A method and device for identifying the axis trajectory of a rotating mechanical rotor

Abstract

An embodiment of the present invention discloses a method and device for identifying the axis trajectory of a rotating machinery rotor. The method uses a vibration sensor to collect two vibration time-domain signal sequences with a phase difference of 90° at the same measuring point and within the same time period while the rotating machinery rotor is running. The two vibration time-domain signal sequences are preprocessed to obtain two vibration amplitude sequences. A matrix space is constructed using the two vibration amplitude sequences. Feature extraction is performed on the matrix space based on a matrix operator to construct a labeled axis trajectory space vector data set. An RBF support vector machine model is trained using the labeled axis trajectory space vector data set to obtain a trained axis trajectory recognition model. The axis trajectory recognition model is used to recognize axis trajectory space vectors with unknown labels. In the embodiment of the present invention, matrix operator feature extraction technology is combined with an RBF support vector machine to jointly identify the axis trajectory of a rotating machinery rotor, thereby improving the accuracy of identifying the axis trajectory of the rotating machinery rotor.

Description

Technical Field

[0001] The present invention relates to the technical field of rotating machinery fault diagnosis, and more particularly to a method and device for identifying the axis trajectory of a rotating machinery rotor. Background Art

[0002] Currently, intelligent learning methods are used to identify the axis trajectory of rotating machinery rotors. This method uses the original axis trajectory signal itself, or one or more abstracted features of the original axis trajectory signal, as samples. Through intelligent learning of the sample space, a trained classifier is obtained, which is then used to intelligently classify new data. However, the recognition accuracy of the classifier obtained by this method is low, and it cannot accurately identify the fault mode of rotating machinery rotors. Summary of the Invention

[0003] In view of this, an embodiment of the present invention discloses a method and device for identifying the axis center trajectory of a rotating machine rotor, so as to improve the accuracy of identifying the axis center trajectory of the rotating machine rotor.

[0004] The technical solutions provided by the embodiments of the present invention are as follows:

[0005] A first aspect of an embodiment of the present invention provides a method for identifying the axis trajectory of a rotor of a rotating machine, comprising:

[0006] Using a vibration sensor to collect two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same period of time when the rotating machinery rotor is running;

[0007] Preprocessing the two vibration time domain signal sequences to obtain two vibration amplitude sequences;

[0008] Constructing a matrix space using the two vibration amplitude sequences, wherein the matrix space is used to describe the axis trajectory represented by the vibration time domain signal sequence;

[0009] Performing feature extraction on the matrix space according to a matrix operator to obtain a seventeen-dimensional space vector;

[0010] Using the seventeen-dimensional space vectors obtained when the rotating machinery rotor operates in different fault modes, a labeled axis trajectory space vector dataset is constructed;

[0011] Using the labeled axis trajectory space vector data set to train a radial basis function (RBF) support vector machine model to obtain a trained axis trajectory recognition model;

[0012] The axis trajectory recognition model is used to recognize the axis trajectory space vector of the unknown label.

[0013] In a possible implementation, the step of training a radial basis function (RBF) support vector machine (SVM) model using the labeled axis trajectory spatial vector dataset to obtain a trained axis trajectory recognition model includes:

[0014] performing batch normalization processing on the labeled axis trajectory space vector dataset to obtain a normalized axis trajectory space vector dataset;

[0015] According to a preset regularization coefficient and parameter shrinking strategy, the RBF support vector machine model is trained using the normalized axis trajectory space vector dataset;

[0016] The K-fold cross-validation method is used to evaluate the generalization ability of the RBF support vector machine model. When the prediction accuracy of the RBF support vector machine model is greater than a preset value, the training of the RBF support vector machine model is terminated to obtain a trained axis trajectory recognition model.

[0017] In a possible implementation, there are two vibration sensors, and the method further includes:

[0018] The two vibration sensors are arranged at the same axial position, and the radial position difference between the two vibration sensors is 90°; wherein, the connection between the vibration sensor and the measuring point is by gluing or rigid mechanical fastening.

[0019] In a possible implementation, preprocessing the two vibration time-domain signal sequences to obtain two vibration amplitude sequences includes:

[0020] The two vibration time domain signals are preprocessed by filtering, time domain averaging, detrending and normalizing to obtain two vibration amplitude sequences.

[0021] In a possible implementation, constructing a matrix space using the two vibration amplitude sequences includes:

[0022] Determining a minimum amplitude and a maximum amplitude according to the two vibration amplitude sequences;

[0023] Constructing an amplitude interval according to the minimum amplitude value and the maximum amplitude value;

[0024] Using any one of the two vibration amplitude sequences as the coordinates of the rows in the matrix, and using the other one of the two vibration amplitude sequences as the coordinates of the columns in the matrix;

[0025] Determining effective amplitude sequence coordinates using the amplitude interval and the two vibration amplitude sequences;

[0026] The position values ​​corresponding to the valid amplitude sequence coordinates in the matrix are set to 1, and the values ​​of the other positions in the matrix except for the position with the value 1 are set to 0, so as to obtain a matrix space with values ​​of 0 or 1.

[0027] In a possible implementation, performing feature extraction on the matrix space according to a matrix operator to obtain a seventeen-dimensional space vector includes:

[0028] Traversing and extracting a submatrix space from the matrix space according to a matrix operator;

[0029] Calculating the gradient of each sub-matrix space;

[0030] According to the gradient of each of the sub-matrix spaces, the number of gradients of the matrix space in seventeen directions is counted to obtain a seventeen-dimensional space vector.

[0031] In one possible implementation, the fault modes include: rotor misalignment characterized by an outer ring ellipse, rotor imbalance characterized by a stable ellipse, rotor rubbing characterized by a divergent rotor, and petal-shaped oil film oscillation.

