CN114282571B - Method, system, equipment and medium for constructing multidimensional health index of bearing - Google Patents
Method, system, equipment and medium for constructing multidimensional health index of bearing Download PDFInfo
- Publication number
- CN114282571B CN114282571B CN202111258215.XA CN202111258215A CN114282571B CN 114282571 B CN114282571 B CN 114282571B CN 202111258215 A CN202111258215 A CN 202111258215A CN 114282571 B CN114282571 B CN 114282571B
- Authority
- CN
- China
- Prior art keywords
- bearing
- moa
- dimensional
- health degree
- imf
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention relates to a method, a system, equipment and a medium for constructing a multi-dimensional health index of a bearing based on MOA-VMD, wherein the method comprises the following steps: firstly, acquiring a bearing vibration signal; secondly, optimizing the optimal parameters of VMD decomposition by an MOA algorithm, and decomposing the bearing vibration signals according to the optimal parameters to obtain k IMF components; then, screening n IMF components to reconstruct signals; then, obtaining a multidimensional feature vector according to the maximum value, the average value and the kurtosis of the extracted reconstruction signal and the singular values, the sample entropy and the energy entropy of the n IMF components; thirdly, reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimension reduction algorithm to obtain health degree features; and finally, respectively inputting the health degree characteristics into SVM and LSTM networks using MOA to optimize parameters for analysis. The scheme provided by the invention can extract the working condition information and the health degree information at the same time, and has good working condition identification and health degree prediction effects.
Description
Technical Field
The invention relates to the technical field of fault monitoring of mechanical equipment, in particular to a method, a system, equipment and a medium for constructing a multidimensional health index based on an MOA-VMD bearing.
Background
In recent years, wind power is fully developed and utilized, and the wind power is gradually becoming the third largest energy source after thermal power and hydroelectric power. The fan is as wind power generation's main part, often works in various abominable environments, and antifriction bearing is the core part of fan, and its accident damages often can make fan equipment seriously destroy, and even the person that can bring the casualties, cause very big loss. To avoid the above danger, it is necessary to predict the health of the bearing. Through the prediction to the health degree of bearing, can master the health degree level of bearing earlier, formulate the maintenance plan of fan equipment according to it rationally, can reduce down time and reduce maintenance cost, avoid the emergence of major incident. Therefore, the prediction of the health degree of the bearing has important significance for the operation and maintenance of equipment such as a fan and the like.
The current basic health prediction techniques can be basically divided into two types, physical model-based methods and data-driven methods. The physical model-based method generally requires a large amount of professional prior knowledge, and the structure, mechanism, operation environment and the like of the equipment are complicated, so that a model with strong applicability is difficult to establish. The method for predicting the health degree by processing the test data is generally called a data-driven method, is a mainstream method for predicting the health degree at present, and establishes a mapping relation model of the vibration signal and the health degree of the bearing by analyzing and processing the vibration signal data of the bearing, so as to predict the health degree of the bearing by the model. With the development of machine learning, the health degree prediction based on data driving becomes the key direction of domestic and foreign research.
However, most of the types of the feature information adopted in the existing scheme are single, and various types of features cannot be comprehensively considered; most scholars choose to construct a one-dimensional monotonic health index to describe the current health degree, so that the health degree information contained in the constructed health index is insufficient, most extracted features can only identify the health degree, and information such as working conditions and the like cannot be extracted at the same time.
Disclosure of Invention
Technical problem to be solved
In view of the above disadvantages and shortcomings of the prior art, the present invention provides a method, a system, a device and a medium for constructing a multidimensional health index of a bearing based on an MOA-VMD, which solves the technical problem that the information contained in the health index constructed for predicting the health degree of the bearing is insufficient.
(II) technical scheme
In order to achieve the purpose, the invention adopts the main technical scheme that:
in a first aspect, an embodiment of the present invention provides a method for constructing a multidimensional health index of a bearing based on an MOA-VMD, including:
acquiring a bearing vibration signal;
optimizing the optimal parameters of VMD decomposition through an MOA algorithm, and decomposing the bearing vibration signals according to the optimized optimal parameters to obtain k IMF components;
reconstructing the n IMF components screened according to the correlation magnitude to obtain a reconstructed signal;
obtaining a multidimensional feature vector according to the extracted maximum value, average value and kurtosis of the reconstructed signal and singular values, sample entropies and energy entropies of the n IMF components;
reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimension reduction algorithm to obtain health degree features;
respectively inputting the health degree characteristics into SVM and LSTM networks using MOA to optimize parameters to obtain bearing working condition information and health degree information;
the SVM and the LSTM network are trained in advance based on a training set under the same working condition.
Optionally, the optimizing the parameters of the VMD decomposition by the MOA algorithm comprises:
initializing parameters of the MOA;
according to a predetermined parameter [ k ] 0 ,α 0 ]Performing VMD decomposition on the bearing vibration signal to obtain k IMF initial components;
calculating the envelope entropy of each IMF initial component, and taking the local minimum envelope entropy as a fitness function;
based on the fitness function, continuously iterating and optimizing until a preset termination condition is met to find out the global minimum envelope entropy and the optimal parameter where the corresponding optimal IMF initial component is located;
the fitness function is:
min F＝min E p
wherein E is p Is the envelope entropy, j =1,2,3, N, a (j) is the envelope signal obtained by Hilbert demodulation; e.g. of a cylinder j Is obtained by normalizing a (j).
Optionally, performing VMD decomposition on the bearing vibration signal according to the optimal parameter to obtain k IMF original components includes:
solving an analytic signal of each IMF original component through Hilbert transformation to further obtain a single-side frequency spectrum;
modulating the single-side frequency to a corresponding base frequency band according to the pre-estimated mixed center frequency to obtain a demodulation signal;
calculating the square of the time gradient norm of the demodulation signal to obtain the estimated bandwidth of the IMF original component;
constructing a variation model with constraints according to the estimated bandwidth and the introduced constraint conditions;
a penalty parameter and a Lagrange multiplication coefficient are introduced to solve the variational model to obtain a frequency domain expression of an IMF original component, an updating formula of a center frequency and an updating formula of a penalty factor;
obtaining an updated frequency domain value, an updated central frequency and an updated Lagrange multiplication coefficient according to the frequency domain expression of the IMF original component, the updated formula of the central frequency and the updated formula of the Lagrange multiplication coefficient;
converting the k IMF original components into amplitude modulation-frequency modulation signals according to the updated frequency domain value, the updated center frequency and the updated Lagrange multiplication coefficient to obtain k IMF components;
wherein the IMF component expression is:
u k (t)＝A k (t)cos[φ k (t)]
in the formula u k (t) represents K IMF components, K ∈ { 1., K }, φ k (t) is a non-decreasing phase function; a. The k (t) represents an envelope function;
the variation model with constraint is as follows:
{u k represents the K IMF components resulting from VMD decomposition; { omega [ [ omega ] ] k Represents the center frequency corresponding to the IMF component; δ (t) is the dirichlet function; * Performing convolution operation; x represents the original signal.
