A method to reduce the pathological nature of dynamic load identification model for rotating machinery bearings

By selecting the measurement point combination with the smallest number of conditions in rotating mechanical bearings and the regularization method of truncated singular value decomposition, the pathological problem of frequency response matrix is ​​solved, the dynamic load recognition error is reduced, and the recognition accuracy is improved.

CN114091200BInactive Publication Date: 2024-11-08SOUTHEAST UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202111390635.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-23
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the recognition of bearing dynamic loads of rotating machinery, the frequency response function direct inversion method has a pathological problem with the frequency response matrix, which leads to unfavorable condition of the solution. When using regularization technology alone, it is easy to accidentally remove the effective measurement point response signal, reducing the dynamics Load recognition accuracy.

Method used

By setting multiple response measurement points, calculate the number of conditions of the frequency response function matrix corresponding to each combination, select the combination with the smallest condition number as the optimal measurement point combination, use the truncated singular value decomposition method to eliminate the smaller singular value, and combine the filter factor to optimize the dynamic Load calculation reduces the pathological nature of the frequency response function matrix.

Benefits of technology

Effectively eliminate response points that are prone to pathological causes, reduce the correlation between response points, reduce dynamic load recognition errors, improve recognition accuracy, eliminate the pathological characteristics of the matrix, and reduce the response error.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114091200B_ABST
    Figure CN114091200B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for reducing the pathological property of a dynamic load identification model of a rotating machinery bearing. The method comprises the following steps: optimizing the response point measurement points based on the condition number, traversing all the measurement point combinations, calculating the condition number thereof, and selecting the optimal combination with the smallest condition number to reduce the pathological property caused by the linear correlation of the response points; in the case of the optimal combination, extracting the frequency response function matrix of the bearing support system, adopting the truncated singular value decomposition regularization method to decompose the frequency response function matrix, eliminating the smaller singular values ​​in the matrix, obtaining a well-conditioned matrix to approximate the original matrix, eliminating the influence of the too small singular values ​​and the corresponding eigenvectors on the solution, thereby weakening the pathological property of the equation and improving the accuracy and stability of the dynamic load identification result of the bearing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of bearing dynamic load identification methods, and in particular to a method for reducing the pathological nature of a dynamic load identification model for a rotating machinery bearing. Background Art

[0002] Dynamic load is an important factor affecting the safe and stable operation of rotating machinery. Dynamic load identification technology was first studied in the 1970s and is mainly divided into time domain identification method and frequency domain identification method. The time domain identification method is highly sensitive to initial values ​​and requires convolution operations, which requires a high amount of calculations; the frequency domain identification method has formed a relatively mature theoretical system and is the most commonly used method for dynamic load identification. The frequency domain identification method includes the direct inversion method of the frequency response function and the modal coordinate transformation method. The modal coordinate transformation method has high requirements for the accuracy of the modal parameters, while the direct inversion method of the frequency response function is simple and widely used.

[0003] In the identification of dynamic loads on bearings of rotating machinery, the frequency response matrix is ​​ill-posed when the dynamic loads are identified by directly inverting the frequency response function, which leads to the ill-posedness of the solution. In the 1980s, regularization techniques were developed to improve the ill-posedness in the dynamic load identification problem, including the Tikhonov regularization method and the truncated singular value decomposition (TSVD regularization method). When regularization is used alone to improve the ill-posedness of the matrix, the problem is that while removing small singular values, it is easy to mistakenly remove the effective measurement point response signal, reducing the accuracy of dynamic load identification. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for reducing the pathological nature of a dynamic load identification model for a rotating machinery bearing, the purpose of which is to reduce the pathological nature of a frequency response function in the dynamic load identification model for a rotating machinery bearing, eliminate response points that are prone to cause the pathological nature of the frequency response function, and reduce the dynamic load identification error caused by the response error.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for reducing the pathological nature of a dynamic load identification model for a rotating mechanical bearing comprises the following steps:

[0007] Step 1: Set n bearings that are simultaneously excited by dynamic loads, set m response measurement points, and obtain the bearing dynamic load excitation array and the corresponding response array;

[0008] Step 2: Establish the relationship between dynamic load and vibration response:

[0009] F=(H H H) -1 H H X

[0010] In the above formula, F is the dynamic load; H is the frequency response function matrix of all response measurement points, H H is the complex conjugate transposed matrix of H; X is the vibration response;

[0011] Step 3: Randomly select m from the m response measurement points r , n≤m r ≤m, then the number of combinations of response measurement points is Calculate the system frequency response function matrix corresponding to the i-th combination The condition number:

[0012]

[0013] In the above formula, ||·|| is the matrix norm induced by the vector norm, is the Moore-Pennos generalized pseudo-inverse matrix of the matrix H;

[0014] Traverse all the measurement point combinations under the number of response measurement points, calculate the condition number of the frequency response function matrix corresponding to each combination, find the matrix with the smallest condition number, and take the corresponding measurement point combination as the optimal measurement point combination;

[0015] Step 4: According to the vibration response and frequency response function matrix corresponding to the optimal measurement point combination, the truncated singular value decomposition method is used to eliminate small singular values ​​and calculate the dynamic load of the bearing.

[0016] Further technical solutions are:

[0017] The step 4 specifically includes:

[0018] 1) From the frequency response function matrix H of all response measurement points, obtain the frequency response function matrix H corresponding to the optimal combination measurement point op , and H op It performs a singular value decomposition:

[0019]

[0020] In the above formula, V nn and U mn is an orthogonal matrix of column vectors, which is the matrix H op The two unitary matrices decomposed, U mn ={u1,u2,…,u n}、V nn ={v1,v2,…,v n}; diagonal matrix∑ nn The diagonal elements of i}, i = 1, 2, ..., n, σ i ≥0 is the matrix H op All singular values ​​of are arranged in descending order;

[0021] 2) The dynamic load to be identified is obtained by using the frequency response function matrix after singular value decomposition. The calculation expression is:

[0022]

[0023] In the above formula, F is the dynamic load, is the matrix H op The Moore-Pennos generalized pseudo-inverse matrix, X op is the vibration response of the optimal combined measuring point extracted from the vibration response X of all the response measuring points;

[0024] 3) The truncated singular value decomposition regularization method is used to identify the dynamic load to be identified, and the regularized solution of the identified dynamic load is obtained:

[0025]

[0026] In the above formula, H k To approximate the original matrix H op The low-rank frequency response function matrix, k<n;

[0027]

[0028] In the above formula, ∑ k It is the diagonal matrix obtained by filtering the nk smallest singular values ​​in the diagonal matrix ∑.

[0029] Further technical solutions are:

[0030] Introducing the filter factor into the dynamic load calculation expression, the optimized dynamic load regularization solution is obtained as follows:

[0031]

[0032] Among them, the filtering factor is:

[0033]

[0034] The beneficial effects of the present invention are as follows:

[0035] The method for optimizing the response measurement points is to traverse all the response measurement point combinations and select the combination with the smallest condition number, which not only eliminates the response points that are prone to cause the pathological characteristics of the frequency response function, but also greatly reduces the correlation between the response points and thus reduces the pathological characteristics of the frequency response function matrix.

[0036] On the basis of measurement point optimization based on condition number, the truncated singular value decomposition (TSVD) regularization method based on measurement point optimization is applied to eliminate the matrix ill-conditioned problem in bearing dynamic load identification and the ill-conditioned characteristics of the frequency response function matrix, thereby reducing the dynamic load identification error caused by response error. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The figure is a flow chart of a method according to a specific embodiment of the present invention.

[0038] Figure 2 It is a schematic diagram of the structure of a test system according to a specific embodiment of the present invention.

[0039] Figure 3 This is a force analysis diagram of a test system according to a specific embodiment of the present invention. DETAILED DESCRIPTION

[0040] The specific implementation of the present invention is described below with reference to the accompanying drawings.