[0032] In a possible implementation, the batch normalization processing of the labeled axis trajectory space vector dataset includes:

[0033] Batch normalization is performed as follows:

[0034] (ab) / c;

[0035] Wherein, a is any axis trajectory space vector in the labeled axis trajectory space vector dataset, b is the mean of the labeled axis trajectory space vector dataset, and c is the standard deviation of the labeled axis trajectory space vector dataset.

[0036] In one possible implementation, the K-fold cross-validation method is used to evaluate the generalization ability of the RBF support vector machine model, and the training of the RBF support vector machine model is terminated when the prediction accuracy of the RBF support vector machine model is greater than a preset value, including:

[0037] Dividing the labeled axis trajectory space vector dataset into K parts;

[0038] Each time, K-1 copies are selected as training sets, and the remaining 1 copy is used as the test set;

[0039] The K-fold cross validation method was used to consider the generalization ability of the RBF support vector machine model;

[0040] When the average prediction accuracy of the test set is greater than 90%, it is determined that the generalization ability of the axis trajectory recognition model meets the standard, and the training of the RBF support vector machine model is terminated.

[0041] A second aspect of the present application provides a device for identifying the axis trajectory of a rotor of a rotating machine, the device comprising:

[0042] An acquisition unit is used to acquire two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same period of time using a vibration sensor when the rotor of the rotating machinery is running;

[0043] An acquisition unit is used to acquire two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same period of time using a vibration sensor when the rotor of the rotating machinery is running;

[0044] A preprocessing unit, configured to preprocess the two vibration time domain signal sequences to obtain two vibration amplitude sequences;

[0045] a matrix construction unit, configured to construct a matrix space using the two vibration amplitude sequences, wherein the matrix space is used to describe the axis trajectory represented by the vibration time domain signal sequence;

[0046] A feature extraction unit, configured to extract features from the matrix space according to a matrix operator to obtain a 17-dimensional space vector;

[0047] A data set construction unit, configured to construct a labeled axis trajectory space vector data set using the seventeen-dimensional space vectors obtained when the rotating machinery rotor operates in different fault modes;

[0048] A training unit, configured to train a radial basis function (RBF) support vector machine (SVM) model using the labeled axis trajectory spatial vector data set to obtain a trained axis trajectory recognition model;

[0049] The recognition unit is used to recognize the axis trajectory space vector of the unknown label by using the axis trajectory recognition model.

[0050] It can be seen from the above technical solution that an embodiment of the present invention discloses a method and device for identifying the axis trajectory of a rotating mechanical rotor, which uses a vibration sensor to collect two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and in the same period of time when the rotating mechanical rotor is running; the two vibration time domain signal sequences are preprocessed to obtain two vibration amplitude sequences; a matrix space is constructed using the two vibration amplitude sequences; features are extracted from the matrix space according to a matrix operator to obtain a seventeen-dimensional space vector; the seventeen-dimensional space vectors obtained when the rotating mechanical rotor is running in different fault modes are used to construct a labeled axis trajectory space vector data set; a radial basis function (RBF) support vector machine model is trained using the labeled axis trajectory space vector data set to obtain a trained axis trajectory recognition model; and the axis trajectory recognition model is used to recognize axis trajectory space vectors with unknown labels. It can be seen that the use of matrix operators to extract features in the embodiment of the present invention can extract more appropriate axis trajectory features. By using the RBF support vector machine model to perform classification learning on a known, labeled axis trajectory space vector data set, a more generalized recognition model with better classification effect can be obtained. In this way, the use of matrix operator feature extraction technology combined with the RBF support vector machine model to jointly identify the axis trajectory of the rotating machinery rotor can improve the accuracy of the axis trajectory identification of the rotating machinery rotor, so that the identified axis trajectory can be used to assist in judging whether the rotating machinery rotor contains misalignment, imbalance, bending and other types of faults, thereby improving the accuracy of rotating machinery rotor fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the disclosed drawings without any creative work.

[0052] Figure 1 This is a flow chart of a method for identifying the axis trajectory of a rotor of a rotating machine disclosed in an embodiment of the present invention;

[0053] Figure 2 A schematic diagram of a set of vibration time domain signal sequences disclosed in an embodiment of the present invention;

[0054] FIG3(a)-(d) are visualization diagrams of a matrix space of divergence, inner ring ellipse, ellipse and outer ring ellipse disclosed in an embodiment of the present invention;

[0055] 3(e)-(h) are visual diagrams of a matrix space of two ellipses at different relative positions in an outer ring ellipse disclosed in an embodiment of the present invention;

[0056] Figure 4This is a schematic diagram of a 3-row x 3-column submatrix space selected at random according to an embodiment of the present invention;

[0057] Figure 5 A schematic diagram of seventeen directions of a matrix space disclosed in an embodiment of the present invention;

[0058] Figure 6 A schematic diagram of a seventeen-dimensional vector in a matrix space disclosed in an embodiment of the present invention;

[0059] Figure 7 A schematic diagram of a confusion matrix of a prediction result disclosed in an embodiment of the present invention;

[0060] Figure 8 This is a flow chart of another method for identifying the axis trajectory of a rotor of a rotating machine disclosed in an embodiment of the present invention;

[0061] Figure 9 The present invention is a schematic structural diagram of a device for identifying the axis trajectory of a rotor of a rotating machine disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] An embodiment of the present invention discloses a method and device for identifying the axis trajectory of a rotating machinery rotor. The method comprises the following steps: using a vibration sensor to collect two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same time period when the rotating machinery rotor is running; preprocessing the two vibration time domain signal sequences to obtain two vibration amplitude sequences; constructing a matrix space using the two vibration amplitude sequences; performing feature extraction on the matrix space according to a matrix operator to obtain a 17-dimensional space vector; constructing a labeled axis trajectory space vector data set using the 17-dimensional space vectors obtained when the rotating machinery rotor is running in different fault modes; training a radial basis function (RBF) support vector machine model using the labeled axis trajectory space vector data set to obtain a trained axis trajectory recognition model; and recognizing axis trajectory space vectors with unknown labels using the axis trajectory recognition model. It can be seen that the use of matrix operators to extract features in the embodiment of the present invention can extract more appropriate axis trajectory features. By using the RBF support vector machine model to perform classification learning on a known, labeled axis trajectory space vector data set, a more generalized recognition model with better classification effect can be obtained. In this way, the use of matrix operator feature extraction technology combined with the RBF support vector machine model to jointly identify the axis trajectory of the rotating machinery rotor can improve the accuracy of the axis trajectory identification of the rotating machinery rotor, thereby improving the accuracy of the rotating machinery rotor fault diagnosis.

