High-speed bearing fault diagnosis method for adaptive parameter completely generalized Gini coefficient

Abstract

The invention provides a high-speed bearing fault diagnosis method for an adaptive parameter completely generalized Gini coefficient. The method comprises the following steps: 1, establishing different state data sets of a bearing; 2, decomposing the signal to obtain a modal component; 3, selecting a part of modal components to calculate a complete generalized Gini coefficient, namely performing sparse processing on the signal; 4, establishing a self-adaptive optimization model, taking the minimum Pearson correlation as a target function, calculating the correlation of the feature fault information in the step 3, and carrying out self-adaptive selection on FGGI correlation parameters a and p; 5, bearing fault feature data are obtained according to the optimal parameters obtained in the step 4; 6, processing the fault feature data, and removing redundant information; 7, establishing a deep extreme learning machine classification model; 8, performing classification and identification on the deep extreme learning machine; according to the method, the FGGI related parameters are adaptively selected, so that the accuracy of fault feature information after subsequent processing can be ensured.

Description

Technical Field

[0001] The present invention relates to a fault diagnosis method, in particular to a high-speed bearing fault diagnosis method oriented to an adaptive parameter complete generalized Gini coefficient. Background Art

[0002] With the rapid development of new energy technologies, as a key core component, the operating state of high-speed rolling bearings directly affects the reliability of the entire mechanical system. Once a fault occurs, it will not only cause equipment downtime, but may also trigger a chain reaction, resulting in serious economic losses and safety accidents. Therefore, achieving accurate positioning and type identification of bearing faults has important engineering significance. In the field of fault diagnosis, feature extraction based on vibration signals has become the mainstream technical means, and its development trend shows a deep integration with professional domain knowledge and a high degree of integration of signal processing technologies, which provides strong support for improving the accuracy and reliability of fault diagnosis.

[0003] In recent years, with the rapid development of nonlinear theories, especially methods based on entropy value feature extraction have been widely used. However, a single entropy value is difficult to accurately reflect the characteristic information of bearing faults. As an effective state detection method, sparse metric can represent a signal as a linear combination of a small number of atoms. The main purpose is to capture the repeated transient characteristic information generated by bearing fault signals during operation and effectively quantify fault characteristics. Therefore, the complete generalized Gini coefficient (FGGI) has been proposed in the field of sparse metric of vibration signals, which has better discrimination ability for impact and repeated transients. It satisfies the typical attributes of sparse metric and has better anti-noise and random transient interference capabilities. However, the selection of its two parameters is too dependent on prior knowledge and lacks self-adaptability. Therefore, it is particularly urgent to study parameter adaptive FGGI to automatically adjust the weight parameters and obtain effective vibration signal fault characteristics. At the same time, an appropriate objective function also needs to be selected as the standard for parameter optimization so that the selected parameters are more scientific and reliable.

[0004] As an efficient and practical statistical tool, the Pearson correlation coefficient can intuitively reflect the linear correlation degree between variables, but its application needs to fully consider the specific research scenario and data characteristics. In this study, the minimum correlation coefficient is used as the objective function to optimize the parameters of the complete generalized Gini index (FGGI), so as to determine the optimal weight parameters and norm orders. Although traditional optimization algorithms (such as genetic algorithms, particle swarm algorithms, etc.) are widely used in parameter optimization, they generally have limitations such as long calculation time and low convergence efficiency. In contrast, the bearing fault diagnosis method based on adaptive complete generalized Gini coefficient proposed in this study not only significantly improves the convergence speed, but also can accurately identify the fault types of high-speed bearings, providing a more efficient and reliable solution for the field of fault diagnosis. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a fault diagnosis method based on the complete generalized Gini coefficient that can adaptively select relevant parameters for feature extraction.

[0006] The technical solution adopted by the present invention to solve the above technical problem is: a high-speed bearing fault diagnosis method for the complete generalized Gini coefficient with adaptive parameters; among them, a complete generalized Gini coefficient fault diagnosis method with adaptive parameters disclosed by the present invention includes the following steps 1 to 13.

[0007] Step (1): Establish a data set of different states of the bearing;

[0008] Step (2): Decompose the signal to obtain modal components;

[0009] Step (3): Select some modal components to calculate their complete generalized Gini coefficients, that is, sparsify the signal;

[0010] Step (4): Establish an adaptive optimization model, use the minimum Pearson correlation as the objective function, calculate the correlation of the characteristic fault information in step (3), and adaptively select the FGGI-related parameters a and p;

[0011] Step (5): According to the optimal parameters obtained in step (4), obtain all the bearing fault feature data;

[0012] Step (6): Perform dimensionality reduction processing on the fault feature data to remove redundant information;

[0013] Step (7): Establish a deep extreme learning machine classification model;

[0014] Step (8): Perform classification and recognition for the deep extreme learning machine;

[0015] Compared with the traditional method, the fault diagnosis method proposed by the present invention uses the minimum correlation coefficient as the objective function, adaptively selects the FGGI-related parameters, can effectively screen the fault feature signals, ensure the accuracy of the fault feature information after subsequent processing, and has a high accuracy rate for fault recognition. Brief Description of the Drawings

[0016] Figure 1 It is the implementation flowchart of the method of the present invention. Detailed Embodiment

[0017] The following combines the attached Figure 1 The method of the present invention is described in detail.