[0041] A method for reducing the pathological nature of a dynamic load identification model for a rotating mechanical bearing comprises the following steps:

[0042] Step 1: Set n bearings that are simultaneously excited by dynamic loads, set m response measurement points, and obtain the bearing dynamic load excitation array and the corresponding response array;

[0043] Step 2: Establish the relationship between dynamic load and vibration response:

[0044] F=(H H H) -1 H H X

[0045] In the above formula, F is the dynamic load; H is the frequency response function matrix of all response measurement points, H∈R m×n , R is the real number field, H H is the complex conjugate transposed matrix of H; X is the vibration response;

[0046] Step 3: Randomly select m from the m response measurement points r , n≤m r ≤m, then the number of combinations of response measurement points is Calculate the system frequency response function matrix corresponding to the i-th combination The condition number:

[0047]

[0048] In the above formula, ||·|| is the matrix norm induced by the vector norm, is the Moore-Pennos generalized pseudo-inverse matrix of the matrix H;

[0049] Traverse all the measurement point combinations under the number of response measurement points, calculate the condition number of the frequency response function matrix corresponding to each combination, find the matrix with the smallest condition number, and take the corresponding measurement point combination as the optimal measurement point combination;

[0050] Step 4: According to the vibration response and frequency response function matrix corresponding to the optimal measurement point combination, the truncated singular value decomposition method is used to eliminate small singular values ​​and calculate the dynamic load of the bearing; specifically, it includes:

[0051] 1) From the frequency response function matrix H of all response measurement points, obtain the frequency response function matrix H corresponding to the optimal combination measurement point op , and H op It performs a singular value decomposition:

[0052]

[0053] In the above formula, V nn and U mn is an orthogonal matrix of column vectors, which is the matrix H op The two unitary matrices decomposed include the matrix H op The singular values ​​of the left and right singular vectors; among them, U mn ={u1,u2,…,u n},V nn ={v1,v2,…,v n}; diagonal matrix∑ nn The diagonal elements of i}, i = 1, 2, ..., n, σ i ≥0 is the matrix H op All singular values ​​of are arranged in descending order.

[0054] 2) The dynamic load to be identified is obtained by using the frequency response function matrix after singular value decomposition. The calculation expression is:

[0055]

[0056] In the above formula, F is the dynamic load, is the matrix H op The Moore-Pennos generalized pseudo-inverse matrix, X op It is the vibration response of the optimal combined measurement point extracted from the vibration response X of all response measurement points.

[0057] 3) The truncated singular value decomposition regularization method is used to identify the dynamic load to be identified, and the regularized solution of the identified dynamic load is obtained:

[0058]

[0059] In the above formula, H k To approximate the original matrix H op The low-rank frequency response function matrix is ​​used to eliminate the influence of too small singular values ​​and corresponding eigenvectors on the regular solution, thereby weakening the ill-conditioned characteristics of the equation, k<n;

[0060] The expression of the regularization matrix is:

[0061]

[0062] In the above formula, ∑ k It is the diagonal matrix obtained by filtering the nk smallest singular values ​​in the diagonal matrix ∑. Where k is the cutoff number. When k is a reasonable value, the matrix H k The condition number is moderate, so that the dynamic load regularization solution F k More stable

[0063] Furthermore, the filter factor f is introduced i :

[0064]

[0065] The filter factor f i Substituting into the dynamic load formula, the dynamic load regularization solution F k Optimized to:

[0066]

[0067] The technical solution of the present application is further illustrated by specific embodiments below.

[0068] Example 1

[0069] For example Figure 2 The casing support system shown, identify the dynamic loads on its three bearings, such as Figure 1 As shown, the following steps are included:

[0070] (1) Figure 2 As shown in the figure, the rotor is removed, and vibration sensor measuring points (X1, X2, X3, X4, X5, X6, X7, X8, X9, X10) are arranged on multiple sections of the casing. At the same time, vibration sensor measuring points (B1, B2, B3) are arranged on the internal bearings. The connection between the casing and the bearing is a cone shell-spoke plate connection.

[0071] Impact excitation is applied to 10 measuring points on the casing surface to obtain the vibration responses of the three bearings. The reciprocity of the frequency response function is used to obtain a 10×3-dimensional frequency response function matrix H from the bearing to the casing surface. The coherence coefficient is greater than 0.7, and the frequency response function is considered to be reliable.

[0072] (2) Install the rotor part, such as Figure 3 The figure shows the force analysis diagram of the rotor, where F1-F3 are the dynamic loads on the three bearings; U1 and U2 are the unbalanced forces on the two wheels; l1-l4 are the distances between each bearing and the wheel. An unbalanced amount is set on the two wheels of the rotor, and the unbalanced force (U1, U2) caused by the unbalanced amount simulates different dynamic load conditions. The sensor measures the vibration response X of all measuring points on the casing surface under the running state.