[0064] See also Figure 1 , a flowchart of a method for identifying the axis trajectory of a rotor of a rotating machine disclosed in an embodiment of the present invention, the method comprising:

[0065] Step S101: using a vibration sensor to collect two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same time period when the rotor of the rotating machinery is running;

[0066] See also Figure 2 , a set of collected vibration time domain signal sequences disclosed in an embodiment of the present invention.

[0067] It is understood that before using the vibration sensor to collect the vibration time series signal sequence, the shaft center trajectory simulation signals of the rotating machinery rotor faults with different faults and densities are first generated. The shaft center trajectory patterns of the rotating machinery rotor faults are first classified into circles, ellipses, double ellipses (including inner and outer ellipses), divergent, and so on.

[0068] For circle and ellipse data, the difference in the axis trajectory graph can be adjusted by adjusting the ratio of the major and minor axes of the ellipse. Among them, the ratio of the major and minor axes a / b can be as small as 1 and as large as 2, and 8 numbers are taken at equal intervals. These 10 ratios of the major and minor axes are used as circle and ellipse data. In this embodiment of the present invention, the number of sampling points can be set to ten numbers with equal intervals between 330 and 630. According to the above classification method and adding noise, a total of 2000 sets of simulated data for circles and ellipses are obtained. It should be noted that the embodiment of the present invention does not limit the specific generation method, number of sampling points, quantity, etc. of the simulated data.

[0069] For double-ellipse figure-of-eight data, first, consider the inner ellipse and outer ellipse classifications. Second, the relative positions of the two ellipses also differ. Specifically, the inner ellipse's larger ellipse contains the smaller ellipse without contact; the inner and outer ellipse's larger and smaller ellipses are tangent at the uppermost point of the larger ellipse; the inner and outer ellipse's larger and smaller ellipses are tangent at the lowermost point of the larger ellipse; the inner and outer ellipse's larger and smaller ellipses are tangent at the leftmost point of the larger ellipse; and the inner and outer ellipse's larger and smaller ellipses are tangent at the rightmost point of the larger ellipse. Finally, the signal density can be divided by adjusting the number of sampling points within a cycle. This embodiment of the present invention can also set the number of sampling points to ten equidistant numbers between 330 and 630. Using the above classification method and adding noise, a total of 2000 sets of simulation data for the double-ellipse figure-of-eight data were generated, including 1000 sets for the inner and outer ellipse rings. It should be noted that this embodiment does not limit the specific generation method, number of sampling points, or quantity of the simulation data.

[0070] For divergent data, a random number within the range [-1, 1] can be selected as the analog signal. In this embodiment of the present invention, the number of sampling points can also be set to ten equidistant numbers between 330 and 630. Based on the above classification method, a total of 2000 sets of divergent analog data are generated. It should be noted that this embodiment of the present invention does not limit the specific generation method, number of sampling points, or quantity of analog data.

[0071] It should be noted that in the embodiments of the present invention, two vibration sensors can be provided, and the two vibration sensors are arranged at the same axial position, with the radial position of the two vibration sensors differing by 90°. Thus, in the embodiment of the present invention, the two vibration sensors are arranged at the same axial position in the measurement point selection to ensure the signals are homologous, and the two vibration sensors are selected to have a 90° angle difference in the measurement point direction to ensure orthogonal signals. This ensures that the collected signals can accurately restore and reflect the axis trajectory of the rotating machinery rotor. In the embodiments of the present invention, two vibration sensors that can simultaneously measure synchronously can also be selected in the time direction to synchronize the signals, further ensuring that the collected signals can accurately restore and reflect the axis trajectory of the rotating machinery rotor.

[0072] It should be noted that the connection between the vibration sensor and the measuring point in the embodiments of the present application is achieved by gluing or rigid mechanical fastening. This makes the vibration sensor's installation and connection method closer to a rigid connection, resulting in better vibration transmission and less signal distortion, ensuring that the process of vibration transmission from the rotating machinery rotor to the vibration sensor is minimally affected.

[0073] Step S102: preprocessing the two vibration time domain signal sequences to obtain two vibration amplitude sequences;

[0074] It should be noted that the inventors, taking into account the presence of noise and vibration interference during the actual operation of rotating machinery, as well as the potential DC component errors during signal sampling and transmission, preprocessed the two vibration time-domain signal sequences before constructing the matrix space. Specifically, the two vibration time-domain signals were preprocessed through filtering, time-domain averaging, detrending, and normalization to obtain two vibration amplitude sequences, thus avoiding the impact of noise and vibration interference on the signals.

[0075] It should be noted that the filtering method can be low-pass filtering with a maximum pass frequency of 100 Hz. Time domain averaging removes the DC component from the signal by removing the mean. Normalization normalizes the frequency value by dividing by the number of sampling points.

[0076] It is understood that the embodiment of the present invention can perform time-domain averaging and low-pass filtering on the vibration time-domain signal sequences collected corresponding to the above 6000 groups of analog signals. Time-domain averaging specifically involves calculating the average value of the data within each group, subtracting the average value of the group from the data within each group, and obtaining data without DC components. Low-pass filtering specifically involves performing a time-frequency domain Fast Fourier Transformation (FFT) on the data to remove components with frequencies higher than f1. In the embodiment of the present invention, f1 can be 100 Hz, etc., without specific limitation.

