Rotating machine image fault diagnosis method based on interval model

Abstract

The invention discloses a rotating machine image fault diagnosis method based on an interval model, and the method comprises the steps: carrying out the resampling according to an original displacement signal, obtaining a plurality of fault samples, and drawing a time domain oscillogram; obtaining a principal component matrix of the fault sample according to histogram of oriented gradients feature extraction and principal component analysis; obtaining an interval feature vector of each fault mode according to the extreme value range of each fault mode sample set of the rotating machine; according to histogram of oriented gradient feature extraction and principal component analysis, obtaining a principal component vector of a to-be-detected sample of the rotating machine; obtaining a matching degree matrix of the to-be-detected sample based on a position relationship between the interval feature vector and the principal component vector, and calculating a matching degree vector; and carrying out weight optimization on the matching degree matrix which does not meet the difference threshold condition, and calculating a final fault diagnosis result according to the weights of the feature vectors in different intervals. According to the method, the credibility of the rotating machine image fault diagnosis result under the dispersive condition can be improved.

Description

Technical Field

[0001] The invention belongs to the field of mechanical engineering, and in particular relates to a rotating machinery image fault diagnosis method based on an interval model. Background Art

[0002] With the rapid development of science and technology and industrial technology, mechanical equipment is becoming larger, more integrated and more automated. Rotating machinery is the most widely used mechanical equipment at present. Common rotating machinery includes engines, generators, blowers, compressors and steam turbines. Rotating machinery usually includes components such as rotors, bearings and gears. As a key component of rotating machinery, the rotor plays a key role in the efficient and stable operation of rotating machinery. However. The rotor works in extreme environments for a long time, such as high speed, high load, high temperature and high pressure, which often causes various forms of failure, such as rotor misalignment, dynamic and static friction, rotor imbalance, etc. Once the rotor fails, the shutdown will cause production interruption at the least, and the company will face huge economic losses, or even endanger life.

[0003] In the process of fault diagnosis of rotating machinery, the vibration signal of rotating machinery can be used to construct a vibration signal image through a variety of extraction methods, such as one-dimensional time domain image, two-dimensional time-frequency image, etc., and the state of rotating machinery during operation can be monitored through the vibration signal image. However, due to problems such as measurement errors and noise, the fault diagnosis of vibration signal images faces the problems of complexity and dispersion, making it very difficult to directly diagnose the fault of rotating machinery from the vibration signal image, and the credibility of the fault diagnosis results is poor. Summary of the invention

[0004] The embodiment of the present invention provides a rotating machinery image fault diagnosis method based on an interval model to improve the credibility of the fault diagnosis result.

[0005] The embodiment of the present invention provides a rotating machinery image fault diagnosis method based on an interval model, comprising:

[0006] Step 1: Under different fault modes, the original displacement signal of the rotating machinery is intercepted by using a resampling method to obtain multiple fault samples, and a time domain waveform corresponding to each fault sample is formed;

[0007] Step 2: extract the directional gradient histogram features of each of the time domain waveforms to obtain the corresponding directional gradient histogram feature vectors, all the feature vectors form the fault sample feature matrix, and use the principal component analysis method to reduce the dimension of the fault sample feature matrix to obtain the principal component matrix of the fault sample feature matrix;

[0008] Step 3: Obtain a sample set under each of the fault modes according to the principal component matrix, establish an interval model for each of the fault modes within an extreme value range, and all of the interval models respectively form interval feature vectors corresponding to the fault modes;

[0009] Step 4: obtaining the directional gradient histogram feature vector of the sample to be tested of the rotating machinery, and reducing the dimension of the directional gradient histogram feature vector of the sample to be tested by using the principal component analysis method to obtain the principal component vector of the sample to be tested;

[0010] Step 5: According to the positional relationship between the interval feature vector of each fault mode and the corresponding element in the principal component vector, a matching degree matrix between the sample to be tested and the fault sample is obtained, and a matching degree vector is calculated for each element of the interval feature vector using the same weight;

[0011] Step six: Calculate the difference between the first matching degree and the second matching degree in the matching degree vector. For the samples to be tested that meet the difference threshold, select the fault mode corresponding to the first matching degree to obtain a fault diagnosis conclusion; for the samples to be tested that do not meet the difference threshold, reallocate the weights according to the discrimination degree of each element of the interval feature vector in the matching degree matrix under each fault mode and calculate the matching degree vector, select the fault mode corresponding to the calculated first matching degree to obtain the fault diagnosis conclusion.