[0073] (3) Optimize the measuring points. The number of measuring points must be no less than the number of dynamic loads to be identified. That is, the number of measuring points selected is at least 3. Select at least 3 measuring points from 10 measuring points, and there are a total of The condition number of the frequency response function matrix of the support system is calculated for each combination, and the measurement point combination with the smallest condition number is the optimized measurement point combination.

[0074] From the 10×3-dimensional frequency response function matrix H of all measurement points, the optimized measurement point frequency response function matrix H is obtained. op ; Extract the optimized measurement point combination response X from all measurement point responses X op .

[0075] (4) Using the truncated singular value decomposition (TSVD) regularization method, the frequency response function matrix H of the optimized measurement point is eliminated. op The influence of the smaller singular values ​​and their eigenvectors on the solution weakens the ill-conditioned characteristics of the matrix and obtains the well-conditioned frequency response function matrix Hk-op of the optimized measurement point. op and H k Substituting into the dynamic load calculation formula, we get Thus, three bearing dynamic loads (F1, F2, F3) are obtained.

[0076] (5) Identify the dynamic loads of the three bearings under any two working conditions, and calculate the unbalanced force on the two wheels equivalent to the dynamic loads on the bearings based on the principle of force and moment balance:

[0077]

[0078]

[0079] Where, U1 and U2 are the unbalanced forces on the two wheels respectively; F1, F2, F3 are the dynamic loads on the three bearings respectively; l1, l2, l3, l4 are the distances between each bearing and the wheel respectively.

[0080] (6) The vector difference of the wheel unbalance force calculated under the two working conditions is compared with the actual unbalance force vector difference to verify the accuracy of dynamic load identification. The results show that the amplitude error of the bearing dynamic load identification does not exceed 20%, and the phase error is about 15°.

[0081] Comparative Example 1

[0082] The test device and conditions used in Example 1 are the same, except that the identification method used does not optimize the measurement points on the casing surface, only inverts the frequency response function H of the support system under all measurement points of the casing, and uses the truncated singular value decomposition (TSVD) regularization method to weaken the matrix ill-conditioning to achieve the identification of the bearing dynamic load, including the following steps:

[0083] (1) After removing the rotor part, impact excitation was applied to 10 measuring points on the casing surface to obtain the vibration responses of the three bearings. The reciprocity of the frequency response function was used to obtain the 10×3-dimensional frequency response function matrix H from the bearing to the casing surface. The rotor part was installed, and the vibration response X of the 10 measuring points on the casing surface was measured by the sensor.

[0084] (2) The frequency response function matrix H is decomposed by the truncated singular value decomposition (TSVD) regularization method, and the smaller singular values ​​are removed to obtain the well-conditioned frequency response function matrix H k .

[0085] (3) Select any two dynamic load conditions and set X and H k Substituting into the dynamic load calculation formula, we get Thus, the dynamic loads of the three bearings without measuring point optimization are identified. According to the principle of force and moment balance, the dynamic loads of the three bearings are equivalent to the unbalanced forces on the two wheels.

[0086] (4) The calculated wheel unbalance force under the two working conditions is taken as the vector difference and compared with the actual unbalance force vector difference to verify the dynamic load identification accuracy. The results show that the amplitude error of the bearing dynamic load identification is about 28%, and the phase error is about 20°.

[0087] Comparative Example 2

[0088] The test device and conditions used in Example 1 are the same, except that the identification method used only optimizes the measurement points on the casing surface, and does not use the truncated singular value decomposition (TSVD) regularization method to process the support system frequency response function to achieve the identification of the bearing dynamic load, including the following steps:

[0089] (1) After removing the rotor part, impact excitation was applied to 10 measuring points on the casing surface to obtain the vibration responses of the three bearings. The reciprocity of the frequency response function was used to obtain the 10×3-dimensional frequency response function matrix H from the bearing to the casing surface. The rotor part was installed, and the vibration response X of the 10 measuring points on the casing surface was measured by the sensor.

[0090] (2) Optimize the measuring points, traverse all measuring point combinations, and calculate the condition number of the frequency response function matrix corresponding to each combination. The measuring point combination with the smallest condition number is the optimized measuring point combination.