[0077] Step S103: constructing a matrix space using the two vibration amplitude sequences, wherein the matrix space is used to describe the axis trajectory represented by the vibration time domain signal sequence;

[0078] It should be noted that a matrix space of N rows × N columns can be constructed. For example, the value of N can be 100 or 200, and there is no specific limitation on the value of N. Optionally, constructing a matrix space of 100 rows × 100 columns may provide better performance in the algorithm.

[0079] Step S104: performing feature extraction on the matrix space according to a matrix operator to obtain a 17-dimensional space vector;

[0080] The 17-dimensional space vector refers to the number of gradients in each of the 17 directions of the matrix space. It can be understood that the 17-dimensional space vector is 17 values.

[0081] Step S105: constructing a labeled axis trajectory space vector dataset using the seventeen-dimensional space vectors obtained when the rotating machinery rotor operates in different fault modes;

[0082] It should be noted that the fault modes include: rotor misalignment with the fault characteristic of an outer ring ellipse, rotor imbalance with the fault characteristic of a stable ellipse, rotor friction with the fault characteristic of a divergent rotor, and oil film oscillation with the fault characteristic of a petal shape, etc.

[0083] It is understandable that by processing the above 6000 sets of data in steps S101 to S104 , a labeled axis trajectory space vector data set is obtained.

[0084] Step S106: using the labeled axis trajectory space vector data set to train a radial basis function (RBF) support vector machine model to obtain a trained axis trajectory recognition model;

[0085] It should be noted that the radial basis function (RBF) is a real-valued function whose value depends solely on the distance from the origin. Support vector machines (SVMs) have significant advantages in handling nonlinear, small sample, and high-dimensional pattern recognition problems. The RBF SVM model uses the RBF kernel as the kernel function. Using the RBF SVM model for classification learning can produce a more generalizable recognition model with better classification results.

[0086] Step S107: using the axis trajectory recognition model to identify the axis trajectory space vector of the unknown tag.

[0087] In summary, the use of matrix operators to extract features in the embodiment of the present invention can extract more appropriate axis trajectory features. By using the RBF support vector machine model to perform classification learning on a known, labeled axis trajectory space vector data set, a more generalized recognition model with better classification effect can be obtained. In this way, the use of matrix operator feature extraction technology combined with the RBF support vector machine model to jointly identify the axis trajectory of the rotating machinery rotor can improve the accuracy of the axis trajectory identification of the rotating machinery rotor, thereby improving the accuracy of the rotating machinery rotor fault diagnosis.

[0088] To further optimize the above embodiment, step S103 of the method for identifying the axis trajectory of the rotor of a rotating machine in the embodiment of the present invention specifically includes:

[0089] Step S1031, determining the minimum amplitude and the maximum amplitude according to the two vibration amplitude sequences;

[0090] It is understandable that since the vibration signal of the rotating machinery rotor may have inconsistent amplitude intervals of the two vibration amplitude sequences, and the matrix space is a fixed N-row x N-column space, it is necessary to adaptively unify the two vibration amplitude sequences and jointly construct the amplitude interval.

[0091] Take a set of data as an example. One vibration amplitude sequence is A1, and the other is A2. Find the maximum and minimum amplitude values ​​common to the two amplitude sequences, and record them as Amax and Amin.

[0092] Step S1032: constructing an amplitude interval according to the minimum amplitude and the maximum amplitude;

[0093] For better understanding, let's take the example of constructing a 100-row x 100-column matrix space. First, construct a 100-row x 100-column all-zero matrix M, and consider Amax as the 99th row and Amin as the 2nd row. This is done to ensure that the amplitude sequence does not overflow the matrix.

[0094] The formula for constructing the amplitude interval is as follows:

[0095] det = (Amax - Amin) / 98;

[0096] Where det is the true amplitude difference between adjacent rows or columns of the matrix.

[0097] It is understandable that because the data needs to be mapped to a 100-row x 100-column matrix, and to prevent data overflow, the actual data only occupies a 98-row x 98-column matrix, which is equivalent to dividing the maximum value minus the minimum value into 98 equal parts, each of which represents the actual amplitude difference between adjacent rows or columns of the matrix.

[0098] Step S1033: using any one of the two vibration amplitude sequences as the coordinates of the rows in the matrix, and using the other one of the two vibration amplitude sequences as the coordinates of the columns in the matrix;

[0099] Step S1034: Determine valid amplitude sequence coordinates using the amplitude interval and the two vibration amplitude sequences;

[0100] The formula for determining the effective amplitude sequence coordinates is as follows:

[0101] Mij=1

[0102] i=a1k / det

[0103] j=a2k / det

[0104] Among them, a1k is the k-th data in the vibration amplitude sequence A1, a2k is the k-th data in the vibration amplitude sequence A2, k is taken across A1 and A2, i is the row coordinate of the matrix M, and j is the column coordinate of the matrix M.

[0105] It should be noted that since the matrix coordinates are integers between 1 and 100, the two formulas i=a1k / det and j=a2k / det need to add a rounding operation after the division.

[0106] Step S1035: Set the position value corresponding to the valid amplitude sequence coordinate in the matrix to 1, and set the values ​​of the other positions in the matrix except the position value 1 to 0, so as to obtain a matrix space with values ​​of 0 or 1.

[0107] For example, suppose the maximum value of two vibration amplitude sequences is 2 and the minimum value is -2, and det = 4 / 98. Assume the first data point of vibration amplitude sequence A1 is 1, and the first data point of vibration amplitude sequence A2 is 2; then 1 / (4 / 98) = 24.5 ≈ 25, and 2 / (4 / 98) = 49, resulting in the valid amplitude sequence coordinates (25, 49). In the 100-row × 100-column matrix space, the value of the 49th position in the 25th row is set to 1.