Optionally, reconstructing the n IMF components screened according to the correlation size to obtain a reconstructed signal includes:
determining a correlation coefficient of each IMF component and the bearing vibration signal by using a Pearson correlation coefficient;
and based on the correlation coefficient, selecting the first n IMF components which are ordered from large to small according to the correlation number, and accumulating the first n IMF components to obtain a reconstructed signal.
Optionally, reducing the multidimensional feature vector to 3 dimensions by using a t-SNE dimension reduction algorithm, and obtaining the health feature includes:
and S51, calculating the joint probability density of the multi-dimensional feature vector.
S52, according to a preset low-dimensional vector initial solution, determining the similarity of any two data points in the low-dimensional vector initial solution.
And S53, measuring the similarity of the distribution of the multidimensional characteristic vectors and the distribution of the low-dimensional vectors by using KL divergence according to the joint probability density of the multidimensional characteristic vectors and the similarity of any two data points in the initial solution of the low-dimensional vectors.
And S54, calculating gradient according to the similarity of the multi-dimensional feature vector distribution and the low-dimensional vector distribution, and obtaining the current low-dimensional vector.
And S55, repeating the steps S52-S55 for multiple times until the preset iteration times T are met, and obtaining the low-dimensional health degree characteristic.
Alternatively, the first and second liquid crystal display panels may be,
the multi-dimensional feature vector is: x = { X 1 ,x 2 ,...,x n }，
Any two data points x in the multi-dimensional feature vector i And x j Similar conditional probabilities are:
in the formula, delta i Is represented by x i Is the gaussian distribution variance of the center point.
The preset low-dimensional vector initial solution is y 0 ＝{y 1 ,y 2 ,...,y n }；
The health degree of the low dimension is characterized by:
wherein eta is the learning rate, alpha (t) is the optimization parameter, t represents the t-th iteration, C represents the objective function using the similarity of high-and low-dimensional data distribution,
representing a gradient formula.
Optionally, the step of inputting the health degree characteristics into SVM and LSTM networks using MOA optimization parameters to obtain bearing condition information and health degree information includes:
dividing 3-dimensional health degree features under the same working condition into a training set and a testing set, inputting the health degree features as input, outputting health degree values as labels, inputting training set data into an LSTM network for training to obtain a health degree prediction model, and inputting testing set data into the health degree prediction model to obtain health degree information;
dividing the 3-dimensional health degree characteristics under different working conditions into a training set and a test set, wherein the health degree characteristics serve as input, working condition categories serve as label output, training set data are input into an SVM to be trained to obtain a working condition category judgment model, and test set data are input into the category judgment model to obtain working condition information.
In a second aspect, an embodiment of the present invention provides a system for constructing a multidimensional health indicator based on an MOA-VMD bearing, including:
the information acquisition module is used for acquiring a bearing vibration signal;
the parameter optimizing and decomposing module is used for optimizing the optimal parameter decomposed by the VMD through an MOA algorithm and decomposing the bearing vibration signal according to the optimized optimal parameter to obtain k IMF components;
the reconstruction signal module is used for reconstructing the n IMF components screened according to the correlation size to obtain a reconstruction signal;
the multi-dimensional feature vector acquisition module is used for obtaining a multi-dimensional feature vector according to the maximum value, the average value and the kurtosis of the reconstruction signal and the singular values, the sample entropies and the energy entropies of the n IMF components obtained through calculation;
the dimensionality reduction module is used for reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimensionality reduction algorithm to obtain health degree features;
and the information output module is used for respectively inputting the health degree characteristics into an SVM (support vector machine) and LSTM network using MOA (motion estimation algorithm) optimization parameters to obtain bearing working condition information and health degree information.
In a third aspect, an embodiment of the present invention provides a device for constructing a multidimensional health index based on an MOA-VMD bearing, including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the MOA-VMD bearing based multi-dimensional health indicator construction method as described above.
In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which computer-executable instructions are stored, and when executed by a processor, the steps of the method for constructing a multidimensional health index based on an MOA-VMD bearing as described above are implemented.
(III) advantageous effects
The invention has the beneficial effects that: the VMD decomposition algorithm optimized by the mayfly algorithm has good decomposition effect on non-stationary and nonlinear signals such as vibration signals of bearings, can better recover the fluctuation characteristics of the signals, and greatly improves the prediction precision. The features extracted by the invention comprehensively consider the features of all aspects, so that the features contain rich feature information. The multi-dimensional health index provided by the invention reserves more bearing related information, is presented as a continuous curve in 3-dimensional visualization, and can extract health degree information and working condition information from one health index.
Drawings
FIG. 1 is a schematic flow chart of a MOA-VMD bearing multi-dimensional health index construction method provided by the invention;
FIG. 2 is an optimization iteration curve of three algorithms of the MOA-VMD bearing multi-dimensional health index construction method provided by the invention;
FIG. 3 is a partial schematic flow chart of step S2 of the MOA-VMD-based bearing multi-dimensional health index construction method provided by the present invention;
FIG. 4 is a partial schematic flow chart of step S2 of the MOA-VMD-based bearing multi-dimensional health index construction method provided by the present invention;
FIG. 5 is a detailed flowchart of step S3 of the MOA-VMD-based bearing multi-dimensional health index construction method provided by the present invention;
FIG. 6 is a detailed flowchart of step S5 of the MOA-VMD-based bearing multi-dimensional health index construction method provided by the present invention;
FIG. 7 is a schematic data sampling diagram of a MOA-VMD bearing multi-dimensional health index construction method according to the present invention;
FIG. 8 is a VMD decomposition result diagram of sampling point data of a MOA-VMD bearing multi-dimensional health index construction method according to the present invention;
FIG. 9 is a schematic diagram of the correlation degree between each IMF component and an original signal of the MOA-VMD bearing multi-dimensional health index construction method provided by the present invention;
FIG. 10 is a time domain diagram before and after data denoising based on the MOA-VMD bearing multi-dimensional health index construction method provided by the invention;
FIG. 11 is a schematic view of a condition feature extraction visualization method based on the MOA-VMD bearing multi-dimensional health index construction method provided by the invention;
12-a, 12-b and 12-c are health degree feature extraction visualizations based on the MOA-VMD bearing multi-dimensional health index construction method provided by the invention;
FIG. 13 is a schematic diagram of SVM classification results based on the MOA-VMD bearing multi-dimensional health index construction method provided by the invention;
FIG. 14 is a schematic diagram of an LSTM health degree prediction result based on an MOA-VMD bearing multi-dimensional health index construction method provided by the present invention;
FIG. 15 is a schematic overall flow chart of a MOA-VMD-based bearing multi-dimensional health index construction method provided by the present invention.
Detailed Description
For a better understanding of the present invention, reference will now be made in detail to the present embodiments of the invention, which are illustrated in the accompanying drawings.