[0018] Step (1), obtain the high-speed bearing fault data set, and form the training data matrix X = {S1, S2,..., S10}, and standardize the data in X respectively to obtain a data matrix with a mean of 0 and a standard deviation of 1 Among them, S is the bearing vibration data of different loads, with a size of 100 * 1024. The arrangement order of bearing conditions is: ball fault, inner and outer ring combined fault, healthy state, inner ring fault, outer ring fault;

[0019] Step (2), set relevant parameters, specifically including: the number of decomposed modes k = 4 and the penalty factor α = 2000. Obtain the fault feature set X = {x1, x2,..., x n}, and the specific implementation process includes the following steps (2.1) to step (2.4);

[0020] Step (2.1), according to the formula Decompose part of the data in into k intrinsic mode components, and then perform Hilbert transform on u k (t). When the sum of each mode component is equal to the input signal, establish a variational constraint model Finally, introduce a quadratic penalty factor and a Lagrange multiplier to solve this problem, and finally obtain k mode components;

[0021] Step (2.2), set the values of a and p, reconstruct the IMF in the phase space, and select m data points from the given time series {x i}, (i = 1, 2, 3,..., N) I to form a new sequence Among them, the acquisition processes of m and r values are as follows steps (A) to step (F);

[0022] Step (A), set the parameter range. The optimization range of the FGGI parameters by the adaptive parameter optimization model is that a is an integer in [0.01, 10], p is a random number in [0.01, 10], the population number is 25, the iteration number i is 10, and the individuals a and p in the search space are randomly initialized;

[0023] Step (B), select part of the signal, decompose it into mode components, combine step (2.2) and step (A) to obtain a new sequence Z, and then obtain the FGGI value X = {x1, x2,..., x n} according to step (2.3) and step (2.4);

[0024] Step (C), according to the formula Calculate the correlation coefficient between X = {x1, x2,..., x n} and the original input signal, and set its absolute value as the objective function, that is

[0025] Step (D), construct an adaptive parameter optimization model to globally search the search space and obtain the optimal region. The formula is as follows

[0026] where P i is the position of the i-th population, and p i,j is the position of the j-th individual in the i-th population. k is a random integer; is its corresponding fitness value; parameters r and I are new random numbers updated during the search and iteration processes, where r is a random number in the interval [0, 1], and the value of I is 1 or 2;

[0027] Step (E), determine the hunting range radius R and determine the objective function value

[0028] where is the position of the j-th individual in the i-th population after the update of P2; t is the current iteration number, and T is the maximum iteration number; refers to the new position of the i-th population in P2, is the objective function value of the i-th population updated by P2.

[0029] Step (F), update the values of a and p according to formulas (1) and (2), and then determine whether i is less than 10; if so, after setting i = i + 1, return to Step (C); if not, finally compare all the target values searched and select the optimal combination of a and p obtained under the minimum correlation coefficient;

[0030] Step (2.3) Calculate the sequence distance and the value minF(|r i |) with the minimum correlation coefficient;

[0031] Step (2.4), repeat the above steps to obtain the minimum correlation coefficient, and at the same time output the signal after FGGI sparsification. Repeat the above steps to obtain the FGGI sparse signals of multiple types of fault signals;

[0032] Step (3), obtain the FGGI information X = {x1, x2,..., x n} of the fault features, a data set with a size of 1000 * 128. Dimension reduction is performed on X = {x1, x2,..., x n} to obtain (x i , y i ) ∈ R n ×R m , where n and m represent the number of features and the number of categories respectively;

[0033] Step (4): Input the obtained FGGI feature information into the Deep Extreme Learning Machine (DELM) to construct a high-speed bearing fault diagnosis model:

[0034] where is the output weight of the i-th hidden layer node, is the input weight of the i-th hidden layer node, b i is the threshold of the i-th hidden layer node, and G(·) is the sigmoid activation function; the matrix expression of ELM is obtained as Hβ = T;

[0035] Step (5): Through the least squares solution such that According to the formula the output weight is obtained;

[0036] Step (6): Apply the Autoencoder AE to ELM, and its weight β is transformed into Stack multiple Extreme Learning Machine Autoencoders ELM-AE to form DELM;

[0037] Step (7): Diagnose the high-speed bearing fault test set to achieve accurate classification of fault types.