[0108] 3(a)-(d), which are respectively a visualization diagram of the matrix space of divergence, inner ellipse, ellipse and outer ellipse. It should be noted that the diagram is intended to facilitate understanding of the present invention, and visualization of the matrix space is not necessary in actual application.

[0109] 3(e)-(h), which are respectively a visualization diagram of the matrix space of two ellipses at different relative positions in the outer ring ellipse. It should be noted that the diagram is intended to facilitate understanding of the present invention, and visualization of the matrix space is not necessary in actual application.

[0110] In an embodiment of the present invention, two vibration amplitude sequences are adaptively de-aggregated and then unified to jointly construct an amplitude interval, so that a matrix space subsequently constructed using the amplitude interval and the values ​​of the two vibration amplitude sequences can accurately describe the axis trajectory represented by the vibration time domain signal sequence, thereby subsequently obtaining more appropriate axis trajectory features, thereby improving the accuracy of axis trajectory recognition of the rotor of the rotating machinery.

[0111] To further optimize the above embodiment, step S104 of the method for identifying the axis trajectory of the rotor of a rotating machine in the embodiment of the present invention specifically includes:

[0112] Step S1041: traverse and extract a sub-matrix space from the matrix space according to a matrix operator;

[0113] For example, according to the matrix operator, a sub-matrix space of 3 rows × 3 columns is extracted from a spatial matrix of 100 rows × 100 columns to obtain multiple sub-matrix spaces of 3 rows × 3 columns.

[0114] Step S1042: Calculate the gradient of each sub-matrix space;

[0115] See also Figure 4 For the matrix space, take any 3-row × 3-column submatrix U, use the value at its center as a reference, and calculate the vectors from the 8 peripheral points to this point to obtain 8 vectors.

[0116] Each vector may have three situations: if the values ​​are equal, that is, both 0 or both 1, it is a 0 vector; the center point is 1, the outer point is 0, the vector points from the center point to the outer point, and the modulus is 1; the center point is 0, the outer point is 1, the vector points from the outer point to the center point, and the modulus is 1.

[0117] See also Figure 5 , which is a schematic diagram of the seventeen directions of the matrix space disclosed in an embodiment of the present invention. It is understandable that in an embodiment of the present invention, the above eight vectors will be added and subtracted to form a vector to represent the gradient direction of the submatrix. There are seventeen possible gradient directions of the submatrix: 0 vector (no direction), up, down, left, and right (a total of 4), the directions of the angle bisectors of the above four directions and adjacent directions (a total of 4), and the directions of the angle bisectors of the above eight directions (except the 0 vector) and adjacent directions (a total of 8).

[0118] Step S1043 : According to the gradient of each sub-matrix space, count the number of gradients of the matrix space in seventeen directions to obtain a seventeen-dimensional space vector.

[0119] See also Figure 6 , is a schematic diagram of a 17-dimensional vector in a matrix space. It can be understood that the value in each row represents the number of gradients in a direction of the matrix space.

[0120] In an embodiment of the present invention, a submatrix space is extracted from a matrix space based on a matrix operator, the gradient of each submatrix space is calculated, and based on the gradient of each submatrix space, the number of gradients of the entire matrix space in seventeen directions is counted to obtain a seventeen-dimensional space vector. In this way, two vibration time-domain signal sequences are processed into seventeen numerical values, which are used to represent the axis center trajectory of the rotating machinery rotor. This allows for rapid identification of the axis center trajectory of the rotating machinery rotor, thereby improving identification speed.

[0121] To further optimize the above embodiment, step S106 of the method for identifying the axis trajectory of the rotor of a rotating machine in the embodiment of the present invention specifically includes:

[0122] Step S1061: performing batch normalization processing on the labeled axis trajectory space vector dataset to obtain a normalized axis trajectory space vector dataset;

[0123] It should be noted that the batch normalization processing of the labeled axis trajectory space vector dataset includes: performing batch normalization processing in the following manner:

[0124] (ab) / c;

[0125] Wherein, a is any axis trajectory space vector in the labeled axis trajectory space vector dataset, b is the mean of the labeled axis trajectory space vector dataset, and c is the standard deviation of the labeled axis trajectory space vector dataset.

[0126] Step S1062: training an RBF support vector machine model using the normalized axis trajectory space vector dataset according to a preset regularization coefficient and parameter shrinking strategy;

[0127] It should be noted that there is no specific limit on the preset regularization coefficient. Optionally, the regularization coefficient of the RBF support vector machine model can be set to 14.5, which will result in better model training. The shrinking strategy is mainly used to accelerate the training process.

[0128] Step S1063: Use the K-fold cross-validation method to evaluate the generalization ability of the RBF support vector machine model. When the prediction accuracy of the RBF support vector machine model is greater than a preset value, the training of the RBF support vector machine model is terminated to obtain a trained axis trajectory recognition model.

[0129] In an embodiment of the present invention, batch normalization is performed on the input data, the regularization coefficient of the RBF support vector machine model is set to 14.5, a shrinking strategy is adopted in the training process, and the K-fold cross-validation method is used to consider the generalization ability of the RBF support vector machine model. The axis trajectory recognition model trained in this way has high recognition accuracy and fast recognition speed, which can further improve the accuracy and speed of axis trajectory recognition of rotating machinery rotors.

[0130] To further optimize the above embodiment, step S1063 of the method for identifying the axis trajectory of the rotor of a rotating machine in the embodiment of the present invention specifically includes:

[0131] Step S10631: Divide the labeled axis trajectory space vector dataset into K parts;

[0132] It should be noted that the value of K can be 5 or 6, etc., and there is no specific limitation on the value of K. Optionally, the algorithm performs better when K is 5.