[0091] Extract the vibration response of the optimized measurement point combination from the vibration responses of 10 measurement points op , extract the optimized measurement point frequency response function matrix H from the 10×3 dimensional frequency response function matrix H op .

[0092] (3) Select any two dynamic load conditions and set X op and H op Substituting into the dynamic load calculation formula, we get In this way, the dynamic loads of the three bearings after the measurement point optimization are identified. Based on the principle of force and moment balance, the unbalanced forces equivalent to the dynamic loads of the three bearings on the two wheels are calculated.

[0093] (4) The wheel unbalance force calculated under the two working conditions is taken as the vector difference and compared with the actual unbalance force vector difference to verify the dynamic load identification accuracy. The results show that the amplitude error of the bearing dynamic load identification is about 45%, and the phase error is about 35°.

[0094] Through the above specific embodiments and comparative examples, it can be seen that the method of reducing the pathological nature of the dynamic load identification model of the rotating mechanical bearing of the present application traverses all measurement point combinations and selects the optimized measurement point combination with the smallest condition number. The pathological characteristics caused by the linear correlation between the response points are eliminated, and the effective measurement point response signals can be retained; the TSVD regularization method is used to eliminate smaller singular values ​​on the basis of measurement point optimization to improve the dynamic load identification accuracy.

Claims

1. A method for reducing the pathological nature of a dynamic load identification model for a rotating mechanical bearing, characterized in that: The following steps are involved: Step 1: Set n bearings that are simultaneously excited by dynamic loads, set m response measurement points, and obtain the bearing dynamic load excitation array and the corresponding response array; Step 2: Establish the relationship between dynamic load and vibration response: F=(H H H) -1 H H X In the above formula, F is the dynamic load; H is the frequency response function matrix of all response measurement points, H H is the complex conjugate transposed matrix of H; X is the vibration response; Step 3: Randomly select m from the m response measurement points r , n≤m r ≤m, then the number of combinations of response measurement points is Calculate the system frequency response function matrix corresponding to the i-th combination The condition number: In the above formula, ||·|| is the matrix norm induced by the vector norm, is the Moore-Pennos generalized pseudo-inverse matrix of the matrix H; Traverse all the measurement point combinations under the number of response measurement points, calculate the condition number of the frequency response function matrix corresponding to each combination, find the matrix with the smallest condition number, and take the corresponding measurement point combination as the optimal measurement point combination; Step 4: According to the vibration response and frequency response function matrix corresponding to the optimal measurement point combination, the truncated singular value decomposition method is used to eliminate small singular values ​​and calculate the dynamic load of the bearing; The step 4 specifically includes: 1) From the frequency response function matrix H of all response measurement points, obtain the frequency response function matrix H corresponding to the optimal combination measurement point op , and H op It performs a singular value decomposition: In the above formula, V nn and U mn is an orthogonal matrix of column vectors, which is the matrix H op The two unitary matrices decomposed, U mn ={u1,u2,…,u n }、V nn ={v1,v2,…,v n }; diagonal matrix∑ nn The diagonal elements of i }, i = 1, 2, ..., n, σ i ≥0 is the matrix H op All singular values ​​of are arranged in descending order; 2) The dynamic load to be identified is obtained by using the frequency response function matrix after singular value decomposition. The calculation expression is: In the above formula, F is the dynamic load, is the matrix H op The Moore-Pennos generalized pseudo-inverse matrix, X op is the vibration response of the optimal combined measuring point extracted from the vibration response X of all the response measuring points; 3) The truncated singular value decomposition regularization method is used to identify the dynamic load to be identified, and the regularized solution of the identified dynamic load is obtained: In the above formula, H k To approximate the original matrix H op The low-rank frequency response function matrix, k<n; In the above formula, ∑ k It is the diagonal matrix obtained by filtering the nk smallest singular values ​​in the diagonal matrix ∑; Introducing the filter factor into the dynamic load calculation expression, the optimized dynamic load regularization solution is obtained as follows: Among them, the filtering factor is:

Citation Information

Cited By

  • Shafting-machinery-hull system coupling dynamic load online identification method and forecasting system

    CN119513524A