLE 3 compressor health status recognition specific results based on BP neural network
As can be seen from table 3, in the compressor health status recognition, three erroneous determinations are made on the BP neural network, and specific erroneous determination results are shown in the above table, wherein only the recognition rate of the severe health status sample reaches 100%, and the other status samples are all erroneous determinations, so that the average recognition rate of the BP neural network in the compressor health status recognition is 92.5%. The BP neural network has the defects of sample dependence, easiness in being trapped in local optimum, low convergence speed and the like, and the problems influence the recognition efficiency and accuracy and further need to improve the effect of the recognition of the health state of the compressor.
SOM neural networks, also known as self-organizing feature mapping networks, are a model of neural networks that do not have supervised learning. The method maps the input data into a two-dimensional or multi-dimensional topological structure through a self-organizing process, realizes the clustering and visualization of the data, and can be used for tasks such as data dimension reduction, cluster analysis and the like. A typical structure thereof is shown in fig. 9 of the accompanying drawings.
The training method of the SOM neural network is different from that of the BP neural network, and adopts a competition learning strategy. The position of the neuron is adjusted by calculating the Euclidean distance between the input sample and the neuron of the competitive layer, determining the most similar neuron as the winning neuron, and then updating the weight vector of the winning neuron and the neurons in the neighborhood range of the winning neuron. After multiple iterations, the SOM neural network can automatically identify the distribution rule of the sample data and realize unsupervised clustering. Compared with other methods, the SOM neural network can intuitively reflect the distribution condition of the input samples. The SOM neural network-based compressor health status recognition flow is shown in fig. 10 of the accompanying drawings. The specific flow is as follows:
1. the network basic settings: determining a network topology structure and a maximum iteration number T according to actual conditions, wherein the network topology structure comprises the number of input layer nodes and the number and distribution of competing layer nodes;
2. network initialization: randomly distributing initial connection weights between the network input layer neurons and the competitive layer neurons;
3. determining a winning neuron and its neighborhood: inputting a normalized degradation state training sample, calculating the distance between each degradation characteristic vector and each neuron of the competitive layer, and taking the neuron with the smallest distance as a winning neuron;
4. updating the network weight;
5. And (3) judging operation stop: judging whether the training times T reach the maximum iteration times T, if so, ending the training, otherwise, returning to the step 3 to continue the operation until the ending condition is met;
6. model test: and inputting the degradation state test sample into the trained model, and checking a state division result.
Specific results of the SOM neural network for identifying each health state of the compressor are shown in table 4.
TABLE 4 SOM neural network based compressor health status recognition specific results
9 Groups of samples in the moderate health state samples are misjudged to be in the heavy state, so that the average recognition rate of the model is only 82.7%. Although the SOM neural network can realize classification of the health state of the compressor, the classification effect of different states similar to sample data is poor, only coarse classification of data samples is realized, and the output result of the SOM neural network is not suitable to be directly used for judging the health state of the compressor.
The method takes the output of the SOM neural network as the input of the BP neural network as the SOM-BP series neural network. The network model integrates the advantages of the SOM neural network and the BP neural network, and the main construction thought is to add a competition layer before an implicit layer of the BP neural network structure, so as to construct a two-stage series neural network, as shown in figure 11 of the drawings.
A flow chart for identifying the health state of the compressor based on the SOM-BP neural network is shown in figure 12 of the accompanying drawings. The operation process is as follows:
1. Data preprocessing: for S health training sample data (x s,ys) (s=1, 2 … … S)
(X s is the input health status feature sample in M dimensions, y s is the expected output sample) to perform normalization processing;
2. Setting parameters of a series neural network: determining the node number of each layer of the network, the maximum iteration times of each level, transfer functions, learning efficiency, error tolerance and the like;
3. Sample primary classification: processing an input M-dimensional compressor health state characteristic sample x s through a first-stage SOM neural network to obtain a preliminary classification result r s; t1 represents the number of iterations of the first stage, and T1 represents the maximum number of iterations of the first stage;
4. Constructing an implicit layer input sample: adding the preliminary classification result r s into x s to form an M+1-dimensional health state characteristic data sample x s', and constructing hidden layer input training sample data as (x s',ys);
5. Second level neural network training: taking (x s',ys) obtained in the step 4 as input for further training; t2 represents the number of iterations of stage 2, T2 represents the maximum number of iterations of stage 2;
6. And (3) judging operation stop: if the training error Es is smaller than or equal to the error margin epsilon or the iteration number T2 reaches the maximum iteration number T 2, stopping the network training, otherwise, turning to the step 5, and continuing the algorithm until convergence.
Specific results of the SOM-BP neural network for identifying each health state of the compressor are shown in Table 5.