As shown in fig. 1, a method for constructing a multidimensional health index based on an MOA-VMD bearing according to an embodiment of the present invention includes: firstly, acquiring a bearing vibration signal; optimizing the optimal parameters of the VMD decomposition through an MOA algorithm, and decomposing the bearing vibration signals according to the optimized optimal parameters to obtain k IMF components; then, reconstructing the n IMF components screened according to the correlation size to obtain a reconstruction signal; then, obtaining a multidimensional feature vector according to the maximum value, the average value and the kurtosis of the extracted reconstruction signal and the singular values, the sample entropy and the energy entropy of the n IMF components; thirdly, reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimension reduction algorithm to obtain health degree features; and finally, respectively inputting the health degree characteristics into an SVM (support vector machine) and LSTM network using MOA (mean of squares) optimization parameters to obtain bearing working condition information and health degree information. The SVM and the LSTM network are trained in advance based on a training set under the same working condition.
The VMD decomposition algorithm optimized by the mayfly algorithm has a good decomposition effect on non-stationary and nonlinear signals such as vibration signals of bearings, can better recover the fluctuation characteristics of the signals, and greatly improves the prediction precision. The features extracted by the invention comprehensively consider the features of all aspects, so that the features contain rich feature information. The multi-dimensional health index provided by the invention reserves more bearing related information, is presented as a continuous curve in 3-dimensional visualization, and can extract health degree information and working condition information from one health index.
For a better understanding of the above-described technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
Specifically, the invention provides a bearing multi-dimensional health index construction method based on an MOA-VMD, which comprises the following steps:
s1, obtaining a bearing vibration signal.
S2, optimizing the optimal parameters of VMD decomposition through an MOA algorithm, and decomposing the bearing vibration signals according to the optimized optimal parameters to obtain k IMF components.
The variational modal decomposition is a completely non-recursive and self-adaptive signal processing method, and has the advantage of determining the number of modal decompositions. The decomposition method determines the frequency center and the bandwidth of each component by iteratively searching the optimal solution of the variation model in the process of acquiring the decomposition components, thereby being capable of adaptively realizing the frequency domain subdivision of the signal and the effective separation of each component.
The VMD decomposition effect is greatly influenced by two parameters of the decomposition number K and the penalty parameter alpha. If the value of K is too large, a false component is generated in the decomposition result; if the value of α is small, modal aliasing occurs. The size of the penalty parameter affects the bandwidth of each IMF component after decomposition. The other parameters, which have less influence on the decomposition effect, are set to empirical values, i.e., tau =0, init =1, dc =0, and e =1e-7. Therefore, before VMD decomposition is performed on the signal, an appropriate decomposition number K and a penalty parameter α need to be selected.
For optimization of VMD parameters, two classical algorithms are common: genetic Algorithm (GA) and Particle Swarm Optimization (PSO) algorithms. The genetic algorithm simulates the principle of 'win-win elimination and survival of the fittest' in the nature, has strong global searching capacity, but is easy to fall into local optimum. The particle swarm algorithm simulates the predation behavior of a bird swarm, has a high convergence rate, is easy to fall into local optimum, and has a complex search path.
MOA is composed of a collection of mayflies and a collection of male mayflies, inspired by the mating behavior of mayflies, mating the optimal individual of male individuals with the optimal individual of female mayflies to obtain an optimal offspring. And similarly, the suboptimal male individual and the optimal female individual are mated to obtain the suboptimal individual. The process is the same as the survival rule of the fittest, and the individuals with poor fitness are gradually eliminated.
Initially, two groups of mayflies were randomly generated, representing male and female populations respectively. That is, each mayfly is randomly placed in the problem space as a candidate solution x = (x 1,.., x d) represented by a d-dimensional vector, and its performance is evaluated according to a predetermined objective function f (x). The speed v = (v 1., vd) of mayflies is defined as the change in its position, the direction of flight of each mayfly is a dynamic interaction of the individual and social flight experiences. Each mayfly will have its trajectory adjusted towards the best position for the individual (pbest) so far, and the best position for any mayfly in a colony so far (gbest). Mayflies and males have different movement characteristics: the aggregation of clusters of male mayflies means that the position of each male mayflies is adjusted according to experience of oneself and the surrounding mayflies, while female mayflies do not aggregate in clusters and they fly to the male for propagation. Mayflies are then represented by a crossover operator, with the parents being selected based on the fitness function of the individual, with best female and best male breeding, and next best female and next best male breeding. The result of the crossover is the generation of two offspring. The optimization procedure is shown in table 1.
TABLE 1
In order to verify the convergence and optimization performance of the MOA, the PSO algorithm, the GA and the MOA are tested for comparison. The performance of the above-mentioned materials is compared as a fitness function by the following functions.
For the PSO algorithm, the maximum number of iterations is set to 100 and the population size is 100. For the genetic algorithm, the maximum number of iterations is set to 100, the population size is 100, the crossover probability is 1, and the mutation probability is 0.01. For MOA, the maximum number of iterations is set to 100 and the population number is set to 50. The optimization iteration curves of the three algorithms are shown in fig. 2, and it can be seen that the optimization iteration speed of the MOA is superior to that of the other two algorithms, the other two algorithms are trapped in a local optimal solution, and the optimal value of the MOA is closer to the theoretical optimal value. The result shows that compared with the PSO algorithm and the genetic algorithm, the MOA has higher convergence rate and stronger global search capability.
As shown in fig. 3, in step S2, the optimizing the parameters of the VMD decomposition by the MOA algorithm includes:
and S21, initializing parameters of the MOA.
S22, according to a preset parameter [ k ] 0 ,α 0 ]And performing VMD decomposition on the bearing vibration signal to obtain k IMF initial components.
And S23, calculating the envelope entropy of each IMF initial component, and taking the local minimum envelope entropy as a fitness function.
And S24, based on the fitness function, continuously iterating and optimizing until a preset termination condition is met to find the global minimum envelope entropy and the optimal parameter where the corresponding optimal IMF initial component is located.
The fitness function is:
min F＝min E p
wherein E is p Is the envelope entropy, j =1,2,3, N, a (j) is the envelope signal resulting from Hilbert demodulation; e.g. of a cylinder j Is obtained by normalizing a (j).
In the above steps, when using the MOA to optimize the VMD parameters, a fitness function needs to be selected. After the bearing vibration signal is subjected to VMD decomposition, the envelope entropy value E of IMF component P Reflecting the sparseness of the component, if the IMF component contains more noise and can cover the impact characteristics of the signal, the sparseness of the IMF component is weaker, and the enveloping entropy is larger; otherwise, the envelope entropy is smaller. Selecting the smallest one of the envelope entropy values in the K IMF components obtained by decomposition as the local minimum envelope entropy minE P The component corresponding to the minimum entropy value has rich characteristic information. Therefore, the optimization process takes the local minimum envelope entropy as the fitness function of parameter optimization, and the whole search process is to find the global minimum envelope entropy and the optimal parameter [ K, α ] where the corresponding optimal component is located]。
As shown in fig. 4, in step S2, performing VMD decomposition on the bearing vibration signal according to the optimal parameters to obtain k IMF original components includes:
and S25, VMD decomposition is carried out on the bearing vibration signal according to the optimal parameters to obtain k IMF original components.