Claims

1. A high-speed bearing fault diagnosis method for an adaptive parameter complete generalized Gini coefficient, characterized in that, Specifically, it includes the following steps: Step (1), obtain the high-speed bearing fault dataset, and form the training data matrix X = {S1, S2,..., S 10}, and perform standardization processing on the data in X respectively to obtain a data matrix with a mean of 0 and a standard deviation of 1 where S is the bearing vibration data under different loads, with a size of 100 * 1024, and the arrangement order of the bearing conditions is: ball fault, inner and outer ring combination fault, healthy state, inner ring fault, outer ring fault; Step (2), set relevant parameters, specifically including: the number of decomposed modes k = 4 and the penalty factor α = 2000. Obtain the fault feature set X = {x1, x2, …, x n} through FGGI. The specific implementation process includes the following steps (2.1) to (2.4); Step (2.1), according to the formula Decompose part of the data in into k intrinsic mode components, and then perform Hilbert transform on u k (t). When the sum of each mode component is equal to the input signal, establish a variational constraint model Finally, introduce the quadratic penalty factor α and the Lagrange multiplier λ to solve the variational problem, and finally obtain k IMF mode components; Step (2.2), set the values of a and p, reconstruct the IMF in the phase space, and select m data points from the given time series {x i}, (i = 1, 2, 3,..., N) I to form a new sequence Step (2.3), calculate the sequence distance and the value minF(|r i |) with the smallest correlation coefficient; Step (2.4), repeat the above steps to obtain the minimum correlation coefficient, and at the same time output the signal after FGGI sparsification. Repeating the above steps will obtain the FGGI sparse signals of multiple types of fault signals; Step (3), obtain the FGGI sparsification feature signal X = {x1, x2,..., x n}, a data set of size 1000 * 128, and perform dimensionality reduction on X = {x1, x2,..., x n} to obtain (x i , y i ) ∈ R n ×R m , where n and m represent the number of features and the number of classes respectively; Step (4), input the obtained feature information into the Deep Extreme Learning Machine (DELM) to construct a high-speed bearing fault diagnosis model: where β i = [β i1 , β i2 , …, β im T is the output weight of the i-th hidden layer node, a i = [a i1 , a i2 , …, a in T is the input weight of the i-th hidden layer node, b i is the threshold of the i-th hidden layer node, G(·) is the sigmoid activation function; the matrix expression of ELM is obtained as Hβ = T;​​ Step (5), through the least squares solution such that According to the formula Output the weights; Step (6), apply the autoencoder AE to ELM, and its weight β is transformed into Stack multiple extreme learning machine autoencoders ELM-AE to form DELM; Step (7), diagnose the high-speed bearing fault test set to achieve accurate classification of fault types.

2. The high-speed bearing fault diagnosis method for an adaptive parameter complete generalized Gini coefficient according to claim 1, characterized in that The specific implementation process of the said Step (2.2) is as shown in the following Steps (A) to (F): Step (A): Set the parameter range. The optimization range of the FGGI parameters by the adaptive parameter optimization model is that \(m\) is an integer in \([1, 5]\), \(r\) is a random number in \([0.01, 1]\), the population number is 25, the number of iterations \(i\) is 10, and individuals \(a\) and \(p\) in the search space are randomly initialized; Step (B): Select part of the signal, decompose it into intrinsic mode components, combine Step (2.2) and Step (A) to obtain a new sequence \(Z\), and then obtain the FGGI value \(X=\{x_1, x_2, \ldots, x\) n \}; Step (C), according to the formula calculate the correlation coefficient between X = {x1, x2,..., x n} and the original input signal, and set its absolute value as the objective function, that is Step (D), construct an adaptive parameter optimization model, perform a global search on the search space, and obtain the optimal region. The formula is as follows: where P i is the position of the i-th population, p i,j is the position of the j-th individual in the i-th population, and k is a random integer; F i new,P1 is its corresponding fitness value; the parameters r and I are new random numbers updated during the search and iteration processes, where r is a random number in the interval [0, 1] and the value of I is 1 or 2; Step (E), second step, determine the hunting range radius R and determine the objective function value In the formula is the position of the j-th individual in the i-th population after the update of P2; t is the current iteration number, and T is the maximum iteration number; refers to the new position of the i-th population in P2, and is the objective function value of the i-th population updated in P2; Step (F), update the values of a and p according to Formula (2) and Formula (3), and then determine whether i is less than 10; if so, after setting i = i + 1, return to Step (C); if not, finally compare all the target values searched and select the optimal combination of a and p obtained under the minimum correlation coefficient.

Images