TABLE 5 SOM neural network based compressor health status recognition specific results
The SOM-BP series neural network has excellent performance in the aspect of compressor health state identification, the average identification rate reaches 98.1%, wherein only the identification of the middle health state is in a group of errors, and the rest are accurately identified in hundred percent.
The compressor health status recognition results of the three neural networks under the same training and testing health status samples are shown in fig. 13 of the accompanying drawings. Compared with single BP and SOM neural network models, the SOM-BP series neural network has higher accuracy in the aspect of health state identification and faster convergence efficiency under the same data volume.
According to the embodiment of the invention, according to the full life cycle vibration data of the compressor, various statistical characteristics and entropy characteristics in the time domain, the frequency domain and the time-frequency domain of the working state of the compressor are extracted, the characterization condition of each characteristic parameter on the health state of the compressor is analyzed, then the health state characteristic screening and the health state division of the compressor based on PCA characteristic fusion and GG fuzzy clustering are carried out, and the characteristic screening and the health state division lay a foundation for the health state identification.
A health state identification model is built based on a Back Propagation (BP) neural network and a Self-organizing map (SOM) neural network respectively, the respective identification performance is verified by utilizing a health state characteristic data set, the advantages and disadvantages of the two models are integrated, a SOM-BP series neural network health state identification model is built, the identification effects of the three models are compared and analyzed under the same health state characteristic data set, and the superiority of the series neural network state identification model is proved.
As shown in fig. 14, an embodiment of the present invention further provides a compressor health status degradation identifying device 200, including:
The parameter extraction module 201 is configured to extract various characteristic parameters of the working state of the compressor according to the full life cycle vibration data of the compressor; wherein the plurality of characteristic parameters includes parameters of at least two domains of a time domain, a frequency domain, and a time-frequency domain;
the feature optimization module 202 is configured to perform feature optimization on the multiple feature parameters based on a preset evaluation index, so as to obtain a preferred feature parameter set;
the state dividing module 203 is configured to divide the health state of the preferred feature parameter set based on a preset fusion clustering manner, so as to obtain a feature data set of the health state;
the model training module 204 is configured to construct a training SOM-BP serial neural network based on the health status feature data set, so as to obtain a health status recognition model;
The state recognition module 205 is configured to perform degradation recognition on the health state of the compressor by using the trained health state recognition model.
The embodiment of the invention also provides a device for identifying the deterioration of the health state of the compressor, which comprises the following steps: the compressor health state degradation identification method provided by the embodiment is characterized in that the processor executes the program.
The embodiment of the present invention also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the method for identifying deterioration of health status of a compressor as provided in the above embodiment.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the apparatus and units described above may refer to corresponding procedures in the foregoing method embodiments, which are not described herein again.
In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of elements is merely a logical functional division, and there may be additional divisions of actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not performed.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment of the present invention.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention is essentially or a part contributing to the prior art, or all or part of the technical solution may be embodied in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (10)
1. A method for identifying deterioration of a health state of a compressor, comprising:
extracting various characteristic parameters of a plurality of domains of the working state of the compressor according to the full life cycle vibration data of the compressor;
performing feature optimization on the multiple feature parameters based on a preset evaluation index to obtain an optimized feature parameter set;
dividing the health state of the preferred characteristic parameter set based on a preset fusion clustering mode to obtain a health state characteristic data set;
Constructing a training SOM-BP series neural network based on the health state characteristic data set to obtain a health state identification model;
and carrying out degradation identification on the health state of the compressor by using the trained health state identification model.
2. The method of claim 1, wherein the plurality of characteristic parameters comprises statistical features and entropy features of at least two of a time domain, a frequency domain, and a time-frequency domain.
3. The method of claim 2, wherein the time domain statistical features in the plurality of feature parameters include dimensional features and dimensionless features, and wherein the time domain entropy features in the plurality of feature parameters include information entropy and permutation entropy.
4. The method according to claim 1, wherein the feature optimization of the plurality of feature parameters based on a preset evaluation index to obtain a preferred feature parameter set includes:
respectively calculating the correlation index, the robustness index and the information utility index of various characteristic parameters;
Performing weighted linear combination on the basis of preset weights of the three performance indexes and each performance index value of the characteristic parameters to obtain a comprehensive evaluation index value;
and performing feature optimization based on the sequence of the comprehensive evaluation index values of the feature parameters to obtain an optimized feature parameter set.
5. The method according to claim 1, wherein the classifying the health state of the preferred feature parameter set based on the preset fusion clustering method to obtain a health state feature data set includes:
Mapping the various characteristic parameters to a preset dimension based on a linear dimension reduction fusion algorithm, wherein the variable in each dimension after mapping is the linear combination of the original variables;
and carrying out fuzzy clustering on the multidimensional characteristic parameters subjected to linear dimension reduction fusion based on a fuzzy clustering algorithm, and carrying out health state division on the multidimensional characteristic parameters based on the distance measure of fuzzy maximum likelihood estimation to obtain a health state characteristic data set.