S26, solving analysis signals of the IMF original components through Hilbert transformation, and further obtaining a single-side frequency spectrum.
And S27, modulating the single-side frequency to the corresponding base frequency band according to the pre-estimated mixed center frequency to obtain a demodulation signal.
And S28, calculating the square of the time gradient norm of the demodulation signal to obtain the estimated bandwidth of the IMF original component.
And S29, constructing a variation model with constraint according to the estimated bandwidth and the introduced constraint condition.
S2-1, a penalty parameter and a Lagrange multiplication coefficient are introduced to solve the variation model, and a frequency domain expression, a center frequency updating formula and a penalty factor updating formula of the IMF original component are obtained.
S2-2, obtaining an updated frequency domain value, an updated center frequency and an updated Lagrange multiplication coefficient according to the frequency domain expression of the IMF original component, the updated formula of the center frequency and the updated formula of the center frequency.
And S2-3, converting the k IMF original components into amplitude modulation-frequency modulation signals according to the updated frequency domain value, the updated central frequency and the updated penalty factor, and obtaining k IMF components.
In the above steps, the VMD algorithm first decomposes the original signal into k IMF original components, and then redefines the IMF original components into an am-fm signal, where the expression of the IMF components is:
u k (t)＝A k (t)cos[φ k (t)] (1)
in the formula u k (t) represents K IMF components, K ∈ { 1., K }, φ ∈ { 1.,. K }, φ k (t) is a non-decreasing phase function; a. The k (t) represents an envelope function.
Firstly, for each IMF component, solving an analysis signal by using Hilbert transform to obtain a single-side frequency spectrum, wherein the frequency spectrum expression is as follows:
secondly, according to the mixed estimated center frequency, modulating the spectrum to the corresponding base band, and recording as:
finally, the time gradient norm L of the demodulated signal is calculated 2 Estimating the bandwidth of the modal component, introducing constraint conditions, and constructing a constrained variation model as follows:
wherein, { u [ [ u ] ] k Denotes K IMF components after VMD decomposition; { omega [ [ omega ] ] k Denotes the center frequency represented by each IMF component; δ (t) is the dirichlet function; * Performing convolution operation; x represents the original signal.
And then, introducing a penalty parameter alpha and a Lagrangian multiplication coefficient lambda (t) to convert the constrained variation problem into the unconstrained variation problem. The punishment parameter has the function of ensuring the reconstruction precision of the signal in the presence of noise, and the Lagrange coefficient has the function of ensuring the strictness of constraint conditions. The augmented Lagrangian function is shown below:
then, the frequency domain expression of the IMF component obtained by extremum solving the above formula by using an Alternating Direction Multiplier Method (ADMM) is as follows:
in the formula (I), the compound is shown in the specification,
is the current surplus
The inverse Fourier transform is then carried out on the wiener filter to obtain a real part which is a time domain signal u k (t) of (d). Converting the center frequency problem to the frequency domain and solving to obtain the center frequency
The update formula of (2) is as follows:
in the formula (I), the compound is shown in the specification,
representing the center of gravity of the current IMF component power spectrum. Lambda n+1 The update formula of (c) is:
through the above analysis, the algorithm flow of the VMD is as follows:
(1) Initialization
And n;
(2) Updating u according to equations (6) and (7) k And omega k ；
(3) Updating lambda according to equation (8);
(4) Setting a decision accuracy e > 0 if
Stopping the iteration, otherwise returning to the step (2).
And S3, reconstructing the n IMF components screened according to the correlation size to obtain a reconstruction signal.
As shown in fig. 5, step S3 includes:
and S31, determining a correlation coefficient between each IMF component and a bearing vibration signal by adopting the Pearson correlation coefficient.
The correlation coefficient is an index reflecting the degree of closeness of the correlation between the variables. The pearson correlation coefficient is used here to analyze the correlation of the individual IMF components with the original vibration signal. Let sample X and sample Y have the correlation coefficient:
wherein r is a correlation coefficient; cov (X, Y) is the covariance of sample X and sample Y; d (X) is the variance of sample X; d (Y) is the variance of sample Y.
The larger the value of the correlation coefficient r, the higher the correlation between samples. And calculating the correlation coefficient of each IMF component and the original vibration signal, selecting partial components which can represent the original signal most to reconstruct the signal, and extracting the correlation characteristics.
S32, based on the correlation coefficient, selecting the first n IMF components which are ordered from large to small according to the correlation coefficient, and accumulating the first n IMF components to obtain a reconstructed signal.
And S4, obtaining a multi-dimensional feature vector according to the maximum value, the average value and the kurtosis of the extracted reconstruction signal, and singular values, sample entropies and energy entropies of the n IMF components. And calculating the maximum value, the average value and the kurtosis value of each sampling point data and the singular value, the sample entropy and the energy entropy of the n IMF component data corresponding to the sampling point according to the extracted reconstructed signal to obtain a multi-dimensional feature vector, wherein the vector dimension is (3 + n).
The method selects the maximum value, the standard value and the kurtosis of the reconstructed signal and singular values, sample entropies and energy entropies of a plurality of main IMF components as the extracted multidimensional characteristics.
Common time domain features: with the gradual degradation of the bearing, the simple features in the ordinary time domain may show obvious changes, but the changes of the simple features often show "stage", may not change for a long time, and may change suddenly in a short time, and the use of such features alone may result in poor extraction of degradation feature information. Three characteristics of maximum value, average value and kurtosis are selected to characterize the simple characteristics. Along with the reduction of the health degree of the bearing, the maximum value and the standard value of the vibration signal of the bearing are gradually increased, and the stage degradation information of the bearing can be obviously embodied; the kurtosis is used for representing the convexity and flatness of the crest of a function graph of a sample, and can be greatly changed when obvious faults occur in a bearing, so that the health degree of an axis can be reflected to a great extent.
Singular values: singular value decomposition, which can decompose a matrix containing signal feature information into different subspaces, is a feature quantity extraction means capable of keeping signal features relatively stable under disturbance and noise. Meanwhile, the singular value can reflect the inherent property and the principal component relation of the matrix. Singular value features are often used in fault diagnosis and health degree prediction of bearings, because VMD decomposition greatly increases data volume while performing adaptive frequency band information description on signals, and high-dimensional related time series can be subjected to dimensionality compression, i.e., concentration of each IMF component information, by using singular value decomposition. Here the singular values of the principal components are taken as part of the multi-dimensional features.