[0133] Step S10632: Select K-1 copies as training sets each time, and the remaining copy as test set;

[0134] Step S10633: using a K-fold cross-validation method to evaluate the generalization ability of the axis trajectory recognition model;

[0135] Step S10634: When the average prediction accuracy of the test set is greater than 90%, it is determined that the generalization ability of the axis trajectory recognition model meets the standard, and the training of the RBF support vector machine model is terminated.

[0136] See also Figure 7 , is the confusion matrix of the prediction results in the embodiment of the present invention. Labels 0, 1, and 2 are divergence, inner and outer ring ellipse, and ellipse, respectively. Among them, the horizontal axis represents the predicted result of the axis trajectory shape, and the vertical axis represents the actual result of the axis trajectory shape. For example: (0,0) represents the data of the axis trajectory shape that is actually divergent, and the axis trajectory is predicted to be a divergent shape. The number 382.00 represents the number of times the prediction result is this. (0,1) represents the data of the axis trajectory shape that is actually divergent, and the axis trajectory is predicted to be an inner and outer ring ellipse shape. The number 4.00 represents the number of times the prediction result is this.

[0137] In the embodiment of the present invention, the K-fold cross-validation method is used to consider the generalization ability of the axis trajectory recognition model, thereby ensuring the recognition accuracy of the axis trajectory recognition model, so that the axis trajectory recognition model can be used to accurately identify the axis trajectory of the rotor of the rotating machinery in the subsequent use.

[0138] See also Figure 8 , a flowchart of another method for identifying the axis trajectory of a rotor of a rotating machine disclosed in an embodiment of the present invention, the method comprising:

[0139] Step S801: using a vibration sensor to collect two vibration time domain signal sequences with a 90° phase difference at the same measuring point and within the same period of time when the pump rotor of the slave unit is running, and obtaining two vibration amplitude sequences after preprocessing;

[0140] Step S802: construct a matrix space of 100 rows × 100 columns to describe the axis trajectory represented by the vibration time domain signal sequence;

[0141] Step S803: extract features of the axis trajectory using a matrix operator to obtain a 17-dimensional space vector;

[0142] Step S804: Repeat steps S801-S803 to collect data under different fault modes of the pump to obtain a labeled axis trajectory space vector data set;

[0143] Step S805: Use the data set to train the RBF support vector machine model, perform batch normalization processing, set the model regularization coefficient to 14.5, adopt the shrinking strategy, and divide the original data set, and use K-fold cross validation;

[0144] Step S806: Use the RBF support vector based on the minimum structural risk obtained through training to infer and identify the axis trajectory space vector of the unknown label.

[0145] An embodiment of the present invention can synchronously convert two sections of rotating machinery rotor vibration time domain signals with a phase difference of 90° into a matrix space model based on the collection of a series of fault vibration signals, use matrix operator feature extraction technology to extract axis trajectory features of the rotating machinery rotor vibration signal, and use RBF support vector machine to perform classification learning on known, labeled axis trajectory vibration signal features. The optimized classifier finally obtained can identify the axis trajectory features of unlabeled rotating machinery rotor vibration signals. This feature can assist in judging whether the rotating machinery rotor contains faults such as misalignment, imbalance, and bending.

[0146] Corresponding to the above method embodiment, the present invention further discloses a device for identifying the axis trajectory of a rotor of a rotating machine.

[0147] See also Figure 9 , a schematic diagram of a device for identifying the axis trajectory of a rotor of a rotating machine disclosed in an embodiment of the present invention, the device comprising:

[0148] The acquisition unit 901 is configured to use a vibration sensor to acquire two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same time period when the rotor of the rotating machinery is running;

[0149] A preprocessing unit 902 is used to preprocess the two vibration time domain signal sequences to obtain two vibration amplitude sequences;

[0150] A matrix construction unit 903 is configured to construct a matrix space using the two vibration amplitude sequences, wherein the matrix space is used to describe the axis trajectory represented by the vibration time domain signal sequence;

[0151] A feature extraction unit 904 is configured to perform feature extraction on the matrix space according to a matrix operator to obtain a 17-dimensional space vector;

[0152] The data set construction unit 905 is configured to construct a labeled axis trajectory space vector data set using the seventeen-dimensional space vectors obtained when the rotating machinery rotor operates in different fault modes;

[0153] A training unit 906 is configured to train a radial basis function (RBF) support vector machine (SVM) model using the labeled axis trajectory spatial vector dataset to obtain a trained axis trajectory recognition model.

[0154] The identification unit 907 is configured to identify the axis trajectory space vector of the unknown label using the axis trajectory identification model.

[0155] An embodiment of the present invention discloses an axis trajectory identification device for a rotating machinery rotor. The device uses a vibration sensor to collect two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same time period when the rotating machinery rotor is running; the two vibration time domain signal sequences are preprocessed to obtain two vibration amplitude sequences; a matrix space is constructed using the two vibration amplitude sequences; features are extracted from the matrix space according to a matrix operator to obtain a 17-dimensional space vector; a labeled axis trajectory space vector data set is constructed using the 17-dimensional space vectors obtained when the rotating machinery rotor is running in different fault modes; a radial basis function (RBF) support vector machine model is trained using the labeled axis trajectory space vector data set to obtain a trained axis trajectory identification model; and the axis trajectory identification model is used to identify axis trajectory space vectors with unknown labels. It can be seen that the use of matrix operators to extract features in the embodiment of the present invention can extract more appropriate axis trajectory features. By using the RBF support vector machine model to perform classification learning on a known, labeled axis trajectory space vector data set, a more generalized recognition model with better classification effect can be obtained. In this way, the use of matrix operator feature extraction technology combined with the RBF support vector machine model to jointly identify the axis trajectory of the rotating machinery rotor can improve the accuracy of the axis trajectory identification of the rotating machinery rotor, thereby improving the accuracy of the rotating machinery rotor fault diagnosis.