6. The method of claim 5, wherein the linear dimension reduction fusion algorithm employs a PCA feature fusion algorithm and the fuzzy clustering algorithm employs a GG fuzzy clustering algorithm.
7. The method according to any one of claims 1 to 6, wherein the SOM-BP series neural network is a layer of contention added before an underlying layer of the BP neural network structure, and an output of the SOM neural network is taken as an input of the BP neural network.
8. A compressor health condition degradation identification device, comprising:
The parameter extraction module is used for extracting various characteristic parameters of a plurality of domains of the working state of the compressor according to the full life cycle vibration data of the compressor;
The feature optimization module is used for performing feature optimization on the various feature parameters based on a preset evaluation index to obtain an optimized feature parameter set;
The state dividing module is used for dividing the health state of the preferred characteristic parameter set based on a preset fusion clustering mode to obtain a health state characteristic data set;
The model training module is used for constructing a training SOM-BP series neural network based on the health state characteristic data set to obtain a health state identification model;
and the state recognition module is used for recognizing the deterioration of the health state of the compressor by using the trained health state recognition model.
9. A compressor health condition degradation identification device, comprising: memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the compressor health status degradation identification method according to any one of claims 1 to 7 when executing the program.
10. A computer readable storage medium comprising instructions which, when run on a computer, cause the computer to perform the compressor health status degradation identification method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410025382.7A CN118035815A (en) | 2024-01-08 | 2024-01-08 | Compressor health state degradation identification method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410025382.7A CN118035815A (en) | 2024-01-08 | 2024-01-08 | Compressor health state degradation identification method, device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118035815A true CN118035815A (en) | 2024-05-14 |
Family
ID=90995884
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410025382.7A Pending CN118035815A (en) | 2024-01-08 | 2024-01-08 | Compressor health state degradation identification method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118035815A (en) |
-
2024
- 2024-01-08 CN CN202410025382.7A patent/CN118035815A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111222549B (en) | Unmanned aerial vehicle fault prediction method based on deep neural network | |
| CN115577275A (en) | Time sequence data anomaly monitoring system and method based on LOF and isolated forest | |
| KR20180120056A (en) | Method and system for pre-processing machine learning data | |
| CN116861354A (en) | Abnormality identification and mode distinguishing method and system for monitoring data of power transformer | |
| CN112039903B (en) | Network security situation assessment method based on deep self-coding neural network model | |
| CN112232244A (en) | Fault diagnosis method for rolling bearing | |
| CN110708318A (en) | Network abnormal flow prediction method based on improved radial basis function neural network algorithm | |
| CN114861788A (en) | Load abnormity detection method and system based on DBSCAN clustering | |
| CN113568368B (en) | Self-adaptive determination method for industrial control data characteristic reordering algorithm | |
| CN110503133A (en) | A kind of centrifugal compressor failure prediction method based on deep learning | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| CN117421994A (en) | Edge application health monitoring method and system | |
| Gond et al. | A survey of machine learning-based approaches for missing value imputation | |
| CN115375026A (en) | Method for predicting service life of aircraft engine in multiple fault modes | |
| CN116108371A (en) | Cloud service abnormity diagnosis method and system based on cascade abnormity generation network | |
| CN115114484A (en) | Abnormal event detection method and device, computer equipment and storage medium | |
| CN117633688A (en) | Large-scale power data anomaly detection method based on ridge regression-k-means clustering-LOF-LSTM fusion algorithm | |
| Syaliman | Enhance the Accuracy of K-Nearest Neighbor (K-Nn) for Unbalanced Class Data Using Synthetic Minority Oversampling Technique (Smote) and Gain Ratio (Gr) | |
| KR20040012285A (en) | System And Method For Detecting Intrusion Using Hidden Markov Model | |
| CN116563690A (en) | Unmanned aerial vehicle sensor type unbalanced data anomaly detection method and detection system | |
| CN118035815A (en) | Compressor health state degradation identification method, device and storage medium | |
| CN116933119A (en) | Signal data trend removal method based on convolutional neural network | |
| CN112733903B (en) | SVM-RF-DT combination-based air quality monitoring and alarming method, system, device and medium | |
| Huang et al. | Outlier detection method based on improved two-step clustering algorithm and synthetic hypothesis testing | |
| Yan et al. | Fault Diagnosis of Rolling Bearing with Small Samples Based on Wavelet Packet Theory and Random Forest |
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 |