Sample entropy and energy entropy: entropy-like features are generally used to characterize the magnitude of various energies contained in a signal. The sample entropy reflects the complexity of the time series, and the higher the complexity of the series, the larger the value of the sample entropy, and generally, the larger the sample entropy of the signal series as the bearing degrades. Similarly, as the health degree of the bearing is lower and lower, the frequency distribution in the vibration signal changes, and the energy distribution of the vibration signal also changes correspondingly, so the energy entropy also changes accordingly. The sample entropy and energy entropy of each principal IMF component are chosen to be computed as part of the multidimensional feature.
And S5, reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimension reduction algorithm to obtain the health degree feature.
As shown in fig. 6, step S5 includes:
and S51, calculating the joint probability density of the multi-dimensional feature vectors.
And S52, determining the similarity of any two data points in the low-dimensional vector initial solution according to the preset low-dimensional vector initial solution.
And S53, measuring the similarity of the distribution of the multidimensional characteristic vectors and the distribution of the low-dimensional vectors by using KL divergence according to the joint probability density of the multidimensional characteristic vectors and the similarity of any two data points in the initial solution of the low-dimensional vectors.
And S54, calculating gradient according to the similarity of the multi-dimensional feature vector distribution and the low-dimensional vector distribution, and obtaining the current low-dimensional vector.
And S55, repeating the steps S52-S55 for multiple times until the preset iteration times T are met, and obtaining the low-dimensional health degree characteristic.
In the steps, the t-distribution random adjacent embedding (t-SNE) algorithm is a deep learning nonlinear popular learning algorithm and has an excellent dimensionality reduction effect on a high-dimensional nonlinear data set. the specific steps of the t-SNE algorithm are as follows:
(1) Setting multidimensional eigenvector as X = { X = 1 ,x 2 ,...,x n Any two sample data points and similar conditional probabilities are:
in the formula, delta i Is given by x i Is the gaussian distribution variance of the center point. Can be determined by binary search according to the user-specified perplexity, which is defined as:
(2) Calculating the joint probability density of the multidimensional feature vector:
(3) The initial solution for the low-dimensional vector is: y is 0 ＝{y 1 ,y 2 ,...,y n }
(4) Calculating the similarity of the low-dimensional vectors:
(5) Measure the similarity Q of the multidimensional feature vector distribution P and the low-dimensional vector distribution using KL divergence:
(6) Calculating the gradient:
(7) Obtaining a low-dimensional vector:
wherein eta is the learning rate, alpha (t) is the optimization parameter, t represents the t-th iteration, C represents the objective function using the similarity of high-and low-dimensional data distribution,
representing a gradient formula.
(8) Repeating the steps (4) to (7) until a preset iteration time T is met, wherein the low-dimensional vector expression after the T iterations is as follows: y is (T) ＝{y 1 ,y 2 ,...,y n }。
And S6, respectively inputting the health degree characteristics into an SVM (support vector machine) and LSTM network using MOA (motion estimation algorithm) optimization parameters to obtain bearing working condition information and health degree information.
Step S6 comprises:
s61, dividing the 3-dimensional health degree features under the same working condition into a training set and a testing set, taking the health degree features as input, taking the health degree values as label output, inputting the training set data into an LSTM network for training to obtain a health degree prediction model, and inputting the testing set data into the health degree prediction model to obtain health degree information.
S62, dividing the 3-dimensional health degree characteristics under different working conditions into a training set and a testing set, taking the health degree characteristics as input, taking the working condition types as labels to be output, inputting training set data into an SVM to train to obtain a working condition type judgment model, and inputting testing set data into the type judgment model to obtain working condition information.
In the specific embodiment, the effectiveness of the proposed method is verified by selecting an XJTU-SY rolling bearing accelerated life experimental data set of the Western Ann university of transportation. Since the degradation failure of the bearing under normal conditions usually takes a very long time and the data of the whole life cycle of the bearing is difficult to obtain, the degradation process of the bearing is accelerated by maintaining a fixed radial force and a fixed rotating speed to keep a high load, so that the full life cycle monitoring data of the tested bearing is obtained.
In the test, the sampling frequency is set to be 25.6kHz, the sampling interval is 1min, the sampling time length is 1.28s each time, and 32768 points are recorded in each time of sampling. The two acceleration sensors are respectively used for measuring signal data in the horizontal direction and the vertical direction of the bearing, and only signals in the horizontal direction in the test are researched. 3 types of working conditions are designed in the test, and 5 bearings are arranged under each type of working conditions. The operating conditions are shown in Table 2, and the bearing life is shown in Table 3
TABLE 2
TABLE 3
The original vibration signal of the bearing often contains a large amount of high-frequency noise, which seriously affects the extraction of the required characteristic information, so that it is necessary to perform 'denoising' on the signal before the characteristic extraction to reduce the influence of the noise on the characteristic extraction.
As shown in fig. 7, the signal is sampled every 1min to obtain 32768 sample points, and 10000 sample points before each sampling point are intercepted as the experimental data.
First, the MOA is used to optimize the optimal parameters of the VMD, and the optimization is aimed at minimizing the minimum envelope entropy of each IMF component after decomposition. Taking the data of one sampling point as an example, the optimal parameter set is α =4000, k =10, and the VMD decomposition is performed on 10000 data of each sampling point to obtain 10 IMF components and the corresponding frequency bands, as shown in fig. 8, the left side in fig. 8 is a time domain diagram of the original vibration signal and the 10 IMF components, and the right side is the corresponding frequency bands.
Then, the correlation coefficient between each IMF component and the original signal is calculated, a histogram is obtained, as shown in fig. 9, and 3 components with the highest correlation with the original signal are selected to reconstruct the signal. The vibration signal and the reconstructed signal of the whole life cycle of Bearing1_1 under the 1 st working condition are obtained as shown in fig. 10, the upper part is a time domain diagram of the original vibration signal, and the lower part is a denoised time domain diagram. The reconstructed signal edge 'burr' signals are obviously reduced, and the relevant characteristic information of the bearing is more easily extracted from the de-noised signals.
And calculating singular values, sample entropies and energy entropies of 3 IMF components with the highest maximum value, average value, kurtosis and correlation of each sampling point of the reconstructed signal to form a 12-dimensional feature vector, and then reducing the feature vector to 3 dimensions by using a t-SNE dimension reduction method to perform feature quantity visualization.
The working condition characteristics are obtained by the following method: the method comprises the steps of taking Bearing1_3, bearing1_4, bearing1_5, bearing2_1, bearing2_2, bearing2_3, bearing3_2, bearing3_3 and Bearing3_5 of bearings at different damage positions under each working condition respectively, taking 3-dimensional features extracted from reconstructed signals of the Bearing to visualize the data of the whole life cycle of 9 bearings, and representing Bearing data under the working conditions 1,2 and 3, wherein the colors from blue to red represent that the health degree of the bearings is gradually reduced. As can be seen from fig. 11, the extracted features may have a better separation effect on the distribution of the operating conditions, with less overlap of 3 operating conditions.