[0156] To further optimize the above embodiment, the number of the vibration sensors is two. The device for identifying the axis trajectory of the rotor of a rotating machine according to the embodiment of the present invention further includes:

[0157] A position setting unit is used to set the two vibration sensors at the same axial position, and the radial position difference between the two vibration sensors is 90°; wherein the connection between the vibration sensor and the measuring point is adhesive or rigid mechanical fastening.

[0158] To further optimize the above embodiment, the preprocessing unit 902 in the rotating machinery rotor axis trajectory identification device in the embodiment of the present invention is specifically used to preprocess the two vibration time domain signals through filtering, time domain averaging, detrending, and normalization to obtain two vibration amplitude sequences.

[0159] To further optimize the above embodiment, the matrix construction unit 903 in the apparatus for identifying the axis trajectory of a rotor of a rotating machinery in the embodiment of the present invention specifically includes:

[0160] an amplitude determination unit, configured to determine a minimum amplitude and a maximum amplitude according to the two vibration amplitude sequences;

[0161] an amplitude interval construction unit, configured to construct an amplitude interval according to the minimum amplitude value and the maximum amplitude value;

[0162] a row and column determining unit, configured to use any one of the two vibration amplitude sequences as the coordinates of the rows in the matrix, and use the other one of the two vibration amplitude sequences as the coordinates of the columns in the matrix;

[0163] A coordinate determining unit, configured to determine a valid amplitude sequence coordinate using the amplitude interval and the two vibration amplitude sequences;

[0164] The value setting unit is used to set the position value corresponding to the valid amplitude sequence coordinate in the matrix to 1, and the remaining position values ​​in the matrix except the position value of 1 to 0, so as to obtain a matrix space with a value of 0 or 1.

[0165] To further optimize the above embodiment, the feature extraction unit 904 in the apparatus for identifying the axis trajectory of a rotor of a rotating machinery in the embodiment of the present invention specifically includes:

[0166] an extraction subunit, configured to traverse and extract a submatrix space from the matrix space according to a matrix operator;

[0167] A calculation unit, configured to calculate the gradient of each sub-matrix space;

[0168] A statistical unit is used to count the number of gradients of the matrix space in seventeen directions according to the gradient of each sub-matrix space to obtain a seventeen-dimensional space vector.

[0169] To further optimize the above embodiment, in the embodiment of the present invention, the device for identifying the axis trajectory of the rotor of a rotating machinery, the fault modes include: rotor misalignment with the fault characteristic of an outer ring ellipse, rotor imbalance with the fault characteristic of a stable ellipse, rotor friction with the fault characteristic of a divergent rotor, and oil film oscillation with the fault characteristic of a petal shape.

[0170] To further optimize the above embodiment, the training unit 906 in the apparatus for identifying the axis trajectory of a rotor of a rotating machinery according to the embodiment of the present invention specifically includes:

[0171] a normalization processing unit, configured to perform batch normalization processing on the labeled axis trajectory space vector dataset to obtain a normalized axis trajectory space vector dataset;

[0172] A training subunit is used to train an RBF support vector machine model using the normalized axis trajectory space vector dataset according to a preset regularization coefficient and parameter shrinking strategy;

[0173] The verification unit is used to consider the generalization ability of the RBF support vector machine model using a K-fold cross-validation method, and to end the training of the RBF support vector machine model when the prediction accuracy of the RBF support vector machine model is greater than a preset value to obtain a trained axis trajectory recognition model.

[0174] To further optimize the above embodiment, the normalization processing unit in the device for identifying the axis trajectory of a rotating machinery rotor in the embodiment of the present invention is specifically configured to perform batch normalization processing in the following manner:

[0175] (ab) / c;

[0176] Wherein, a is any axis trajectory space vector in the labeled axis trajectory space vector dataset, b is the mean of the labeled axis trajectory space vector dataset, and c is the standard deviation of the labeled axis trajectory space vector dataset.

[0177] To further optimize the above embodiment, the verification unit in the device for identifying the axis trajectory of the rotor of a rotating machinery in the embodiment of the present invention specifically includes:

[0178] a division unit, configured to divide the labeled axis trajectory space vector dataset into K parts;

[0179] The selection unit is used to select K-1 copies as training sets each time and the remaining 1 copy as test set;

[0180] A validation subunit, configured to evaluate the generalization capability of the RBF support vector machine model using a K-fold cross-validation method;

[0181] The standard-reaching determination unit is used to determine that the generalization ability of the axis trajectory recognition model reaches the standard when the average prediction accuracy of the test set is greater than 90%, and to end the training of the RBF support vector machine model.

[0182] It should be noted that the specific working principles of each component in the device embodiment can be found in the corresponding part of the method embodiment and will not be repeated here.

[0183] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0184] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0185] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying the axis trajectory of a rotating machinery rotor, characterized in that: The method comprises: Using a vibration sensor to collect two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same period of time when the rotating machinery rotor is running; Preprocessing the two vibration time domain signal sequences to obtain two vibration amplitude sequences; Constructing a matrix space using the two vibration amplitude sequences, wherein the matrix space is used to describe the axis trajectory represented by the vibration time domain signal sequence; Performing feature extraction on the matrix space according to a matrix operator to obtain a seventeen-dimensional space vector, where the seventeen-dimensional space vector refers to the number of gradients corresponding to the matrix space in seventeen directions; Using the seventeen-dimensional space vectors obtained when the rotating machinery rotor operates in different fault modes, a labeled axis trajectory space vector dataset is constructed; Using the labeled axis trajectory space vector data set to train a radial basis function (RBF) support vector machine model to obtain a trained axis trajectory recognition model; Identify the axis trajectory space vector of the unknown label using the axis trajectory recognition model; The constructing of a matrix space by using the two vibration amplitude sequences includes: Determining a minimum amplitude and a maximum amplitude according to the two vibration amplitude sequences; Constructing an amplitude interval according to the minimum amplitude value and the maximum amplitude value; Using any one of the two vibration amplitude sequences as the coordinates of the rows in the matrix, and using the other one of the two vibration amplitude sequences as the coordinates of the columns in the matrix; Determining effective amplitude sequence coordinates using the amplitude interval and the two vibration amplitude sequences; The position values ​​corresponding to the valid amplitude sequence coordinates in the matrix are set to 1, and the values ​​of the other positions in the matrix except for the position with the value 1 are set to 0, so as to obtain a matrix space with values ​​of 0 or 1.

