Monitoring and early warning method, system and storage medium for working status of six-sided top press

Abstract

This invention discloses a method, system, and storage medium for monitoring and early warning of the working status of a six-sided top press, belonging to the field of machine structure testing technology. This invention constructs a Hankel matrix from vibration signals, thereby capturing the dynamic characteristics of the equipment during actual operation and improving monitoring accuracy. Based on the Hankel matrix, a dominant frequency matrix is ​​constructed to reflect the dominant frequency in different signal segments. The constructed dominant frequency mean sequence reflects the average level of the dominant frequency, and the constructed dominant frequency change sequence reflects the changes in the dominant frequency of different signal segments, facilitating better capture of the working status of the six-sided top press. The use of a dual-channel feature fusion BP neural network processing method can more effectively fuse the frequency mean sequence and frequency change sequence, achieving a comprehensive evaluation of the working status and improving the sensitivity and accuracy of potential fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of machine structure testing, and in particular to a monitoring and early warning method, system and storage medium for the working state of a six-sided top press. Background Art

[0002] The six-sided top press is an important equipment used for high-pressure experiments and material research, and is widely used in geology, material science, metallurgy and other fields. Its main function is to study the physical and chemical properties of materials under extreme conditions by applying pressure evenly. With the advancement of science and technology, the design and manufacturing technology of the six-sided top press has been continuously improved, but in practical applications, the safety and reliability of the equipment still face challenges. The existing technology mainly relies on traditional monitoring methods, such as pressure sensors and displacement sensors. These methods can often only provide limited static data and it is difficult to fully reflect the dynamic working status of the equipment. At the same time, the existing technology monitors the sensor data and only uses a simple threshold judgment method to determine the normal and abnormal of the corresponding monitoring structure, which has the problem of low monitoring accuracy. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a monitoring and early warning method, system and storage medium for the working status of a six-sided top press, which solves the problem of low monitoring accuracy in the prior art.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a monitoring and early warning method for the working state of a six-sided top press, comprising the following steps:

[0005] S1, collecting the vibration signal of the six-sided top press when the press head moves, and forming a Hankel matrix;

[0006] S2. Construct the dominant frequency matrix based on the Hankel matrix;

[0007] S3, extracting frequency eigenvalues ​​from each column of the dominant frequency matrix, and constructing a dominant frequency mean sequence and a dominant frequency change sequence;

[0008] S4. Use dual-channel feature fusion BP neural network to process the dominant frequency mean sequence and the dominant frequency change sequence to obtain the working state value of the six-sided top press.

[0009] Furthermore, the S2 comprises the following sub-steps:

[0010] S21, performing Fourier transform on each row of the Hankel matrix to obtain spectrum information of each row;

[0011] S22, dividing the frequency value in each row of the spectrum information into four areas, wherein the four areas include: a low frequency area, a medium frequency area, a high frequency area and an ultra-high frequency area;

[0012] S23. Extract the dominant frequency in each area to form a dominant frequency matrix, wherein each row in the dominant frequency matrix includes four elements, the first element is the dominant frequency of the low frequency area, the second element is the dominant frequency of the medium frequency area, the third element is the dominant frequency of the high frequency area, and the fourth element is the dominant frequency of the ultra-high frequency area.

[0013] Furthermore, the S23 comprises the following sub-steps:

[0014] S231, calculating the average value of the amplitude in each zone to obtain the average amplitude;

[0015] S232, marking an amplitude greater than the average amplitude in a zone as a candidate amplitude;

[0016] S233, in one area, extracting the frequency value corresponding to the candidate amplitude, and calculating the mean of each extracted frequency value to obtain the dominant frequency;

[0017] S234. Construct a dominant frequency matrix from each dominant frequency.

[0018] Furthermore, S3 comprises the following steps:

[0019] S31, taking the average value of each column element of the dominant frequency matrix, and using the average value of each column element as an element to construct a dominant frequency mean sequence;

[0020] S32. Calculate the frequency change coefficient for each column element of the dominant frequency matrix, and use the frequency change coefficient of each column element as an element to construct a dominant frequency change sequence.

[0021] Furthermore, the formula for calculating the frequency variation coefficient in S32 is: , where μ i is the i-th frequency variation coefficient, f i,j is the jth element in the i-th column of the dominant frequency matrix, where i and j are positive integers and N is the number of elements in a column.