The health degree characteristic is obtained by the following method: and 3-dimensional features extracted from the reconstructed signals of the 5 bearings under the working condition 1 are visualized, and compared with other two feature extraction methods for verifying the superiority of the extracted health degree features. As shown in FIGS. 12-a, 12-b and 12-c, the color spectrum bars at the lower part of the graph from blue to red represent that the bearing data are normal to fault, different shapes represent 5 different bearing data in the 1 st working condition, the graph a is a visual view obtained by the method, and the vibration signals of the whole working condition form a continuous curve in a three-dimensional space from a normal state to damage, the continuous curve has good aggregation and continuity, and the position of the curve where the characteristic point is located can reflect the health condition of the bearing to a great extent. And a graph b is a visual graph obtained by calculating a 9-dimensional feature vector consisting of the maximum value, the standard value and the kurtosis of 3 IMF components with the strongest correlation with the bearing original vibration signal and reducing the dimensions, and can basically depict the trend from failure to damage, but the visual results cannot be well gathered together, and the effect is not ideal. And a graph c is obtained by calculating singular values, energy entropies and sample entropies of 3 IMF components with strongest correlation with the original vibration signals of the bearing and reducing the formed 9-dimensional characteristic vector to a low dimension, and can be seen that the characteristic curve of the bearing has no good continuity and can not well represent the degradation condition of the bearing.
Then, working condition classification is carried out: and taking the data of the 9 bearings, wherein 4498 sampling points are counted, the extracted characteristic data is R4498 x 3, the sequence of the data is disordered, the first 4000 data are taken as training set training models, and the last 498 data are taken as test set testing models to test the classification effect of the models. The classifier selects a Support Vector Machine (SVM) of a relatively classical machine learning algorithm. When the SVM is used for classification prediction, a penalty parameter c and a kernel function parameter g need to be adjusted, and an optimal parameter [ c ] is selected 0 ,g 0 ]The classification prediction effect can be optimized. The MOA is also used here to optimize two parameters of the SVM, the steps are the same as the above mentioned optimal parameter optimization step of VMD decomposition by MOA algorithm, and the fitness function is the inverse number of the classification accuracy, when it is minimum, the classification accuracy is the highest. Through MOA optimization, an optimal parameter combination c =100,g =0.01 is obtained. The classification result is shown in FIG. 13, and there are (452/498) points in the test set that can be realizedThe current accurate classification is realized, the classification accuracy is 90.7631%, and the effect is ideal.
Then, health degree prediction is performed: and (3) taking the data of the 5 bearings under the 1 st working condition, totaling 616 sampling points, taking the extracted characteristic data as R616 x 3, taking 123 data of the first bearing as a test set, and training a life prediction model by the last four bearings 493. Setting the current remaining life X of the bearing now And total life X all The ratio is used as the current health label Y:
i.e., the initial health label is 0, and finally Y becomes 1 as the bearing becomes progressively damaged, thus normalizing the health label of the bearing to between 0 and 1.
The LSTM, as a variant of the recurrent neural network, can well capture characteristics on a time sequence, and is a classical neural network model for predicting the residual life. Here, LSTM is used to predict the health of the bearing, the extracted 3-dimensional features are used as the input of the neural network, the corresponding health label is used as the output of the neural network, the prediction model is trained using the training set, and the performance of the model is tested using the test set, with the experimental results shown in fig. 14. As can be seen from the figure, the bearing health degree predicted by the model is substantially accurate and can reflect the health degree, and the error is within an acceptable range.
In addition, the invention also provides a system for constructing the multidimensional health index of the bearing based on the MOA-VMD, which comprises the following components:
and the information acquisition module is used for acquiring a bearing vibration signal.
And the parameter optimizing and decomposing module is used for optimizing the optimal parameters of the VMD decomposition through an MOA algorithm and decomposing the bearing vibration signals according to the optimized optimal parameters to obtain k IMF components.
And the signal reconstruction module is used for reconstructing the n IMF components screened according to the correlation size to obtain a reconstruction signal.
And the multi-dimensional feature vector acquisition module is used for obtaining a multi-dimensional feature vector according to the maximum value, the average value and the kurtosis of the reconstructed signal, and the singular values, the sample entropies and the energy entropies of the n IMF components obtained by calculation.
And the dimensionality reduction module is used for reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimensionality reduction algorithm to obtain the health degree feature.
And the information output module is used for respectively inputting the health degree characteristics into the SVM and LSTM networks using the MOA to optimize the parameters to obtain the bearing working condition information and the health degree information.
Meanwhile, the invention also discloses a multi-dimensional health index construction device based on the MOA-VMD bearing, which comprises the following steps: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the MOA-VMD bearing multidimensional health indicator based construction method as described above.
The invention also provides a computer-readable storage medium, on which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the steps of the MOA-VMD bearing-based multidimensional health index construction method are realized.
Since the system/apparatus described in the above embodiments of the present invention is a system/apparatus used for implementing the method of the above embodiments of the present invention, a person skilled in the art can understand the specific structure and modification of the system/apparatus based on the method described in the above embodiments of the present invention, and thus the detailed description is omitted here. All systems/devices employed in the method of the above embodiment of the present invention are within the scope of the present invention.
To sum up, the present invention provides a method, a system, a device and a medium for constructing a multidimensional health index of a bearing, as shown in fig. 15, the specific process is as follows: firstly, acquiring a vibration signal of a bearing; secondly, optimizing the optimal parameters of the VMD decomposition by using an MOA algorithm, and decomposing the vibration signals by using the optimized optimal parameters to obtain IMF components; then, calculating the correlation between the IMF components and the original signal to obtain 3 IMF components with the highest correlation for signal reconstruction; then, calculating the maximum value, the average value and the kurtosis of the reconstructed signal, and singular values, sample entropies and energy entropies of 3 IMF components to obtain a 12-dimensional feature vector, and reducing the 12-dimensional feature vector to 3 dimensions by using a t-SNE dimension reduction method; then, inputting the feature vectors of the training set into an SVM (support vector machine) and LSTM neural network which use MOA (model of optimization) to optimize parameters for training, and respectively obtaining a working condition classification model and a health degree prediction model; and finally, inputting the feature vectors of the test set into the two models to verify the reliability of the models.
The method takes accelerated degradation data of the bearing as a research object, optimizes parameters of the VMD by utilizing a mayfly algorithm, decomposes an original vibration signal of the bearing by utilizing the VMD with optimized parameters, selects a part of IMF components with the highest values of the original signal to reconstruct a signal, calculates the maximum value, the average value and the kurtosis of the reconstructed signal, and calculates singular values, sample entropies and energy entropies of main IMF components to form multidimensional characteristic vectors, and then reduces the maximum value, the average value and the kurtosis to a low dimension by utilizing a t-SNE method to obtain multidimensional HI. The multi-dimensional HI is sent into the SVM and the LSTM for verification, and experiments show that the HI constructed by the method contains more kinds of bearing related information than HI constructed by a general method, such as working condition information and health degree information, and has higher practical value.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention has been described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions.