2. The method according to claim 1, characterized in that The method of using the labeled axis trajectory space vector data set to train a radial basis function (RBF) support vector machine model to obtain a trained axis trajectory recognition model includes: performing batch normalization processing on the labeled axis trajectory space vector dataset to obtain a normalized axis trajectory space vector dataset; According to a preset regularization coefficient and parameter shrinking strategy, the RBF support vector machine model is trained using the normalized axis trajectory space vector dataset; The K-fold cross-validation method is used to evaluate the generalization ability of the RBF support vector machine model. When the prediction accuracy of the RBF support vector machine model is greater than a preset value, the training of the RBF support vector machine model is terminated to obtain a trained axis trajectory recognition model.

3. The method according to claim 1, characterized in that There are two vibration sensors, and the method further includes: The two vibration sensors are arranged at the same axial position, and the radial position difference between the two vibration sensors is 90°; wherein, the connection between the vibration sensor and the measuring point is by gluing or rigid mechanical fastening.

4. The method according to claim 1, wherein The preprocessing of the two vibration time domain signal sequences to obtain two vibration amplitude sequences includes: The two vibration time domain signals are preprocessed by filtering, time domain averaging, detrending and normalizing to obtain two vibration amplitude sequences.

5. The method according to claim 1, characterized in that The feature extraction of the matrix space is performed according to the matrix operator to obtain a seventeen-dimensional space vector, including: Traversing and extracting a submatrix space from the matrix space according to a matrix operator; Calculating the gradient of each sub-matrix space; According to the gradient of each of the sub-matrix spaces, the number of gradients of the matrix space in seventeen directions is counted to obtain a seventeen-dimensional space vector.

6. The method according to claim 1, characterized in that The fault modes include: rotor misalignment with the fault characteristic of an outer ring ellipse, rotor imbalance with the fault characteristic of a stable ellipse, rotor collision with the fault characteristic of a divergent rotor, and oil film oscillation with the fault characteristic of a petal shape.

7. The method according to claim 2, characterized in that The batch normalization processing of the labeled axis trajectory space vector dataset includes: Batch normalization is performed as follows: (ab) / c; Wherein, a is any axis trajectory space vector in the labeled axis trajectory space vector dataset, b is the mean of the labeled axis trajectory space vector dataset, and c is the standard deviation of the labeled axis trajectory space vector dataset.

8. The method according to claim 2, characterized in that The K-fold cross-validation method is used to consider the generalization ability of the RBF support vector machine model, and the training of the RBF support vector machine model is terminated when the prediction accuracy of the RBF support vector machine model is greater than a preset value, including: Dividing the labeled axis trajectory space vector dataset into K parts; Each time, K-1 copies are selected as training sets, and the remaining 1 copy is used as the test set; The K-fold cross validation method was used to consider the generalization ability of the RBF support vector machine model; When the average prediction accuracy of the test set is greater than 90%, it is determined that the generalization ability of the axis trajectory recognition model meets the standard, and the training of the RBF support vector machine model is terminated.

9. A device for identifying the axis trajectory of a rotating machine rotor, characterized in that: The device comprises: An acquisition unit is used to acquire two vibration time domain signal sequences with a phase difference of 90° at the same measuring point and within the same period of time using a vibration sensor when the rotor of the rotating machinery is running; A preprocessing unit, configured to preprocess the two vibration time domain signal sequences to obtain two vibration amplitude sequences; a matrix construction unit, configured to construct a matrix space using the two vibration amplitude sequences, wherein the matrix space is used to describe the axis trajectory represented by the vibration time domain signal sequence; a feature extraction unit, configured to perform feature extraction on the matrix space according to a matrix operator to obtain a seventeen-dimensional space vector, wherein the seventeen-dimensional space vector refers to the number of gradients corresponding to the matrix space in seventeen directions; A data set construction unit, configured to construct a labeled axis trajectory space vector data set using the seventeen-dimensional space vectors obtained when the rotating machinery rotor operates in different fault modes; A training unit, configured to train a radial basis function (RBF) support vector machine (SVM) model using the labeled axis trajectory spatial vector data set to obtain a trained axis trajectory recognition model; an identification unit, configured to identify the axis trajectory space vector of an unknown label using the axis trajectory identification model; The matrix construction unit specifically includes: an amplitude determination unit, configured to determine a minimum amplitude and a maximum amplitude according to the two vibration amplitude sequences; an amplitude interval construction unit, configured to construct an amplitude interval according to the minimum amplitude value and the maximum amplitude value; a row and column determining unit, configured to use any one of the two vibration amplitude sequences as the coordinates of the rows in the matrix, and use the other one of the two vibration amplitude sequences as the coordinates of the columns in the matrix; A coordinate determining unit, configured to determine a valid amplitude sequence coordinate using the amplitude interval and the two vibration amplitude sequences; The value setting unit is used to set the position value corresponding to the valid amplitude sequence coordinate in the matrix to 1, and the remaining position values ​​in the matrix except the position value of 1 to 0, so as to obtain a matrix space with a value of 0 or 1.

Images