[0022] Further, the dual-channel feature fusion BP neural network in S4 includes: a first channel, a second channel, a channel coupling layer and a BP neural network;

[0023] The input end of the first channel is used to input the dominant frequency mean sequence; the input end of the second channel is used to input the dominant frequency change sequence; the input end of the channel fusion layer is respectively connected to the output end of the first channel and the output end of the second channel, and its output end is connected to the input end of the BP neural network; the output end of the BP neural network serves as the output end of the dual-channel feature fusion BP neural network.

[0024] Furthermore, the first channel and the second channel both include: a tanh layer and a BN layer connected in sequence, wherein the expression of the tanh layer is: , where s n is the nth eigenvalue output by the tanh layer, x n is the nth element of the tanh layer input, w x,n For x n The weight of b x,n For x n The bias of , tanh is the hyperbolic tangent activation function, the BN layer is used to normalize the multiple eigenvalues ​​output by the tanh layer, and n is a positive integer.

[0025] Furthermore, the expression of the channel coupling layer is:

[0026] ,

[0027] Among them, g n is the nth eigenvalue output by the channel coupling layer, g 1,n is the nth eigenvalue of the first channel output, g 2,n is the nth eigenvalue of the second channel output, w g1,n g 1,n The weight, w g2,n g 2,n The weight of b g1,n g 1,n The bias, b g2,n g 2,n The bias of the channel coupling layer is W1, W2 is the weight of the second channel, n is a positive integer, σ is the Sigmoid activation function, and the four eigenvalues ​​output by the channel coupling layer are used as the input of the BP neural network.

[0028] A monitoring and early warning system for the working state of a six-sided top press comprises: a Hankel matrix construction unit, a frequency matrix construction unit, a feature extraction unit and a state value prediction unit;

[0029] The Hankel matrix construction unit is used to collect vibration signals of the six-sided top press when the press head moves, and form a Hankel matrix;

[0030] The frequency matrix construction unit is used to construct a dominant frequency matrix according to the Hankel matrix;

[0031] The feature extraction unit is used to extract frequency feature values ​​from each column of the dominant frequency matrix, and construct a dominant frequency mean value sequence and a dominant frequency change sequence;

[0032] The state value prediction unit is used to process the dominant frequency mean value sequence and the dominant frequency change sequence by using a dual-channel feature fusion BP neural network to obtain the working state value of the six-sided top press.

[0033] A storage medium stores a program, which implements the above method when executed by a processor.

[0034] The beneficial effects of the present invention are as follows: the present invention constructs a Hankel matrix for the vibration signal, thereby being able to capture the dynamic characteristics of the equipment in actual operation and improve the accuracy of monitoring; and then constructs a dominant frequency matrix based on the Hankel matrix to reflect the dominant frequencies in different signal segments; the constructed dominant frequency mean sequence reflects the average level of the dominant frequencies; the constructed dominant frequency change sequence reflects the changes in the dominant frequencies of different signal segments, thereby better capturing the working conditions of the six-sided top press; and the dual-channel feature fusion BP neural network processing method can more effectively fuse the frequency mean sequence and the frequency change sequence, thereby achieving a comprehensive evaluation of the working status and improving the sensitivity to potential faults and the accuracy of diagnosis.

[0035] The present invention retains the time sequence characteristics of the signal by constructing the Hankel matrix, so that the constructed dominant frequency matrix can reflect the changes of the dominant frequencies in different time periods, can accurately capture the frequency changes of the vibration signal, and improves the monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a monitoring and early warning method for the working status of a six-sided top press;

[0037] Figure 2 Schematic diagram of the structure of the dual-channel feature fusion BP neural network. DETAILED DESCRIPTION

[0038] 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.

[0039] Embodiment 1, as Figure 1 As shown, a monitoring and early warning method for the working status of a six-sided top press includes the following steps:

[0040] S1, collecting the vibration signal of the six-sided top press when the press head moves, and forming a Hankel matrix;

[0041] S2. Construct the dominant frequency matrix based on the Hankel matrix;

[0042] S3, extracting frequency eigenvalues ​​from each column of the dominant frequency matrix, and constructing a dominant frequency mean sequence and a dominant frequency change sequence;

[0043] S4. Use dual-channel feature fusion BP neural network to process the dominant frequency mean sequence and the dominant frequency change sequence to obtain the working state value of the six-sided top press.

[0044] In this embodiment, a vibration sensor on the ram of the six-sided top press is used to collect vibration signals when the ram of the six-sided top press moves, and the amplitude corresponding to each moment of the vibration signal is filled into the Hankel matrix. The Hankel matrix is: , where r1 is the first amplitude of the vibration signal, r2 is the second amplitude of the vibration signal, r3 is the third amplitude of the vibration signal, r4 is the fourth amplitude of the vibration signal, and r m is the mth amplitude in the vibration signal, r m+1 is the m+1th amplitude in the vibration signal, r m+2 is the m+2th amplitude in the vibration signal, r 2m-1 is the 2m-1th amplitude in the vibration signal, where m is a positive integer.

[0045] In the present invention, the vibration signal of a pressing head during movement constitutes a Hankel matrix, and the working state value of the six-sided top press reflects the working condition of a pressing head.

[0046] In this embodiment, S2 includes the following sub-steps:

[0047] S21, performing Fourier transform on each row of the Hankel matrix to obtain spectrum information of each row;

[0048] S22, dividing the frequency value in each row of the spectrum information into four areas, wherein the four areas include: a low frequency area, a medium frequency area, a high frequency area and an ultra-high frequency area;

[0049] S23. Extract the dominant frequency in each area to form a dominant frequency matrix, wherein each row in the dominant frequency matrix includes four elements, the first element is the dominant frequency of the low frequency area, the second element is the dominant frequency of the medium frequency area, the third element is the dominant frequency of the high frequency area, and the fourth element is the dominant frequency of the ultra-high frequency area.

[0050] In the present invention, the low frequency zone ranges from 0 to 100 Hz, the medium frequency zone ranges from 100 Hz to 1000 Hz, the high frequency zone ranges from 1 kHz to 10 kHz, and the ultra-high frequency zone is above 10 kHz. The low frequency zone captures the basic working frequency and mechanical structure vibration, the medium frequency zone identifies common mechanical failure characteristics, the high frequency zone reflects early damage signals of components, and the ultra-high frequency zone detects microscopic defects and material fatigue.

[0051] The present invention performs Fourier transform on each row of the Hankel matrix to obtain the spectrum information of each row, and divides the frequency value of each row into four areas, which not only reduces the mutual interference between frequency bands and improves the signal-to-noise ratio, but also realizes the effective dimensionality reduction of data and retains key diagnostic information.

[0052] In this embodiment, S23 includes the following sub-steps:

[0053] S231, calculating the average value of the amplitude in each zone to obtain the average amplitude;

[0054] S232, marking an amplitude greater than the average amplitude in a zone as a candidate amplitude;

[0055] S233, in one area, extracting the frequency value corresponding to the candidate amplitude, and calculating the mean of each extracted frequency value to obtain the dominant frequency;

[0056] S234. Construct a dominant frequency matrix from each dominant frequency.

[0057] The dominant frequency matrix is: , where f 1,1 is the first row and first column element in the dominant frequency matrix, f 1,2 is the element in the first row and second column of the dominant frequency matrix, f 1,3 is the element in the first row and third column of the dominant frequency matrix, f 1,4 is the element in the first row and fourth column of the dominant frequency matrix, f 2,1 is the element in the 2nd row and 1st column of the dominant frequency matrix, f 2,2 is the element in the second row and second column of the dominant frequency matrix, f 2,3 is the element in the 2nd row and 3rd column of the dominant frequency matrix, f 2,4 is the element in the 2nd row and 4th column of the dominant frequency matrix, f m,1 is the element in the mth row and first column of the dominant frequency matrix, f m,2 is the element in the mth row and second column of the dominant frequency matrix, f m,3 is the element in the mth row and third column of the dominant frequency matrix, f m,4 It is the element in the mth row and 4th column of the dominant frequency matrix.

[0058] The dominant frequency matrix has a total of 4 columns. The elements in the first column are all the dominant frequencies in the low frequency area, the elements in the second column are all the dominant frequencies in the medium frequency area, the elements in the third column are all the dominant frequencies in the high frequency area, and the elements in the fourth column are all the dominant frequencies in the ultra-high frequency area.

[0059] The present invention ensures that the frequency components with real significant characteristics are captured by marking the candidate values ​​higher than the average amplitude, and filters out the interference of noise and weak vibration; then the candidate frequency values ​​are averaged to obtain the dominant frequency, and the main vibration characteristics in the frequency band are retained.

[0060] In this embodiment, S3 includes the following steps:

[0061] S31, taking the average value of each column element of the dominant frequency matrix, and using the average value of each column element as an element to construct a dominant frequency mean sequence;

[0062] S32. Calculate the frequency change coefficient for each column element of the dominant frequency matrix, and use the frequency change coefficient of each column element as an element to construct a dominant frequency change sequence.

[0063] The formula for calculating the frequency variation coefficient in S32 is: , where μ i is the i-th frequency variation coefficient, f i,j is the jth element in the i-th column of the dominant frequency matrix, i and j are positive integers, and N is the number of elements in a column. In this embodiment, N is equal to m.

[0064] The present invention takes an average value of the elements in each column to obtain the mean value in each frequency range, which is reflected in the overall frequency level in each area of ​​the low frequency area, the medium frequency area, the high frequency area and the ultra-high frequency area. The frequency variation coefficient is calculated for each column element to reflect the degree of data dispersion. When the frequency variation coefficient is larger, the frequency changes faster and the stability is worse.

[0065] like Figure 2 As shown, the dual-channel feature fusion BP neural network in S4 includes: a first channel, a second channel, a channel coupling layer and a BP neural network;

[0066] The input end of the first channel is used to input the dominant frequency mean sequence; the input end of the second channel is used to input the dominant frequency change sequence; the input end of the channel fusion layer is respectively connected to the output end of the first channel and the output end of the second channel, and its output end is connected to the input end of the BP neural network; the output end of the BP neural network serves as the output end of the dual-channel feature fusion BP neural network.

[0067] In this embodiment, the first channel and the second channel both include: a tanh layer and a BN layer connected in sequence, wherein the expression of the tanh layer is: , where s n is the nth eigenvalue output by the tanh layer, x n is the nth element of the tanh layer input, w x,n For x n The weight of b x,n For x n The bias of , tanh is the hyperbolic tangent activation function, the BN layer is used to normalize the multiple eigenvalues ​​output by the tanh layer, and n is a positive integer.

[0068] In this embodiment, the expression of the channel coupling layer is:

[0069] ,

[0070] Among them, g n is the nth eigenvalue output by the channel coupling layer, g 1,n is the nth eigenvalue of the first channel output, g 2,n is the nth eigenvalue of the second channel output, w g1,n g 1,n The weight, w g2,n g 2,n The weight of b g1,n g 1,n The bias, b g2,n g 2,n The bias of the channel coupling layer is W1, W2 is the weight of the second channel, n is a positive integer, σ is the Sigmoid activation function, and the four eigenvalues ​​output by the channel coupling layer are used as the input of the BP neural network.

[0071] A single eigenvalue is processed in both the tanh layer and the BN layer. Therefore, in the channel coupling layer, the feature corresponding to the dominant frequency mean of the low-frequency area is fused with the feature corresponding to the dominant frequency change of the low-frequency area, the feature corresponding to the dominant frequency mean of the medium-frequency area is fused with the feature corresponding to the dominant frequency change of the medium-frequency area, the feature corresponding to the dominant frequency mean of the high-frequency area is fused with the feature corresponding to the dominant frequency change of the high-frequency area, and the feature corresponding to the dominant frequency mean of the ultra-high frequency area is fused with the feature corresponding to the dominant frequency change of the ultra-high frequency area, so that each value output by the channel coupling layer reflects the characteristics of a frequency zone.

[0072] In this embodiment, the weights and biases in the dual-channel feature fusion BP neural network can be trained using the existing gradient descent method.

[0073] The present invention captures the overall level and dynamic change characteristics of the vibration signal through two channels, provides a more comprehensive state representation, and uses the tanh activation function to effectively handle nonlinear relationships, improving the model's ability to fit complex vibration modes. The introduction of the BN layer realizes feature standardization, accelerates the network training process, and improves the stability and generalization ability of the model. The channel coupling layer of the present invention realizes the adaptive fusion of the features of the two channels through learnable weights, and can automatically adjust the importance of each channel according to different working conditions. The entire network structure supports end-to-end training, can automatically learn the optimal feature representation and fusion strategy, and reduces the reliance on artificial feature engineering.

[0074] Embodiment 2, a monitoring and early warning system for the working state of a six-sided top press, comprising: a Hankel matrix construction unit, a frequency matrix construction unit, a feature extraction unit and a state value prediction unit;

[0075] The Hankel matrix construction unit is used to collect vibration signals of the six-sided top press when the press head moves, and form a Hankel matrix;

[0076] The frequency matrix construction unit is used to construct a dominant frequency matrix according to the Hankel matrix;

[0077] The feature extraction unit is used to extract frequency feature values ​​from each column of the dominant frequency matrix, and construct a dominant frequency mean value sequence and a dominant frequency change sequence;

[0078] The state value prediction unit is used to process the dominant frequency mean value sequence and the dominant frequency change sequence by using a dual-channel feature fusion BP neural network to obtain the working state value of the six-sided top press.

[0079] The specific implementation process of Example 2 is the same as that of Example 1.

[0080] Embodiment 3 is a storage medium on which a program is stored. When the program is executed by a processor, the method in Embodiment 1 is implemented.

[0081] In embodiments 1, 2 and 3, monitoring and early warning according to the working status value of the six-sided top press can be carried out in the following manner:

[0082] Set three levels of warning thresholds: Normal state: working state value ≤ 0.3; Mild warning: 0.3 < working state value ≤ 0.6, indicating that the equipment has a slight abnormality and needs to be monitored more closely; Moderate warning: 0.6 < working state value ≤ 0.8, indicating that the equipment may have hidden dangers of failure and inspection and maintenance are recommended; Severe warning: working state value > 0.8, indicating that the equipment has a serious failure risk and needs to be shut down for maintenance immediately.

[0083] Specific warning scenario examples: When it is detected that the working status value of a pressure head suddenly rises from 0.2 to 0.45, the system issues a mild warning, prompting the operator to pay attention to the operating status of the pressure head; if the status value continues to rise to 0.75 in a short period of time, a moderate warning is triggered, and the system recommends that the equipment be inspected after the current working cycle ends; if the status value quickly climbs to 0.85, the system immediately issues a severe warning signal, indicating that an emergency shutdown is required to prevent equipment damage; Warning response measures: Mild warning: increase monitoring frequency and record abnormal data; Moderate warning: arrange professional personnel to inspect the equipment and prepare spare parts; Severe warning: immediately shut down for maintenance and replace any damaged parts.

[0084] This hierarchical early warning mechanism can detect equipment abnormalities in a timely manner, prevent major failures from occurring, and ensure the safe and stable operation of the equipment.

[0085] The present invention constructs a Hankel matrix for vibration signals, thereby capturing the dynamic characteristics of the equipment in actual operation and improving the accuracy of monitoring. Then, a dominant frequency matrix is ​​constructed based on the Hankel matrix to reflect the dominant frequencies in different signal segments. The constructed dominant frequency mean sequence reflects the average level of the dominant frequencies, and the constructed dominant frequency change sequence reflects the changes in the dominant frequencies of signals in different segments, thereby better capturing the working conditions of the six-sided top press. The dual-channel feature fusion BP neural network processing method can more effectively fuse the frequency mean sequence and the frequency change sequence, thereby achieving a comprehensive evaluation of the working status and improving the sensitivity to potential faults and the accuracy of diagnosis.

[0086] The present invention retains the time sequence characteristics of the signal by constructing the Hankel matrix, so that the constructed dominant frequency matrix can reflect the changes of the dominant frequencies in different time periods, can accurately capture the frequency changes of the vibration signal, and improves the monitoring accuracy.

[0087] 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 monitoring and early warning method for the working status of a six-sided top press, characterized in that: The following steps are involved: S1, collecting the vibration signal of the six-sided top press when the press head moves, and forming a Hankel matrix; S2. Construct the dominant frequency matrix based on the Hankel matrix; S3, extracting frequency eigenvalues ​​from each column of the dominant frequency matrix, and constructing a dominant frequency mean sequence and a dominant frequency change sequence; S4, using a dual-channel feature fusion BP neural network to process the dominant frequency mean sequence and the dominant frequency change sequence to obtain the working state value of the six-sided top press; The S2 comprises the following sub-steps: S21, performing Fourier transform on each row of the Hankel matrix to obtain spectrum information of each row; S22, dividing the frequency value in each row of the spectrum information into four areas, wherein the four areas include: a low frequency area, a medium frequency area, a high frequency area and an ultra-high frequency area; S23, extracting the dominant frequency in each area to form a dominant frequency matrix, wherein each row in the dominant frequency matrix includes four elements, the first element is the dominant frequency of the low frequency area, the second element is the dominant frequency of the medium frequency area, the third element is the dominant frequency of the high frequency area, and the fourth element is the dominant frequency of the ultra-high frequency area; The S23 comprises the following sub-steps: S231, calculating the average value of the amplitude in each zone to obtain the average amplitude; S232, marking an amplitude greater than the average amplitude in a zone as a candidate amplitude; S233, in one area, extracting the frequency value corresponding to the candidate amplitude, and calculating the mean of each extracted frequency value to obtain the dominant frequency; S234, forming a dominant frequency matrix from each dominant frequency; The S3 comprises the following steps: S31, taking the average value of each column element of the dominant frequency matrix, and using the average value of each column element as an element to construct a dominant frequency mean sequence; S32, calculating the frequency change coefficient for each column element of the dominant frequency matrix, and using the frequency change coefficient of each column element as an element to construct a dominant frequency change sequence; The formula for calculating the frequency variation coefficient in S32 is: , where μ i is the i-th frequency variation coefficient, f i,j is the jth element in the i-th column of the dominant frequency matrix, where i and j are positive integers and N is the number of elements in a column.

2. The monitoring and early warning method for the working state of a six-sided top press according to claim 1 is characterized in that: The dual-channel feature fusion BP neural network in S4 includes: a first channel, a second channel, a channel coupling layer and a BP neural network; The input end of the first channel is used to input the dominant frequency mean sequence; the input end of the second channel is used to input the dominant frequency change sequence; the input end of the channel fusion layer is respectively connected to the output end of the first channel and the output end of the second channel, and its output end is connected to the input end of the BP neural network; the output end of the BP neural network serves as the output end of the dual-channel feature fusion BP neural network.

3. The monitoring and early warning method for the working state of a six-sided top press according to claim 2 is characterized in that: The first channel and the second channel both include: a tanh layer and a BN layer connected in sequence, wherein the expression of the tanh layer is: , where s n is the nth eigenvalue output by the tanh layer, x n is the nth element of the tanh layer input, w x,n For x n The weight of b x,n For x n The bias of , tanh is the hyperbolic tangent activation function, the BN layer is used to normalize the multiple eigenvalues ​​output by the tanh layer, and n is a positive integer.

4. The monitoring and early warning method for the working state of a six-sided top press according to claim 2 is characterized in that: The expression of the channel coupling layer is: ， Among them, g n is the nth eigenvalue output by the channel coupling layer, g 1,n is the nth eigenvalue of the first channel output, g 2,n is the nth eigenvalue of the second channel output, w g1,n g 1,n The weight, w g2,n g 2,n The weight of b g1,n g 1,n The bias, b g2,n g 2,n The bias of the channel coupling layer is W1, W2 is the weight of the second channel, n is a positive integer, σ is the Sigmoid activation function, and the four eigenvalues ​​output by the channel coupling layer are used as the input of the BP neural network.

5. A monitoring and early warning system for the working status of a six-sided top press, characterized in that: The monitoring and early warning method for the working state of the six-sided top press according to any one of claims 1 to 4 is implemented, characterized in that it includes: a Hankel matrix construction unit, a frequency matrix construction unit, a feature extraction unit and a state value prediction unit; The Hankel matrix construction unit is used to collect vibration signals of the six-sided top press when the press head moves, and form a Hankel matrix; The frequency matrix construction unit is used to construct a dominant frequency matrix according to the Hankel matrix; The feature extraction unit is used to extract frequency feature values ​​from each column of the dominant frequency matrix, and construct a dominant frequency mean value sequence and a dominant frequency change sequence; The state value prediction unit is used to process the dominant frequency mean value sequence and the dominant frequency change sequence by using a dual-channel feature fusion BP neural network to obtain the working state value of the six-sided top press.

6. A storage medium having a program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images