It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the terms first, second, third and the like are for convenience only and do not denote any order. These words are to be understood as part of the name of the component.
Furthermore, it should be noted that in the description of the present specification, the description of the term "one embodiment", "some embodiments", "examples", "specific examples" or "some examples", etc., means that a specific feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Moreover, various embodiments or examples and features of various embodiments or examples described in this specification can be combined and combined by one skilled in the art without being mutually inconsistent.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, the claims should be construed to include preferred embodiments and all changes and modifications that fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include such modifications and variations.
Claims (9)
1. A multi-dimensional health index construction method based on an MOA-VMD bearing is characterized by comprising the following steps:
acquiring a bearing vibration signal;
optimizing the optimal parameter of VMD decomposition through an MOA algorithm, and decomposing the bearing vibration signal according to the optimized optimal parameter to obtain K IMF components;
reconstructing the n IMF components screened according to the correlation magnitude to obtain a reconstructed signal;
obtaining a multidimensional feature vector according to the extracted maximum value, average value and kurtosis of the reconstructed signal and singular values, sample entropies and energy entropies of the n IMF components;
reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimension reduction algorithm to obtain health degree features;
respectively inputting the health degree characteristics into SVM and LSTM networks using MOA to optimize parameters to obtain bearing working condition information and health degree information; inputting the health degree characteristics into an SVM network which is trained by training sets under different working conditions and uses MOA to optimize parameters to obtain bearing working condition information; and inputting the health degree characteristics into an LSTM network trained by a training set under the same working condition specified by the bearing working condition information to obtain health degree information.
2. The method of claim 1, wherein optimizing the optimal parameters of the VMD decomposition by the MOA algorithm comprises:
initializing parameters of the MOA;
according to a predetermined parameter [ k ] 0 ,α 0 ]Performing VMD decomposition on the bearing vibration signal to obtain K IMF original components;
calculating the envelope entropy of each IMF original component, and taking the local minimum envelope entropy as a fitness function;
based on the fitness function, continuously iterating and optimizing until a preset termination condition is met to find out the global minimum envelope entropy and the optimal parameter where the corresponding optimal IMF original component is located;
the fitness function is:
minF＝minE p
wherein E is p Is the envelope entropy, j =1,2,3, N, a (j) is the envelope signal obtained by Hilbert demodulation; e.g. of the type j Is obtained by normalizing a (j), k 0 ,α 0 Respectively a preset decomposition number and a punishment parameter.
3. The MOA-VMD-based multi-dimensional bearing health index construction method of claim 2, wherein decomposing the bearing vibration signal according to the optimized optimal parameters to obtain K IMF components comprises:
solving an analytic signal of each IMF original component through Hilbert transformation to further obtain a single-side frequency spectrum;
modulating the single-side frequency spectrum to a corresponding base frequency band according to the pre-estimated mixed center frequency to obtain a demodulation signal;
calculating the square of the time gradient norm of the demodulation signal to obtain the estimated bandwidth of the IMF original component;
constructing a variation model with constraints according to the estimated bandwidth and the introduced constraint conditions;
a penalty parameter and a Lagrange multiplication coefficient are introduced to solve the variation model to obtain a frequency domain expression of an IMF original component, an updating formula of a center frequency and an updating formula of the Lagrange multiplication coefficient;
obtaining an updated frequency domain value, an updated central frequency and an updated Lagrange multiplication coefficient according to the frequency domain expression of the IMF original component, the update formula of the central frequency and the update formula of the Lagrange multiplication coefficient;
converting the K IMF original components into amplitude modulation-frequency modulation signals according to the updated frequency domain value, the updated center frequency and the updated Lagrangian multiplication coefficient to obtain K IMF components;
wherein the IMF component expression is:
u k (t)＝A k (t)cos[φ k (t)]
in the formula u k (t) denotes the kth IMF component, K ∈ { 1., K }, φ k (t) is a non-decreasing phase function; a. The k (t) represents an envelope function;
the variational model with constraint is as follows:
{u k represents the K IMF components resulting from VMD decomposition; { omega [ (. Omega.) ] k Represents the center frequency corresponding to the IMF component; δ (t) is the dirichlet function; * Performing convolution operation; x represents the original signal.
4. The method for constructing the multi-dimensional health index of the bearing based on the MOA-VMD as claimed in claim 1, wherein the step of reconstructing the n IMF components screened according to the correlation size to obtain the reconstructed signal comprises the steps of:
determining a correlation coefficient of each IMF component and the bearing vibration signal by using a Pearson correlation coefficient;
and based on the correlation coefficient, selecting the first n IMF components which are ordered from large to small according to the correlation number, and accumulating the first n IMF components to obtain a reconstructed signal.
5. The MOA-VMD bearing-based multi-dimensional health index construction method of claim 1, wherein the reducing the multi-dimensional feature vector to 3 dimensions by a t-SNE dimension reduction algorithm to obtain the health feature comprises:
s51, calculating the joint probability density of the multi-dimensional feature vectors;
s52, according to a preset low-dimensional vector initial solution, determining the similarity of any two data points in the low-dimensional vector initial solution;
s53, measuring the similarity of the distribution of the multidimensional characteristic vectors and the distribution of the low-dimensional vectors by using KL divergence according to the joint probability density of the multidimensional characteristic vectors and the similarity of any two data points in the initial solution of the low-dimensional vectors;
s54, calculating gradient and obtaining a current low-dimensional vector according to the similarity of the multi-dimensional feature vector distribution and the low-dimensional vector distribution;
and S55, repeating the steps S52-S55 for multiple times until the preset iteration times T are met, and obtaining the low-dimensional health degree characteristic.
6. The MOA-VMD based bearing multi-dimensional health index construction method of any one of claims 1-5, wherein the inputting the health degree characteristics into SVM and LSTM networks using MOA optimization parameters respectively to obtain bearing condition information and health degree information comprises:
dividing 3-dimensional health degree features under different working conditions into a training set and a testing set, wherein the health degree features serve as input, working condition categories serve as label output, training set data are input into an SVM to be trained to obtain a working condition category judgment model, and testing set data are input into the category judgment model to obtain working condition information;
dividing 3-dimensional health degree features under the same working condition into a training set and a testing set, wherein the health degree features serve as input, the health degree value serves as label output, training set data is input into an LSTM network to be trained to obtain a health degree prediction model, and testing set data is input into the health degree prediction model to obtain health degree information.
