A method for detecting motor shaft damage
By processing the motor shaft vibration signal, Hankel matrix is constructed and singular value decomposition and Fourier transform are performed, combined with the dual-channel enhanced compensation model, the problem of low ultrasonic detection accuracy in motor shaft damage detection is solved, and the more High-precision damage degree prediction.
Patent Information
- Application Number
- CN202411182904.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-27
AI Technical Summary
In the prior art, in motor shaft damage detection, ultrasonic waves are used when ultrasonic is used, and the motor structure is easily disturbed by structural interference, resulting in low detection accuracy.
Collect the vibration signals of the motor axis, build the Hankel matrix and perform singular value decomposition, calculate the difference coefficients between the singular value vector and the standard vector, extract the spectrum eigenvalues through Fourier transform, separate the main feature sequence and the non-main feature sequence, and combine it with Dual-channel enhanced compensation model predicts the degree of damage.
Through detailed analysis of vibration signals, the degree of damage to the motor shaft is accurately predicted, avoiding the problem of low accuracy of ultrasonic detection and improving the detection accuracy.
Smart Images

Figure CN118706950B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of motor shaft damage detection, and in particular to a motor shaft damage detection method. Background Art
[0002] The motor is an indispensable power source in modern industry. Its performance and reliability directly affect the stable operation of the production line. As a key component for transmitting torque, the health of the motor shaft has an important impact on the performance of the entire motor. Therefore, it is very important to effectively detect and evaluate the damage of the motor shaft.
[0003] In the prior art, ultrasonic waves are often used for damage detection for mechanical damage. However, the motor structure is complex and there are seams at the joints of the components. Therefore, when ultrasonic waves are used for damage detection, they may be interfered by the structure, thereby affecting the accuracy and efficiency of the detection. Summary of the invention
[0004] In view of the above-mentioned deficiencies in the prior art, a motor shaft damage detection method provided by the present invention solves the problem of low detection accuracy in the prior art damage detection using ultrasonic waves.
[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a motor shaft damage detection method, comprising the following steps:
[0006] S1, collecting the vibration signal of the motor shaft, segmenting the vibration signal, and constructing a Hankel matrix for each segment of the signal;
[0007] S2. Perform singular value decomposition on each Hankel matrix to construct a singular value vector, where the singular value vector is composed of each singular value corresponding to a Hankel matrix;
[0008] S3, calculating the difference coefficient between each singular value vector and the singular value standard vector, and constructing a difference coefficient vector;
[0009] S4, performing Fourier transform on each row in the Hankel matrix, extracting the spectral eigenvalues, and separating the main feature sequence and the non-main feature sequence;
[0010] S5. According to the main feature sequence and the non-main feature sequence, based on the compensation of the difference coefficient vector, a dual-channel enhanced compensation model is used to predict the degree of motor shaft damage.
[0011] The beneficial effects of the present invention are as follows: the present invention collects the vibration signal of the motor shaft, then segments the vibration signal, constructs a Hankel matrix for each signal segment, performs singular value decomposition on the Hankel matrix, constructs a singular value vector, and constructs a difference coefficient vector based on the difference coefficient between the singular value vector and the singular value standard vector, reflecting the slight changes and potential damage characteristics of each signal segment. The present invention performs Fourier transform on each row in the Hankel matrix, extracts each spectrum eigenvalue, and divides each spectrum eigenvalue into a main feature sequence and a non-main feature sequence. A dual-channel enhanced compensation model is applied to combine the main feature sequence and the non-main feature sequence and the compensation of the difference coefficient vector to accurately predict the degree of damage to the motor shaft. The present invention is based on the analysis of vibration signals, avoiding the problem of low detection accuracy caused by using ultrasonic waves for damage detection.
[0012] Furthermore, S3 includes the following sub-steps:
[0013] S31, calculating the Pearson correlation coefficient between each singular value vector and the singular value standard vector;
[0014] S32, calculating the intersection-and-union ratio of each singular value vector and the singular value standard vector;
[0015] S33, calculating the difference coefficient corresponding to each singular value vector according to the Pearson correlation coefficient and the intersection-over-union ratio;
[0016] S34. Use each difference coefficient as an element to construct a difference coefficient vector.
[0017] Furthermore, the formula for calculating the coefficient of difference in S33 is:
[0018] ,
[0019] Among them, ε is the difference coefficient, α1 is the first weighting coefficient, α2 is the second weighting coefficient, X p is the Pearson correlation coefficient, X I It is the intersection and comparison.
[0020] The beneficial effect of the above further scheme is: the present invention reflects the difference between the singular value vector and the singular value standard vector through the Pearson correlation coefficient and the intersection and union ratio of each singular value vector and the singular value standard vector. The larger the difference coefficient, the greater the deviation of the vibration signal from the normal range.
[0021] Furthermore, the S4 comprises the following sub-steps:
[0022] S41, performing Fourier transform on each row in the Hankel matrix to obtain spectrum information of each row;
[0023] S42, calculating the spectrum characteristic value according to each line of spectrum information;
[0024] S43, forming a spectrum feature set from spectrum feature values corresponding to a Hankel matrix;
[0025] S44, taking the intersection of each spectrum feature set;
[0026] S45, counting the number of times the spectrum feature value in the intersection occurs in each spectrum feature set;
[0027] S46, taking each spectrum feature value whose occurrence number is greater than the threshold as an element to construct a main feature sequence;
[0028] S47, taking each spectrum feature value whose occurrence number is less than or equal to the threshold as an element to construct a non-main feature sequence.
[0029] The beneficial effect of the above further scheme is as follows: the present invention performs Fourier transform on each row in the Hankel matrix to obtain the spectrum information of each row, calculates the spectrum eigenvalue for the spectrum information of each row, and forms a spectrum feature set with the spectrum eigenvalues corresponding to a Hankel matrix. There are multiple Hankel matrices, and therefore, there are multiple spectrum feature sets. The intersection of each spectrum feature set is taken to find the spectrum eigenvalues that exist in each signal segment, and the number of occurrences of the existing spectrum eigenvalues is counted to find the spectrum eigenvalues that appear more often to form a main feature sequence, and the spectrum eigenvalues that appear less often to form a non-main feature sequence.
[0030] The spectral eigenvalues in the main feature sequence reflect the main signal components of the vibration signal and represent the most significant damage-related features. The non-main feature sequence reflects the non-main signal components of the vibration signal and represents the non-significant damage-related features in the vibration signal.
[0031] Furthermore, the formula for calculating the spectrum characteristic value in S42 is:
[0032] ,
[0033] Among them, s is the spectral characteristic value, e is the natural constant, θ i is the phase of the i-th frequency component in each line of spectrum information, F i is the amplitude of the i-th frequency component in each line of spectrum information, f i is the i-th frequency component in each line of spectrum information, where i is a positive integer.
[0034] The beneficial effect of the above further solution is that the spectrum eigenvalue in the present invention is a fusion of phase, amplitude and frequency components, so that the spectrum eigenvalue reflects the component characteristics of each row of data in the Hankel matrix.
[0035] Furthermore, the dual-channel enhancement compensation model in S5 includes: a main feature processing channel, a non-main feature processing channel, a Concat layer, a fully connected layer and a compensation layer;
[0036] The input end of the main feature processing channel is used to input the main feature sequence; the input end of the non-main feature processing channel is used to input the non-main feature sequence; the input end of the Concat layer is respectively connected to the output end of the main feature processing channel and the output end of the non-main feature processing channel, and its output end is connected to the input end of the fully connected layer; the input end of the compensation layer is connected to the output end of the fully connected layer, and its output end serves as the output end of the dual-channel enhanced compensation model.
[0037] The beneficial effects of the above-mentioned further scheme are as follows: the present invention adopts the main feature processing channel to process the main feature sequence, and adopts the non-main feature processing channel to process the non-main feature sequence, so as to realize the separate processing of the two sequences, and can simultaneously process the significant features related to damage and possible secondary features, thereby ensuring the comprehensive capture of damage information and reducing information omission. The Concat layer merges the outputs of the main feature processing channel and the non-main feature processing channel, realizes the effective fusion of different types of features, and improves the model's recognition ability for complex damage patterns. The fully connected layer calculates the initial damage degree according to the features merged by the Concat layer, and then uses the compensation layer to compensate for the initial damage degree, thereby improving the detection accuracy of the motor shaft damage degree.
[0038] Furthermore, the main feature processing channel and the non-main feature processing channel both include: a first convolution layer, an enhancement layer, a second convolution layer, a Sigmoid layer, a multiplier M1 and a pooling layer;
[0039] The input end of the first convolution layer serves as the input end of the main feature processing channel or the non-main feature processing channel, and its output end is connected to the first input end of the enhancement layer; the output end of the enhancement layer is respectively connected to the input end of the second convolution layer and the first input end of the multiplier M1; the output end of the second convolution layer is connected to the input end of the Sigmoid layer; the second input end of the multiplier M1 is connected to the output end of the Sigmoid layer, and its output end is connected to the input end of the pooling layer; the output end of the pooling layer serves as the output end of the main feature processing channel or the non-main feature processing channel.
[0040] Furthermore, the first input end of the enhancement layer is used to input a feature vector composed of feature values output by the first convolutional layer, and the second input end thereof is used to input an enhancement vector, wherein the expression of the enhancement coefficient in the enhancement vector is:
[0041] ,
[0042] Among them, j is the jth enhancement coefficient in the enhancement vector, j is a positive integer. In the main feature processing channel, r j is the number of occurrences of the jth spectral eigenvalue in the main feature sequence. In the non-main feature processing channel, rj is the number of occurrences of the jth spectral feature value in the non-main feature sequence.
[0043] Furthermore, the expression of the enhancement layer is:
[0044] ,
[0045] Among them, R is the output of the enhancement layer, EN is the enhancement vector, X is the feature vector composed of the eigenvalues output by the first convolutional layer, It is Hadamard.
[0046] The beneficial effect of the above further scheme is: after the first convolutional layer extracts features, the present invention uses an enhancement layer to enhance the features output by the first convolutional layer. The enhancement vector comes from the number of times the spectral eigenvalues appear. The more times the spectral eigenvalues appear, the larger the enhancement coefficient is, and the higher the attention to important eigenvalues is. The second convolutional layer, the Sigmoid layer and the self-attention channel of the multiplier M1 are set, so that the model has the ability to screen key features and improve the prediction accuracy of the model.
[0047] Furthermore, the expression of the compensation layer is:
[0048] ,
[0049] Among them, y is the damage degree of the motor shaft, y0 is the initial damage degree output by the fully connected layer, e is a natural constant, ε n is the nth difference coefficient in the difference coefficient vector, ω n is n The weight of , N is the number of difference coefficients in the difference coefficient vector, and n is a positive integer.
[0050] The beneficial effect of the above further solution is: the present invention compensates the initial damage degree according to each difference coefficient of the difference coefficient vector, and the larger the difference coefficient is, the greater the damage degree of the motor shaft is. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 is a flow chart of a motor shaft damage detection method;
[0052] Figure 2 It is a structural diagram of a dual-channel enhanced compensation model;
[0053] Figure 3 Schematic diagram of the structure of the main feature processing channel and the non-main feature processing channel. DETAILED DESCRIPTION
[0054] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0055] like Figure 1 As shown, a motor shaft damage detection method comprises the following steps:
[0056] S1, collecting the vibration signal of the motor shaft, segmenting the vibration signal, and constructing a Hankel matrix for each segment of the signal;
[0057] S2. Perform singular value decomposition on each Hankel matrix to construct a singular value vector, where the singular value vector is composed of each singular value corresponding to a Hankel matrix;
[0058] S3, calculating the difference coefficient between each singular value vector and the singular value standard vector, and constructing a difference coefficient vector;
[0059] S4, performing Fourier transform on each row in the Hankel matrix, extracting the spectral eigenvalues, and separating the main feature sequence and the non-main feature sequence;
[0060] S5. According to the main feature sequence and the non-main feature sequence, based on the compensation of the difference coefficient vector, a dual-channel enhanced compensation model is used to predict the degree of motor shaft damage.
[0061] In this embodiment, the singular value standard vector is a singular value vector obtained according to a vibration signal obtained from a normal motor, that is, the singular value standard vector is a stored normal singular value vector.
[0062] The S3 comprises the following sub-steps:
[0063] S31, calculating the Pearson correlation coefficient between each singular value vector and the singular value standard vector;
[0064] S32, calculating the intersection-and-union ratio of each singular value vector and the singular value standard vector;
[0065] S33, calculating the difference coefficient corresponding to each singular value vector according to the Pearson correlation coefficient and the intersection-over-union ratio;
[0066] S34. Use each difference coefficient as an element to construct a difference coefficient vector.
[0067] The formula for calculating the coefficient of difference in S33 is:
[0068] ,
[0069] Among them, ε is the difference coefficient, α1 is the first weighting coefficient, α2 is the second weighting coefficient, X p is the Pearson correlation coefficient, X I It is the intersection and comparison.
[0070] The present invention reflects the difference between a singular value vector and a singular value standard vector through the Pearson correlation coefficient and the intersection-and-union ratio of each singular value vector and a singular value standard vector. The larger the difference coefficient, the greater the deviation of the vibration signal from the normal range.
[0071] The S4 comprises the following sub-steps:
[0072] S41, performing Fourier transform on each row in the Hankel matrix to obtain spectrum information of each row;
[0073] S42, calculating the spectrum characteristic value according to each line of spectrum information;
[0074] S43, forming a spectrum feature set from spectrum feature values corresponding to a Hankel matrix;
[0075] S44, taking the intersection of each spectrum feature set;
[0076] S45, counting the number of times the spectrum feature value in the intersection occurs in each spectrum feature set;
[0077] S46, taking each spectrum feature value whose occurrence number is greater than the threshold as an element to construct a main feature sequence;
[0078] S47, taking each spectrum feature value whose occurrence number is less than or equal to the threshold as an element to construct a non-main feature sequence.
[0079] The present invention constructs each signal segment as a Hankel matrix, and by reorganizing the time series data into a matrix form, the data of each time step can be compared and correlated with the data of other time steps. This representation helps to reveal the hidden patterns and periodicity in the signal, thereby more comprehensively capturing the operating status of the motor shaft.
[0080] In step S44 of the present invention, the intersection of each spectrum feature set is taken to extract the signal components existing in the vibration signal and eliminate accidental factors.
[0081] In this embodiment, the threshold is set according to demand.
[0082] The present invention performs Fourier transform on each row in the Hankel matrix to obtain the spectrum information of each row, calculates the spectrum eigenvalue for the spectrum information of each row, and forms a spectrum feature set with the spectrum eigenvalues corresponding to a Hankel matrix. There are multiple Hankel matrices, and therefore, there are multiple spectrum feature sets. The intersection of each spectrum feature set is taken to find the spectrum eigenvalues existing in each signal segment, and the number of occurrences of the existing spectrum eigenvalues is counted. The spectrum eigenvalues with a large number of occurrences are found to form a main feature sequence, and the spectrum eigenvalues with a small number of occurrences form a non-main feature sequence.
[0083] The spectral eigenvalues in the main feature sequence reflect the main signal components of the vibration signal and represent the most significant damage-related features. The non-main feature sequence reflects the non-main signal components of the vibration signal and represents the non-significant damage-related features in the vibration signal.
[0084] The formula for calculating the spectrum characteristic value in S42 is:
[0085] ,
[0086] Among them, s is the spectral characteristic value, e is the natural constant, θ i is the phase of the i-th frequency component in each line of spectrum information, F i is the amplitude of the i-th frequency component in each line of spectrum information, f i is the i-th frequency component in each line of spectrum information, where i is a positive integer.
[0087] The spectrum eigenvalue in the present invention is a fusion of phase, amplitude and frequency components, so that the spectrum eigenvalue reflects the component characteristics of each row of data in the Hankel matrix.
[0088] like Figure 2 As shown, the dual-channel enhancement compensation model in S5 includes: a main feature processing channel, a non-main feature processing channel, a Concat layer, a fully connected layer and a compensation layer;
[0089] The input end of the main feature processing channel is used to input the main feature sequence; the input end of the non-main feature processing channel is used to input the non-main feature sequence; the input end of the Concat layer is respectively connected to the output end of the main feature processing channel and the output end of the non-main feature processing channel, and its output end is connected to the input end of the fully connected layer; the input end of the compensation layer is connected to the output end of the fully connected layer, and its output end serves as the output end of the dual-channel enhanced compensation model.
[0090] The present invention adopts the main feature processing channel to process the main feature sequence, and adopts the non-main feature processing channel to process the non-main feature sequence, so as to realize the separate processing of the two sequences, and can simultaneously process the significant features related to the damage and the possible secondary features, thereby ensuring the comprehensive capture of the damage information and reducing the information omission. The Concat layer merges the outputs of the main feature processing channel and the non-main feature processing channel, realizes the effective fusion of different types of features, and improves the model's recognition ability for complex damage patterns. The fully connected layer calculates the initial damage degree according to the features merged by the Concat layer, and then uses the compensation layer to compensate for the initial damage degree, thereby improving the detection accuracy of the motor shaft damage degree.
[0091] like Figure 3 As shown, the main feature processing channel and the non-main feature processing channel both include: a first convolution layer, an enhancement layer, a second convolution layer, a Sigmoid layer, a multiplier M1 and a pooling layer;
[0092] The input end of the first convolution layer serves as the input end of the main feature processing channel or the non-main feature processing channel, and its output end is connected to the first input end of the enhancement layer; the output end of the enhancement layer is respectively connected to the input end of the second convolution layer and the first input end of the multiplier M1; the output end of the second convolution layer is connected to the input end of the Sigmoid layer; the second input end of the multiplier M1 is connected to the output end of the Sigmoid layer, and its output end is connected to the input end of the pooling layer; the output end of the pooling layer serves as the output end of the main feature processing channel or the non-main feature processing channel.
[0093] The first input terminal of the enhancement layer is used to input a feature vector composed of feature values output by the first convolutional layer, and the second input terminal is used to input an enhancement vector, wherein the expression of the enhancement coefficient in the enhancement vector is:
[0094] ,
[0095] Among them, j is the jth enhancement coefficient in the enhancement vector, j is a positive integer. In the main feature processing channel, r j is the number of occurrences of the jth spectral eigenvalue in the main feature sequence. In the non-main feature processing channel, r j is the number of occurrences of the jth spectral feature value in the non-main feature sequence.
[0096] The expression of the enhancement layer is:
[0097] ,
[0098] Among them, R is the output of the enhancement layer, EN is the enhancement vector, and X is the feature vector composed of the eigenvalues output by the first convolutional layer. It is Hadamard.
[0099] After the first convolutional layer extracts features, the present invention uses an enhancement layer to enhance the features output by the first convolutional layer. The enhancement vector comes from the number of times the spectrum eigenvalue appears. When the number of times the spectrum eigenvalue appears is greater, the enhancement coefficient is greater, and the attention to important eigenvalues is higher. In addition, a second convolutional layer, a Sigmoid layer and a self-attention channel of the multiplier M1 are set, so that the model has the ability to screen key features and improve the prediction accuracy of the model.
[0100] The expression of the compensation layer is:
[0101] ,
[0102] Among them, y is the damage degree of the motor shaft, y0 is the initial damage degree output by the fully connected layer, e is a natural constant, ε n is the nth difference coefficient in the difference coefficient vector, ω n is n The weight of , N is the number of difference coefficients in the difference coefficient vector, and n is a positive integer.
[0103] In this embodiment, the dual-channel enhancement compensation model is trained by gradient descent method.
[0104] The present invention performs compensation processing on the initial damage degree according to each difference coefficient of the difference coefficient vector. When each difference coefficient is larger, the damage degree of the motor shaft is larger.
[0105] The present invention collects the vibration signal of the motor shaft, then segments the vibration signal, constructs a Hankel matrix for each signal segment, performs singular value decomposition on the Hankel matrix, constructs a singular value vector, and constructs a difference coefficient vector according to the difference coefficient between the singular value vector and the singular value standard vector, reflecting the slight changes and potential damage characteristics of each signal segment. The present invention performs Fourier transform on each row in the Hankel matrix, extracts each spectrum characteristic value, and divides each spectrum characteristic value into a main characteristic sequence and a non-main characteristic sequence. A dual-channel enhanced compensation model is applied to combine the main characteristic sequence and the non-main characteristic sequence and the compensation of the difference coefficient vector to accurately predict the degree of damage to the motor shaft. The present invention is based on the analysis of the vibration signal, avoiding the problem of low detection accuracy caused by using ultrasonic waves for damage detection.
[0106] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A motor shaft damage detection method, characterized in that: The following steps are involved: S1, collecting the vibration signal of the motor shaft, segmenting the vibration signal, and constructing a Hankel matrix for each segment of the signal; S2. Perform singular value decomposition on each Hankel matrix to construct a singular value vector, where the singular value vector is composed of each singular value corresponding to a Hankel matrix; S3, calculating the difference coefficient between each singular value vector and the singular value standard vector, and constructing a difference coefficient vector; S4, performing Fourier transform on each row in the Hankel matrix, extracting the spectral eigenvalues, and separating the main feature sequence and the non-main feature sequence; S5, according to the main feature sequence and the non-main feature sequence, based on the compensation of the difference coefficient vector, a dual-channel enhanced compensation model is used to predict the degree of motor shaft damage; The dual-channel enhancement compensation model in S5 includes: a main feature processing channel, a non-main feature processing channel, a Concat layer, a fully connected layer and a compensation layer; The input end of the main feature processing channel is used to input the main feature sequence; the input end of the non-main feature processing channel is used to input the non-main feature sequence; the input end of the Concat layer is connected to the output end of the main feature processing channel and the output end of the non-main feature processing channel respectively, and its output end is connected to the input end of the fully connected layer; the input end of the compensation layer is connected to the output end of the fully connected layer, and its output end is used as the output end of the dual-channel enhanced compensation model; The expression of the compensation layer is: , Among them, y is the damage degree of the motor shaft, y0 is the initial damage degree output by the fully connected layer, e is a natural constant, ε n is the nth difference coefficient in the difference coefficient vector, ω n is n The weight of , N is the number of difference coefficients in the difference coefficient vector, and n is a positive integer.
2. The motor shaft damage detection method according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, calculating the Pearson correlation coefficient between each singular value vector and the singular value standard vector; S32, calculating the intersection-and-union ratio of each singular value vector and the singular value standard vector; S33, calculating the difference coefficient corresponding to each singular value vector according to the Pearson correlation coefficient and the intersection-over-union ratio; S34. Use each difference coefficient as an element to construct a difference coefficient vector.
3. The motor shaft damage detection method according to claim 2, characterized in that: The formula for calculating the coefficient of difference in S33 is: , Among them, ε is the difference coefficient, α1 is the first weighting coefficient, α2 is the second weighting coefficient, X p is the Pearson correlation coefficient, X I It is the intersection and comparison.
4. The motor shaft damage detection method according to claim 1, characterized in that: The S4 comprises the following sub-steps: S41, performing Fourier transform on each row in the Hankel matrix to obtain spectrum information of each row; S42, calculating the spectrum characteristic value according to each line of spectrum information; S43, forming a spectrum feature set from spectrum feature values corresponding to a Hankel matrix; S44, taking the intersection of each spectrum feature set; S45, counting the number of times the spectrum feature value in the intersection occurs in each spectrum feature set; S46, taking each spectrum feature value whose occurrence number is greater than the threshold as an element to construct a main feature sequence; S47, taking each spectrum feature value whose occurrence number is less than or equal to the threshold as an element to construct a non-main feature sequence.
5. The motor shaft damage detection method according to claim 4, characterized in that: The formula for calculating the spectrum characteristic value in S42 is: , Among them, s is the spectral characteristic value, e is the natural constant, θ i is the phase of the i-th frequency component in each line of spectrum information, F i is the amplitude of the i-th frequency component in each line of spectrum information, f i is the i-th frequency component in each line of spectrum information, where i is a positive integer.
6. The motor shaft damage detection method according to claim 1, characterized in that: The main feature processing channel and the non-main feature processing channel both include: a first convolution layer, an enhancement layer, a second convolution layer, a Sigmoid layer, a multiplier M1 and a pooling layer; The input end of the first convolution layer serves as the input end of the main feature processing channel or the non-main feature processing channel, and its output end is connected to the first input end of the enhancement layer; the output end of the enhancement layer is respectively connected to the input end of the second convolution layer and the first input end of the multiplier M1; the output end of the second convolution layer is connected to the input end of the Sigmoid layer; the second input end of the multiplier M1 is connected to the output end of the Sigmoid layer, and its output end is connected to the input end of the pooling layer; the output end of the pooling layer serves as the output end of the main feature processing channel or the non-main feature processing channel.
7. The motor shaft damage detection method according to claim 6, characterized in that: The first input terminal of the enhancement layer is used to input a feature vector composed of feature values output by the first convolutional layer, and the second input terminal is used to input an enhancement vector, wherein the expression of the enhancement coefficient in the enhancement vector is: , Among them, j is the jth enhancement coefficient in the enhancement vector, j is a positive integer. In the main feature processing channel, r j is the number of occurrences of the jth spectral eigenvalue in the main feature sequence. In the non-main feature processing channel, r j is the number of occurrences of the jth spectral feature value in the non-main feature sequence.
8. The motor shaft damage detection method according to claim 7, characterized in that: The expression of the enhancement layer is: , Among them, R is the output of the enhancement layer, EN is the enhancement vector, X is the feature vector composed of the eigenvalues output by the first convolutional layer, It is Hadamard.