[0012] In some possible embodiments, an interval model is used to perform interval estimation on the sample set under each of the fault modes in the principal component matrix to form the interval feature vector corresponding to the fault mode.

[0013] The rotating machinery image fault diagnosis method based on interval model provided by the embodiment of the present invention has at least the following advantages:

[0014] (1) The method in the embodiment of the present invention fully considers the problem that data is dispersed due to various uncertain factors in actual engineering, and the reliability of the fault diagnosis result is relatively high.

[0015] (2) The statistical principles are used to perform interval estimation on sample sets of different fault modes, fully extracting the effective information of the sample points, which can effectively deal with the dispersion problem in fault diagnosis.

[0016] (3) Compared with the traditional matching degree calculation method which mostly selects the maximum matching degree as the diagnosis result, the embodiment of the present invention fully considers the situation where the maximum matching degree and the second matching degree are close, and uses the method of calculating the distinguishing degree of each feature in the matching degree matrix under different fault modes to reallocate the weights of the features. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of a rotating machinery image fault diagnosis method based on an interval model in an embodiment of the present invention;

[0018] Figure 2 is a simplified flow chart of a rotating machinery image fault diagnosis method based on an interval model in an embodiment of the present invention;

[0019] Figure 3 Schematic diagram of a rotating machine in an embodiment of the present invention.

[0020] Description of reference numerals:

[0021] 10- Turntable;

[0022] 21- first support;

[0023] 22- second support;

[0024] 31- first coupling;

[0025] 32- second coupling;

[0026] 40-Rotation shaft. DETAILED DESCRIPTION

[0027] In order to make the above-mentioned purposes, features and advantages of the embodiments of the present invention more understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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 work belong to the scope of protection of the present invention.

[0028] refer to Figure 1 and Figure 2 The embodiment of the present invention provides a rotating machinery image fault diagnosis method based on an interval model, comprising the following steps:

[0029] Step 1: Under different fault modes, the original displacement signal of the rotating machinery is intercepted by using the resampling method to obtain multiple fault samples, and a time domain waveform corresponding to each fault sample is formed.

[0030] In the embodiment of the present invention, there are N fault modes in the rotating machinery. The original displacement signal of each fault mode is intercepted by a resampling method to obtain M fault samples of the fault mode. The M fault samples of each fault mode are plotted into a time domain waveform diagram with an image size of W×H by an image grayscale processing method.

[0031] In one possible example, a rotating machine such as Figure 3As shown. The rotating machine comprises a turntable 10, a rotating shaft 40, a first coupling 31, a second coupling 32, a first support 21 and a second support 22. There are two rotating shafts 40, which are respectively located on both sides of the turntable 10, and the turntable 10 is connected to one of the rotating shafts 40 through the first coupling 31, and the turntable 10 is connected to the other rotating shaft 40 through the second coupling 32. The first coupling 31 and the second coupling 32 are also connected to the first support 21 and the second support 22 respectively, and are supported by the first support 21 and the second support 22.

[0032] like Figure 3 As shown, a rotating shaft 40 is respectively disposed on the left and right sides of the rotating disk 10, the rotating disk 10 and the rotating shaft 40 on the left side thereof are connected via a first coupling 31, and the rotating disk 10 and the rotating shaft 40 on the right side thereof are connected via a second coupling 32. The end of the rotating shaft 40 on the left side of the rotating disk 10 away from the rotating disk 10 is connected to the first support 21, and the end of the rotating shaft 40 on the right side of the rotating disk 10 away from the rotating disk 10 is connected to the second support 22.

[0033] A fault simulation experiment was carried out on the above-mentioned rotating machinery, and a total of four fault modes of rotating machinery were simulated, namely rotor imbalance, rotor misalignment, loose support, and dynamic-static friction. The displacement response (i.e., original displacement signal) of the measuring point corresponding to the fault mode was obtained, with a sampling time of 1 min and a sampling frequency of 12.8 kHz.

[0034] The original displacement signals of the four fault modes are intercepted by the resampling method to obtain 13 fault samples in each fault mode, and the 13 fault samples in each fault mode are plotted into corresponding time domain waveforms by the image grayscale processing method. The image size of the time domain waveform is 100×200. It can be understood that each fault sample corresponds to a time domain waveform. Figure 1 There are 4×13 in total.

[0035] Step 2: Extract the directional gradient histogram features of each time domain waveform to obtain the corresponding directional gradient histogram feature vector. All feature vectors form a fault sample feature matrix, and use the principal component analysis method to reduce the dimension of the fault sample feature matrix to obtain the principal component matrix of the fault sample feature matrix.

[0036] In the embodiment of the present invention, there are N kinds of fault modes in the rotating machinery, and each fault mode has M fault samples obtained according to the resampling method. The fault samples corresponding to each fault mode are plotted into corresponding fault sample time domain waveforms. For each fault sample time domain waveform, directional gradient histogram feature extraction is performed, and the specific steps are as follows:

[0037] A c×c window in the pixel matrix of the fault sample time domain waveform is called a cell. The gray value of any pixel in the cell is I(x, y). Its horizontal gradient I is calculated. x (x,y) and vertical gradient I y (x,y) are:

[0038] I x (x,y)=I(x+1,y)-I(x-1,y);

[0039] I y (x,y)=I(x,y+1)-I(x,y-1);

[0040] The gradient magnitude m(x,y) and gradient direction of the pixel point (x,y) They are:

[0041]

[0042] The gradient direction is equally divided into 9 signed directions, each with a size of 20°, and the directional gradient histogram of the current cell is obtained by weighting the current pixel gradient direction and amplitude in the cell.

[0043] Every 4 adjacent cells form a block, and a 36-dimensional directional gradient histogram feature vector is obtained. The directional gradient histogram feature vector of the current block is normalized.

[0044] Then slide the block on the time domain waveform in the horizontal and vertical directions with a step size of c, repeat the above steps to obtain the directional gradient histogram feature vector of the sliding block, and finally concatenate the directional gradient histogram feature vectors of all blocks as the final directional gradient histogram feature vector.

[0045] For the convenience of description, the failure mode F j The i-th element of the eigenvector of the histogram of oriented gradients is called X i,j Therefore, the failure mode F j The directional gradient histogram feature vector X j It is expressed as:

[0046] X j =(X 1,j ,X 2,j ,…,X p,j ) T ,j=1,2,…,N;

[0047] in, Indicates the dimension size of the final directional gradient histogram feature vector, and [*] is the rounding down function.

[0048] For each failure mode F j (j=1,2,…,N), after M times of resampling, a total of M fault samples are obtained, and M time domain waveforms are drawn. The total number of fault samples is n=N×M, the dimension of the directional gradient histogram feature vector is p-dimensional, and the feature vectors corresponding to all fault samples form a fault sample feature matrix, which can be expressed as:

[0049]

[0050] The principal component analysis method is used to reduce the dimension of the above fault sample feature matrix.

[0051] Through data standardization, the fault sample feature matrix is ​​de-averaged with a center of 0, eliminating the influence of the differences in dimensions and value ranges between the feature vectors. The formula for data standardization is:

[0052]

[0053] in, is the average value of each column feature, var(x j ) is the characteristic standard deviation of each column.

[0054] The new matrix after normalization is The correlation coefficient matrix, also known as the covariance matrix, describes the correlation between two or more variables. The correlation coefficient matrix R can be expressed as:

[0055]

[0056] in, and are the matrix elements after normalization.

[0057] Calculate the eigenvalue λ of the correlation coefficient matrix R i And the corresponding eigenvector q i After the principal component analysis method is used for dimensionality reduction, the first t principal components are selected according to the cumulative principal component percentage. The expression of the contribution rate of the jth eigenvalue can be expressed as:

[0058]

[0059] The first t eigenvectors are selected in order to form the matrix Q = [q 1 ,q 2 ,…,q t ], the final principal component matrix expression is:

[0060] Z=[Z 1 ,Z 2 ,…,Zt ] = XQ;

[0061] Among them, Z 1 is the first principal component, Z 2 is the second principal component, Z t is the tth principal component.

[0062] In a possible example, 13 fault samples are collected for each of the four fault modes of the rotating machinery, 52 time domain waveforms of size 100×200 are drawn, and directional gradient histogram features are extracted for each time domain waveform. The specific steps are as follows:

[0063] Take the cell size of the pixel matrix of the fault sample time domain waveform as 32×32, and the gray value of any pixel in the cell is I(x,y), and its horizontal gradient I is obtained. x (x,y) and vertical gradient I y (x,y), through I x (x,y) and I y (x,y) Calculate the gradient magnitude m(x,y) and gradient direction of the pixel point (x,y)

[0064] The gradient direction is equally divided into 9 signed directions, each with a size of 20°, and the directional gradient histogram of the current cell is obtained by weighting the current pixel gradient direction and amplitude in the cell.

[0065] Every 4 adjacent cells form a block, and a 36-dimensional directional gradient histogram feature vector is obtained, and the directional gradient histogram feature vector of the block is normalized.

[0066] Then slide the block on the image horizontally and vertically with a step size of 32, repeat the above steps to obtain the directional gradient histogram feature vector of the sliding block, and finally concatenate the directional gradient histogram feature vectors of all blocks as the final directional gradient histogram feature vector, and the final directional gradient histogram feature vector dimension is 360 dimensions.

[0067] After the above steps, for the failure mode F j , the eigenvector of the directional gradient histogram can be specifically expressed as X j =(X 1,j ,…,X 360,j ) T (j=1,2,3,4), j is the fault mode, and the characteristic matrix obtained after series connection is:

[0068]

[0069] The fault sample feature matrix is ​​de-meaned with 0 as the center. After standardization, the new matrix can be expressed as Calculate the correlation coefficient matrix R = (X * ) T X * .

[0070] Calculate the eigenvalue λ of the correlation coefficient matrix R i And the corresponding eigenvector q i After the principal component analysis method is used for dimensionality reduction, the first 10 principal components are selected according to the eigenvalue contribution rate, and the first 10 principal component eigenvectors are selected in order to form the matrix Q = [q 1 ,q 2 ,…,q 10 ], the expression of the 10th-order principal component matrix after final dimension reduction is:

[0071] Z=[Z 1 ,Z 2 ,…,Z 10 ] = XQ;

[0072] Step 3: Obtain the sample set under each fault mode according to the principal component matrix, establish the interval model of each fault mode within the extreme value range, and all interval models form the interval feature vectors of the corresponding fault mode.

[0073] In the embodiment of the present invention, the t-order principal component matrix Z of the feature matrix obtained in step 2 can be expressed as Z=[Z 1 ,Z 2 ,…,Z N ],Z j (j=1,2,…,N) represents the failure mode F j The set of all reduced-dimensional feature vectors in includes M feature vectors of resampled fault samples after reduced-dimensionality.

[0074] Using Z j The feature vectors of the M resampled fault samples after dimension reduction obtain the numerical distribution range of each feature component and are expressed in interval form. j The i-th principal component of is called the characteristic parameter Z i,j , the sample set of M resampled fault modes is:

[0075]

[0076] Among them, Λ i,j Represents the characteristic parameter Z i,j A collection of samples.

[0077] The characteristic parameter Z is adjusted within the extreme value range i,jPerform interval estimation and use the maximum value of the sample set as the characteristic parameter Z i,j The upper limit of the interval ), taking the minimum value of the sample set as the characteristic parameter Z i,j The lower limit of the interval ). Thus, the characteristic parameter Z is obtained i,j The number of intervals is:

[0078]

[0079] Wherein, the superscript I represents the characteristic parameter Z i,j is the number of intervals.

[0080] The failure mode F j The interval eigenvector of the characteristic parameter is recorded as It is expressed as:

[0081]

[0082] Among them, t is the number of principal component analyses selected, that is, the dimension of the eigenvector of the directional gradient histogram after dimensionality reduction.

[0083] In a possible example, the feature matrix Z after dimension reduction obtained in step 2 can be expressed as Z = [Z 1 ,Z 2 ,Z 3 ,Z 4 ], including 4 fault modes and 13 groups of resampled fault samples. Using the 13 resampled fault sample sets under each fault mode, the numerical distribution range of the characteristic parameters under the fault mode can be obtained. For fault mode F j The i-th principal component Z i , that is, the characteristic parameter Z i,j , the corresponding sample set is:

[0084]

[0085] Among them, Λ i,j Represents the characteristic parameter Z i,j A collection of samples.

[0086] The interval estimation of each characteristic parameter is performed through the extreme value method, and the maximum value of the sample set is used as the upper limit of the characteristic parameter (i.e., the upper limit of the interval), and the minimum value of the sample set is used as the lower limit of the characteristic parameter (i.e., the lower limit of the interval). Taking principal component 1, principal component 2, and principal component 3 as examples, the interval model information established by 13 fault samples is shown in Table 1:

[0087] Table 1 Interval model under extreme value method (taking principal components 1, 2.3 as an example)

[0088]

[0089]

[0090] The failure mode F j The interval feature vector of the sample set is recorded as It is expressed as:

[0091]

[0092] Step 4: Obtain the directional gradient histogram feature vector of the sample to be tested of the rotating machinery, and use the principal component analysis method to reduce the dimension of the directional gradient histogram feature vector of the sample to be tested to obtain the principal component vector of the sample to be tested.

[0093] In the embodiment of the present invention, for the sample to be tested U, the same directional gradient histogram feature vector extraction method as in step 2 is selected to obtain the directional gradient histogram feature vector X corresponding to the sample to be tested. U =(x 1,U ,x 2,U ,…,x p,U ) T , and perform the same processing as the principal component analysis of the fault samples in step 2.

[0094] Normalize using the mean and variance in step 2:

[0095]

[0096] in, is the column mean in step 2, var(x j ) is the column variance in step 2.

[0097] Using the matrix Q composed of the first t principal components in step 2 = [q 1 ,q 2 ,…,q t ]Get the t-order principal component vector Z of the eigenvector of the directional gradient histogram of the sample to be tested U :

[0098] Z U =[Z 1,U ,Z 2,U ,…,Z t,U ]=X U Q;

[0099] For example, for the sample U to be tested, the same method as in step 2 is selected to obtain its directional gradient histogram feature vector X U =(x 1,U ,x 2,U ,...,x 360,U ) TUsing the mean and variance of the fault sample feature matrix in step 2 for normalization, select the matrix Q composed of the first 10 principal component feature vectors of the fault feature matrix in step 2 to obtain the 10th-order principal component vector of the directional gradient histogram feature vector of the sample to be tested:

[0100] Z U =[Z 1,U ,Z 2,U ,…,Z 10,U ]=X U Q=[3.60,4.22,4.56,-12.96,3.26,2.30,5.83,-7.60,0.86,0.49];

[0101] Step 5: According to the positional relationship between the interval feature vector of each fault mode and the corresponding element in the principal component vector, the matching degree matrix between the sample to be tested and the fault sample is obtained, and the matching degree vector is calculated using the same weight for each element of the interval feature vector.

[0102] For failure mode F j (j=1,2,…,N), according to the interval feature vector of the fault sample in step 3 And the directional gradient histogram feature vector Z of the sample to be tested in step 4 U The t-order principal component vector, for the characteristic parameter Z i,j , define the detection mode U and the failure mode F j The interval matching degree of this characteristic parameter is:

[0103]

[0104] Among them, P ij Represents the characteristic parameter Z i The next waiting mode is the same as the fault mode F j The interval matching degree of Z i,U belong The number of intervals, i.e. Represents Z i,U Not The number of intervals, i.e. or

[0105] For each failure mode F j (j=1,2,…,N), using the above interval matching calculation method, we get the matching matrix P=P ij (i=1,2,…,t;j=1,2,…,N). The overall matching degree is calculated by the same weight:

[0106]

[0107] P(U∈Fj ) represents the relationship between the sample to be tested U and the fault mode F under t principal components. j The overall matching degree of all fault modes is combined to form the matching degree vector P of the sample to be tested U. U :

[0108] P U =P(U∈F j ),j=1,2,…,N;

[0109] For example, for failure mode F j (j=1,2,3,4), the first 10 principal components were selected as the principal component matrix.

[0110] According to the interval feature vector of the fault sample in step 3 And the directional gradient histogram feature vector Z of the sample to be tested in step 4 U The t-order principal component vector of the matching degree can be obtained by the above matching degree calculation method. ij (i=1,2,…,10;j=1,2,3,4), as shown in the following table:

[0111] Table 2 Matching matrix

[0112]

[0113] Calculate the overall matching degree with the same weights:

[0114]

[0115] P(U∈F j ) represents the overall matching degree between the sample to be tested U and the fault mode Fj under 10 principal components. The overall matching degrees of all fault modes are combined together to form the matching degree vector P of the sample to be tested U. U :

[0116] P U =[0.4,0.6,0.6,0.5];

[0117] Step six: Calculate the difference between the first matching degree and the second matching degree in the matching degree vector. For the samples to be tested that meet the difference threshold, select the fault mode corresponding to the first matching degree to obtain the fault diagnosis conclusion; for the samples to be tested that do not meet the difference threshold, redistribute the weights according to the discrimination of each element of the interval feature vector in the matching degree matrix under each fault mode and calculate the matching degree vector, and select the fault mode corresponding to the calculated first matching degree to obtain the fault diagnosis conclusion.

[0118] According to the matching degree vector P of the sample U to be tested in step 5 U , sort them from large to small as P′U :

[0119] P′ U = sort(P U );

[0120] The first matching degree and the second matching degree are P' U (1) and P′ U (2), set the difference threshold to 0.3, that is, when P′ U (1) -P′ U (2) ≥ 0.3, the fault diagnosis result is considered reasonable, and the maximum matching degree P′ is U (1) The corresponding fault mode obtains the fault diagnosis result; when P′ U (1)-P′ U When (2) is less than 0.3, it is considered that the fault diagnosis result has decision-making difficulties and weight optimization is required. The specific steps are as follows:

[0121] Filter the matching matrix P by column, remove the columns that are all 0 or 1 (corresponding principal components cannot be used as features to distinguish fault modes) to form an optimized matching matrix t * is the number of principal components after screening.

[0122] In optimizing the matching matrix P * In the definition, the discrimination degree is defined as the number of elements corresponding to the feature vector of the directional gradient histogram to be detected in each column that do not belong to the fault mode F j (j=1,2,…,N) corresponds to the number of intervals (i.e., the number of 0s in each column). The discrimination D represents the importance of the element in distinguishing different failure modes and can be expressed as:

[0123]

[0124] Use the discrimination to redistribute the weights and get the optimized matching degree:

[0125] P * (U∈F j )=P * D i T ,j=1,2,3,4;

[0126] P * (U∈F j ) characterizes the relationship between the sample to be tested U and the failure mode F j The optimized matching degree of all fault modes is combined to form the optimized matching degree vector of the sample to be tested U.

[0127]

[0128] Selecting the optimized matching vector The first match in The corresponding fault mode is taken as the optimized fault diagnosis result.

[0129] In a possible example, in step 5, the matching degree vector P of the sample to be tested U is U , comparing the difference between the first matching degree 0.6 and the second matching degree 0.6, it is less than the difference threshold, and no fault diagnosis conclusion can be drawn.

[0130] The matching degree matrix P is filtered by column, and the columns that are all 0 or 1 (corresponding principal components cannot be used as features to distinguish fault modes) are removed to form an optimized matching degree matrix, as shown in the following table:

[0131] Table 3 Optimized matching matrix

[0132]

[0133] After weight optimization, the discrimination degree D = [1, 1, 3, 1, 2, 3] is obtained through the discrimination degree calculation formula.

[0134] After optimization, the matching degree vector The threshold condition is met (0.81-0.45>0.3), so F is selected. 2 As the final fault diagnosis result.

[0135] In summary, the rotating machinery image fault diagnosis method based on the interval model provided in the embodiment of the present invention resamples the original displacement signal under different fault modes to form multiple fault samples and draw the time domain waveform corresponding to each fault sample, and uses directional gradient feature extraction and principal component analysis to reduce the dimension of the time domain waveform to establish a feature matrix, and establishes an interval model to form an interval feature vector to improve the accuracy of uncertainty quantification. By performing the same processing on the sample to be tested, the principal component vector of the sample to be tested is obtained, and the positional relationship between the principal component vector and the corresponding element of the fault interval feature vector is calculated to establish a matching matrix, thereby improving the ability to handle the dispersion of image features. The rationality of the fault diagnosis result is judged by the difference threshold, and the weight optimization based on the discrimination is performed on the sample to be tested with unreasonable fault diagnosis results to obtain the final fault diagnosis result. By establishing a matching degree calculation and weight optimization method, the credibility of the rotating machinery image fault diagnosis result under dispersion conditions is improved.

[0136] In this specification, each embodiment or implementation is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referenced to each other. In the description of this specification, the description of reference terms "one implementation", "some implementations", "illustrative implementations", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the implementation or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same implementation or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more implementations or examples in a suitable manner.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rotating machinery image fault diagnosis method based on interval model, characterized in that: include: Step 1: Under different fault modes, the original displacement signal of the rotating machinery is intercepted by using a resampling method to obtain multiple fault samples, and a time domain waveform corresponding to each fault sample is formed; Step 2: extract the directional gradient histogram features of each of the time domain waveforms to obtain the corresponding directional gradient histogram feature vectors, all the feature vectors form the fault sample feature matrix, and use the principal component analysis method to reduce the dimension of the fault sample feature matrix to obtain the principal component matrix of the fault sample feature matrix; Step 3: Obtain a sample set under each of the fault modes according to the principal component matrix, establish an interval model for each of the fault modes within an extreme value range, and all of the interval models respectively form interval feature vectors corresponding to the fault modes; Step 4: obtaining the directional gradient histogram feature vector of the sample to be tested of the rotating machinery, and reducing the dimension of the directional gradient histogram feature vector of the sample to be tested by using the principal component analysis method to obtain the principal component vector of the sample to be tested; Step 5: According to the positional relationship between the interval feature vector of each fault mode and the corresponding element in the principal component vector, a matching degree matrix between the sample to be tested and the fault sample is obtained, and a matching degree vector is calculated for each element of the interval feature vector using the same weight; Step six: Calculate the difference between the first matching degree and the second matching degree in the matching degree vector. For the samples to be tested that meet the difference threshold, select the fault mode corresponding to the first matching degree to obtain a fault diagnosis conclusion; for the samples to be tested that do not meet the difference threshold, reallocate the weights according to the discrimination degree of each element of the interval feature vector in the matching degree matrix under each fault mode and calculate the matching degree vector, select the fault mode corresponding to the calculated first matching degree to obtain the fault diagnosis conclusion.

2. The rotating machinery image fault diagnosis method based on interval model according to claim 1 is characterized in that: In the step three, the interval model is used to perform interval estimation on the sample set under each of the fault modes in the principal component matrix to form the interval feature vector corresponding to the fault mode.

Images