7. A multi-dimensional health index construction system based on an MOA-VMD bearing is characterized by comprising the following steps:
the information acquisition module is used for acquiring a bearing vibration signal;
the parameter optimizing and decomposing module is used for optimizing the optimal parameter decomposed by the VMD through an MOA algorithm and decomposing the bearing vibration signal according to the optimized optimal parameter to obtain K IMF components;
the reconstruction signal module is used for reconstructing the n IMF components screened according to the correlation size to obtain a reconstruction signal;
the multi-dimensional feature vector acquisition module is used for obtaining a multi-dimensional feature vector according to the maximum value, the average value and the kurtosis of the reconstruction signal and the singular values, the sample entropies and the energy entropies of the n IMF components obtained through calculation;
the dimensionality reduction module is used for reducing the multidimensional feature vector to 3 dimensions through a t-SNE dimensionality reduction algorithm to obtain health degree features;
the information output module is used for respectively inputting the health degree characteristics into an SVM (support vector machine) and LSTM (least squares metric) network using MOA (motion estimation of arrival) optimization parameters to obtain bearing working condition information and health degree information; firstly, inputting health degree characteristics into an SVM network which is trained by training sets under different working conditions and uses MOA to optimize parameters to obtain bearing working condition information; and inputting the health degree characteristics into the LSTM network trained by the training set under the same working condition specified by the bearing working condition information to obtain the health degree information.
8. A multi-dimensional health index construction device based on an MOA-VMD bearing is characterized by comprising the following steps: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the MOA-VMD bearing based multi-dimensional health indicator construction method of any one of claims 1-6.
9. A computer-readable storage medium having stored thereon computer-executable instructions, which when executed by a processor, implement the steps of the MOA-VMD bearing based multi-dimensional health index construction method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111258215.XA CN114282571B (en) | 2021-10-27 | 2021-10-27 | Method, system, equipment and medium for constructing multidimensional health index of bearing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111258215.XA CN114282571B (en) | 2021-10-27 | 2021-10-27 | Method, system, equipment and medium for constructing multidimensional health index of bearing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114282571A CN114282571A (en) | 2022-04-05 |
| CN114282571B true CN114282571B (en) | 2022-10-14 |
Family
ID=80868750
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111258215.XA Active CN114282571B (en) | 2021-10-27 | 2021-10-27 | Method, system, equipment and medium for constructing multidimensional health index of bearing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114282571B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114812796B (en) * | 2022-04-22 | 2024-05-31 | 中国人民解放军63811部队 | Equipment condition assessment method and related equipment |
| CN115560966B (en) * | 2022-09-30 | 2023-09-15 | 新黎明科技股份有限公司 | Weak fault diagnosis method and system for key components of reinforced sparse filtering fusion motor |
| CN116226646B (en) * | 2023-05-05 | 2023-07-21 | 国家石油天然气管网集团有限公司 | Method, system, equipment and medium for predicting health state and residual life of bearing |
| CN116881636B (en) * | 2023-09-08 | 2023-12-08 | 山东远盾网络技术股份有限公司 | Frequency domain signal analysis and detection method based on artificial intelligence |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100784072B1 (en) * | 2003-09-22 | 2007-12-10 | 김형윤 | Sensors and systems for structural health monitoring |
| CN107766882A (en) * | 2017-10-13 | 2018-03-06 | 哈尔滨理工大学 | Epicyclic gearbox method for diagnosing faults based on the more granularities of data-driven quantization characteristic |
| CN108168924B (en) * | 2017-12-20 | 2020-04-14 | 桂林航天工业学院 | Reciprocating compressor service life prediction method based on VMD (model-based decomposition) and MFSS (model-based decomposition) models |
| CN108444709B (en) * | 2018-04-18 | 2020-11-27 | 北京信息科技大学 | Rolling bearing fault diagnosis method combining VMD and FastICA |
| CN111238807B (en) * | 2020-01-17 | 2021-09-28 | 福州大学 | Fault diagnosis method for planetary gear box |
| CN111488710B (en) * | 2020-04-01 | 2023-02-21 | 武汉理工大学 | Structure damage identification method based on self-adaptive WOA-VMD and range entropy |
| CN111413089A (en) * | 2020-04-08 | 2020-07-14 | 北华大学 | Gear fault diagnosis method based on combination of VMD entropy method and VPMCD |
| CN112348185A (en) * | 2020-11-04 | 2021-02-09 | 上海海事大学 | Lithium battery residual life prediction method based on variational modal decomposition and integrated depth model |
| CN112613646A (en) * | 2020-12-08 | 2021-04-06 | 上海交通大学烟台信息技术研究院 | Equipment state prediction method and system based on multi-dimensional data fusion |
-
2021
- 2021-10-27 CN CN202111258215.XA patent/CN114282571B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN114282571A (en) | 2022-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114282571B (en) | Method, system, equipment and medium for constructing multidimensional health index of bearing | |
| Ma et al. | A novel bearing fault diagnosis method based on 2D image representation and transfer learning-convolutional neural network | |
| Kong et al. | Machine learning in seismology: Turning data into insights | |
| CN110792563B (en) | Wind turbine generator blade fault audio monitoring method based on convolution generation countermeasure network | |
| Del Pezzo et al. | Discrimination of earthquakes and underwater explosions using neural networks | |
| CN101303764B (en) | Method for self-adaption amalgamation of multi-sensor image based on non-lower sampling profile wave | |
| US20110191076A1 (en) | Error detection method and system | |
| Esposito et al. | Automatic discrimination among landslide, explosion-quake, and microtremor seismic signals at Stromboli volcano using neural networks | |
| CN112561165A (en) | Multidimensional time series data prediction method based on combined model | |
| CN108765189B (en) | Open type partial discharge big data management system based on intelligent diagnosis algorithm | |
| Zhang et al. | Bearing performance degradation assessment based on time-frequency code features and SOM network | |
| CN109948726B (en) | Power quality disturbance classification method based on deep forest | |
| CN116560895B (en) | Fault diagnosis method for mechanical equipment | |
| Cárdenas-Peña et al. | Selection of time-variant features for earthquake classification at the Nevado-del-Ruiz volcano | |
| Koley et al. | Wavelet-aided SVM tool for impulse fault identification in transformers | |
| Staszewski | Advanced data pre-processing for damage identification based on pattern recognition | |
| Chen et al. | Fault identification method for planetary gear based on DT-CWT threshold denoising and LE | |
| Yu et al. | Whittle networks: A deep likelihood model for time series | |
| Ishkov et al. | Energy theft detection in smart grids via explainable attention maps | |
| Wang et al. | An automatic feature extraction method and its application in fault diagnosis | |
| CN117782198A (en) | Highway electromechanical equipment operation monitoring method and system based on cloud edge architecture | |
| CN116595468A (en) | Hidden space optimized self-encoder anomaly detection method and system | |
| CN115982621A (en) | Rotary machine residual service life prediction method based on time convolution network | |
| Otero et al. | Intercomparison of deep learning architectures for the prediction of precipitation fields with a focus on extremes | |
| Wu et al. | Classification of complex power quality disturbances based on modified empirical wavelet transform and light gradient boosting machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |