Interference elimination regulation and control system and method for bounce detection

Abstract

The invention belongs to the technical field of signal processing, and discloses an interference elimination regulation and control system and method for bounce detection. The method comprises the following steps: collecting an equipment vibration signal, and preprocessing the equipment vibration signal to obtain m sections of input signals; performing discretization processing on the m sections of input signals to obtain a signal state matrix; segmenting the signal state matrix to obtain a plurality of potential interference segments; determining an optimal interference elimination parameter based on the plurality of potential interference segments; and performing interference elimination processing on the potential interference section by using the optimal interference elimination parameter, and outputting an interference elimination regulation and control signal, so that a real jumping signal is clearer, and the accuracy and reliability of detection are improved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and more specifically, to an interference elimination control system and method for beat detection. Background Art

[0002] With the rapid development of modern industry, the condition monitoring and fault diagnosis of mechanical equipment have become increasingly important. Vibration signal analysis is a commonly used condition monitoring and fault diagnosis method. By analyzing the vibration signals generated by the equipment during operation, potential faults of the equipment can be discovered, the occurrence of faults can be prevented, and the reliability and service life of the equipment can be improved.

[0003] In complex industrial environments, equipment vibration signals are often affected by a variety of interference sources, such as environmental noise, electromagnetic interference, vibration of adjacent equipment, etc., which seriously affect the accuracy of detection. Traditional methods lack targeted processing of different types of interference, resulting in high false alarm rates and low reliability. For example, in a large manufacturing workshop, in a complex vibration scenario generated by the simultaneous operation of multiple devices, it is difficult for existing technologies to accurately identify abnormal vibrations of a single device. In addition, the vibration characteristics of equipment under different working conditions vary significantly, and existing technologies lack adaptive capabilities and are difficult to adjust detection strategies according to real-time working conditions. This is particularly evident in equipment with variable speeds, such as wind turbines, which have large variations in vibration characteristics at different wind speeds. At the same time, existing technologies have low computational efficiency when processing long time series data and are difficult to meet the needs of real-time monitoring. This has caused serious problems for key equipment that requires 24-hour uninterrupted monitoring. In addition, existing methods often require a lot of manual intervention to adjust parameters, which increases the difficulty of operation and the possibility of human error.

[0004] In view of this, the present invention proposes an interference elimination control system and method for jitter detection to solve the above-mentioned problem. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: an interference elimination control system for vibration detection, comprising: a signal acquisition and processing module, for acquiring a vibration signal of the device, and preprocessing the vibration signal of the device to obtain m-segment input signals; A discrete processing module is used to discretize the m-segment input signal to obtain a signal state matrix; A segmentation module is used to segment the signal state matrix to obtain a number of potential interference segments; The parameter fitting module determines the optimal interference elimination parameters based on several potential interference segments; uses the optimal interference elimination parameters to perform interference elimination processing on the potential interference segments, and outputs interference elimination control signals to the control terminal.

[0006] Furthermore, the method of preprocessing the device vibration signal includes: Perform preliminary processing on the equipment vibration signal to obtain a preliminary accurate signal, remove noise in the preliminary accurate signal to obtain a preliminary input signal, standardize or normalize the preliminary input signal to obtain a standardized vibration signal; divide the standardized vibration signal into m time periods to obtain m input signals; m is an integer greater than 1.

[0007] Furthermore, the method of performing preliminary processing on the device vibration signal includes: Collect several wavelet basis function families, construct wavelet basis functions of different orders for each wavelet basis function family to form a wavelet basis function subset, classify and encode all wavelet basis functions in the wavelet basis function subset according to the family and order to form a wavelet basis function library; Define a time period, extract a signal of a time period from any position in the equipment vibration signal as the scoring signal; traverse each wavelet basis function in the wavelet basis function library, perform wavelet decomposition on the scoring signal x(n), and obtain the approximate component A and the detail component D; where n is the index of the discrete sampling point; based on the approximate component A and the detail component D obtained by wavelet decomposition, calculate the comprehensive score of the corresponding wavelet basis function; the calculation formula for the comprehensive score is: ; in, is the approximate entropy value of the approximate component A, is the scoring function of energy distribution, is the energy distribution, i.e., the set of energy value vectors of each scale component; is the energy retention rate; is the feature consistency scoring function, is the weight parameter; is the weight of the scoring function for energy distribution, is the fast Fourier transform, represents inverse fast Fourier transform; Scoring function for energy distribution ;in, is the maximum order of wavelet decomposition, For the The normalized energy of each component; The wavelet basis function with the lowest comprehensive score in the wavelet basis function library is used as the best matching wavelet basis. For the best matching wavelet basis, the wavelet entropy threshold Tapen and the energy threshold en are set; Using the best matching wavelet basis, the equipment vibration signal is decomposed by level I wavelet to obtain the corresponding level I approximate component A(I) and level I detail component D(I); the wavelet entropy Apen(A(I)) of the level I approximate component A(I) is calculated; and the energy loss rate is calculated; the value of I is continuously adjusted until Apen(A(I)) is less than Tapen and the energy loss rate is less than en; the value of I obtained at this time is the optimal series; the optimal series is denoted as q; using the best matching wavelet basis and the optimal series, the equipment vibration signal is decomposed by level q wavelet to obtain the low-frequency approximate component A(q); the original equipment vibration signal is subtracted from the low-frequency approximate component A(q) point by point to obtain a preliminary accurate signal.

[0008] Furthermore, the method for obtaining the feature consistency scoring function includes: Extract time domain features and frequency domain features from the scoring signal x(n) and construct a signal feature vector X; the time domain features include mean, root amplitude, peak value and peak factor; the frequency domain features include power spectrum features, frequency center and bandwidth; extract the same time domain features and frequency domain features from the approximate component A and detail component D after wavelet decomposition, and construct corresponding feature vectors respectively; the feature vector corresponding to the approximate component A is recorded as the approximate feature vector A', and the feature vector corresponding to the detail component D is recorded as the detail feature vector D'; Collect several groups of labeled vibration signals and their corresponding fault modes; extract three groups of vectors X, A', and D' from each vibration signal, use X as input and the fault mode as a label, train a supervised learning model, and evaluate the correlation score S(X) between X and the fault mode by the trained supervised learning model; Input A' and D' into the trained supervised learning model to obtain the relevance scores respectively and ; Calculate feature consistency scoring function based on S(X), S(A') and S(D') ;in, is the fusion function, which is weighted average or product.

[0009] Furthermore, the method of dividing the standardized vibration signal into m time periods includes: Defines the length of a time window ; Preset overlap ratio ; ; in, and is the adjustment parameter, is the first-order derivative of the normalized vibration signal in the current time window, is the standardized vibration signal in the current time window; is the change response parameter; is the index of the sampling point of the normalized vibration signal; Starting from the starting position of the standardized vibration signal, extract the first time window of length L; according to the overlap rate , calculate the starting position of the next time window, and repeat until the entire standardized vibration signal is covered, and obtain m time windows, that is, m time periods.

[0010] Furthermore, the method of discretizing the m-segment input signal includes: The input signal is defined as a one-dimensional lattice chain, and each lattice point in the one-dimensional lattice chain corresponds to a sampling point of the input signal; the spin value range of the lattice points and the interaction rules between the lattice points are defined; the interaction rules between the lattice points are to calculate the total interaction energy and minimize the total interaction energy when simulating evolution; The total interaction energy is calculated as: ; in, is the total interaction energy, is the preset nearest neighbor interaction strength parameter, For the The spin state value of each grid point; For the The spin state of each lattice point; is the preset interaction benchmark strength, For the and The distance between the grid points, is the correlation length parameter; is the intensity parameter of the external magnetic field; According to the value of the input signal, the initial spin state of each grid point is assigned a value, and a linear or nonlinear mapping relationship is used to map the value of the continuous input signal to the value of the discrete spin state; through simulated evolution, the one-dimensional lattice chain converges to the ground state; When the one-dimensional lattice chain converges to the ground state, the current spin states of all lattice points are recorded, and the spin states are arranged in chronological order to form a signal state matrix M1; each row of the signal state matrix M1 corresponds to a time step, and each column corresponds to a lattice point.

[0011] Furthermore, the method of making the one-dimensional lattice chain converge to the ground state by simulating evolution includes: Set an initial temperature T0; traverse each grid point in chronological order. grid points, calculate their spin states Energy changes before and after flipping ;in, is the total interaction energy after flipping, is the total interaction energy before flipping; For the The current spin state of each grid point The spin state after flipping; the total interaction energy before and after flipping is calculated and obtained according to the interaction rules; like , then accept the flip; if , then with probability Accept flip, where is the temperature of the current round; repeat a round of traversal of all grid points, and reduce the temperature after each N1 round; the formula for reducing the temperature is: ;in, is the basic cooling rate, is the current number of cooling steps, is the cooling index; is the oscillation intensity parameter, and ; is the oscillation angular frequency, is the initial phase; In the process of lowering the temperature, the total interaction energy is continuously calculated. When the change in the total interaction energy within M consecutive cooling steps is less than the preset change threshold, the one-dimensional lattice chain has converged to the ground state.

[0012] Furthermore, the method of segmenting the signal state matrix includes: The signal state matrix M1 is regarded as a two-dimensional rectangular area. Initially, M1 is divided into A small rectangular block, is the initial preset segmentation granularity; Create a complete binary tree Y, where the leaf nodes of the complete binary tree Y correspond to small rectangular blocks; each internal node represents the union of the rectangular areas covered by its two child nodes; the root node of the complete binary tree Y corresponds to the entire signal state matrix M1; For each leaf node, calculate its node feature vector; the node feature vector is a joint vector composed of mean, variance, peak, wavelet energy, wavelet coefficient and Fourier coefficient; Preset the similarity threshold and calculate the Mahalanobis distance between each node in the complete binary tree Y and its sibling nodes; Starting from the bottom layer of the complete binary tree Y, traverse the pairs of sibling nodes at each layer; if the Mahalanobis distance of a pair of sibling nodes is greater than the similarity threshold, merge them into a parent node, continue to traverse to the upper layer, and repeat the merging until no further merging is possible; retrieve all unmerged nodes in the complete binary tree Y, use them as potential nodes, and map a rectangular area corresponding to each potential node back to the space of the original signal state matrix to obtain the corresponding potential interference segment.

[0013] Further, the method of determining the optimal interference elimination parameter includes: A hierarchical membrane structure is constructed, the outermost layer of the membrane structure is the skin membrane, and the inside of the membrane structure has several nested regional membranes; the signal characteristics of the potential interference segment are encoded as material objects and placed in the skin membrane; the signal characteristics include wavelet packet decomposition coefficients, harmonic components and spectral entropy; the encoding adopts binary encoding; Each regional membrane represents a type of interference exclusion parameter; a material object is placed in each regional membrane, and the material object represents a candidate value of the corresponding type of interference exclusion parameter; The membrane evolution rules are defined to include object mutation rules, object selection rules, object transfer rules and stop rules. The object mutation rules are used to perform mutation operations on material objects in the regional membrane to generate new objects; the object selection rules are used to perform cross operations on two material objects in the regional membrane to generate new objects; the object transfer rules are used to transfer material objects between different membranes; the stop rule is that the evolution in the regional membrane stops when the maximum number of iterations is reached; an output interface is set to read the values ​​of various interference exclusion parameters from the membrane structure after each iteration; The evolution rule is defined as randomly perturbing the material objects of the regional membrane in each iteration, calculating the value of the objective function, and retaining the material object with the smallest value of the objective function; The formula of the objective function is: ; in, For the The value of the objective function at the iteration, is the potential interference segment, is the index of the signal value within the potential interference segment; In order to process the potential interference segment by using the values ​​of various interference elimination parameters read from the membrane structure during iteration, the processed potential interference segment is obtained; is the covariance calculation function; is the standard deviation of the original potential interference segment, For the The standard deviation of the potential interference segment after processing at the iteration; For the iteration Class interference exclusion parameters; For the iteration weights of class interference exclusion parameters; is a nonlinear parameter; According to the membrane evolution rules and the evolution rules, the material objects are exchanged and evolved between the regional membranes, and each iteration generates a new combination of material objects; the value of the objective function and the corresponding material objects of each iteration are recorded; after each iteration, it is checked whether the stopping rule is satisfied. If so, a combination of material objects that minimizes the value of the objective function is selected from all the regional membranes, and the interference exclusion parameters corresponding to this group of material objects are used as the optimal interference exclusion parameters.

[0014] An interference elimination and control method for beat detection is implemented based on the interference elimination and control system for beat detection, comprising: S1, collecting equipment vibration signals, and preprocessing the equipment vibration signals to obtain m-segment input signals; S2, discretize the m-segment input signal to obtain a signal state matrix; S3, segmenting the signal state matrix to obtain several potential interference segments; S4. Based on several potential interference segments, determine the optimal interference elimination parameters; use the optimal interference elimination parameters to perform interference elimination processing on the potential interference segments, and output the interference elimination control signal to the control terminal.

[0015] The technical effects and advantages of the interference elimination control system and method for jitter detection of the present invention are as follows: The present invention can intelligently adapt to different types of vibration signals and interference environments, greatly improving the accuracy and reliability of vibration detection; through adaptive preprocessing and dynamic segmentation, it can more accurately capture signal characteristics and effectively reduce the false alarm rate; the multi-parameter collaborative optimization mechanism significantly improves the interference elimination effect, so that stable and efficient performance can be maintained under complex working conditions; in addition, the self-learning ability can continuously optimize the processing strategy, and the detection effect will continue to improve with the increase of usage time; this can not only reduce maintenance costs and extend the service life of equipment, but also provide more reliable data support, thereby improving production efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of an interference elimination and control system for beat detection of the present invention; Figure 2 A schematic diagram of an interference elimination and control method for beat detection according to the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work are within the scope of protection of the present invention.

[0018] Example 1 See also Figure 1 As shown, the interference elimination control system for vibration detection described in this embodiment includes: a signal acquisition and processing module, which is used to collect device vibration signals and pre-process the device vibration signals to obtain m-segment input signals; A discrete processing module is used to discretize the m-segment input signal to obtain a signal state matrix; A segmentation module is used to segment the signal state matrix to obtain a number of potential interference segments; The parameter fitting module determines the optimal interference elimination parameters based on several potential interference segments; uses the optimal interference elimination parameters to perform interference elimination processing on the potential interference segments, and outputs the interference elimination control signal to the control terminal, so that the real beating signal is clearer, thereby improving the accuracy and reliability of detection; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0019] Ways to collect equipment vibration signals include: Vibration sensors, such as acceleration sensors, velocity sensors or displacement sensors, are installed on the equipment that needs to be tested for vibration. The vibration sensor converts the vibration generated by the equipment during operation into analog electrical signals or digital electrical signals. The original analog electrical signals or digital electrical signals collected by the vibration sensor are quantified through a data acquisition card or a microcontroller to obtain discrete digital vibration signals, which are the equipment vibration signals.

[0020] The methods for preprocessing the equipment vibration signal include: Perform preliminary processing on the equipment vibration signal to obtain a preliminary accurate signal, remove noise in the preliminary accurate signal to obtain a preliminary input signal, standardize or normalize the preliminary input signal to obtain a standardized vibration signal; divide the standardized vibration signal into m time periods to obtain m input signals; m is an integer greater than 1.

[0021] The methods for preliminary processing of equipment vibration signals include: Collect several wavelet basis function families, such as Daubechies, Coiflet, Symlet, Biorthogonal, etc.; for each wavelet basis function family, construct wavelet basis functions of different orders (such as 2-10 orders) to form a wavelet basis function subset, classify and encode all wavelet basis functions in the wavelet basis function subset according to family and order to form a wavelet basis function library.

[0022] Define a time period (a random time step with a length greater than 1 time step), extract a signal of a time period from any position in the equipment vibration signal as a scoring signal; traverse each wavelet basis function in the wavelet basis function library, perform wavelet decomposition on the scoring signal x(n) to obtain the approximate component A and the detail component D; where n is the index of the discrete sampling point; based on the approximate component A and the detail component D obtained by wavelet decomposition, calculate the comprehensive score of the corresponding wavelet basis function; the calculation formula for the comprehensive score is: ; in, is the approximate entropy value of the approximate component A, is the scoring function of energy distribution (reflecting the distribution of signal energy among components of each scale after wavelet decomposition), is the energy distribution, i.e., the set of energy value vectors of each scale component; The energy retention rate is obtained by summing the energy of each component and normalizing it with the energy of the original scoring signal; is the feature consistency scoring function, is the weight parameter, controlling the feature consistency scoring function Weight in the overall rating; is the weight of the scoring function for energy distribution, is the fast Fourier transform, represents the inverse fast Fourier transform, Represents the weight Perform a fast Fourier transform and By converting from the time domain to the frequency domain, the frequency domain convolution method can be calculated more efficiently without the need to perform computationally intensive convolution operations directly in the time domain, which can greatly improve computing efficiency.

[0023] It should be noted that the calculation process of the approximate entropy value is: Sample the approximate component A to obtain a time series data, divide the time series data into multiple small segments in a sliding window manner, each small segment contains m3 continuous data points, and form an m3-dimensional vector; calculate the maximum difference between any two m3-dimensional vectors, that is, find the maximum absolute difference between the corresponding dimensions of the two vectors in these m3 dimensions; set a similarity tolerance, and for each m3-dimensional vector, count how many other m3-dimensional vectors have a maximum difference with it that is less than this similarity tolerance, and obtain a count value; divide this count value by the total number of m3-dimensional vectors to obtain the m3-dimensional vector Similarity; calculate the mean of the similarities of all m3-dimensional vectors, which is the average similarity in the m3-dimensional case; then increase the dimension to m3+1, repeat the above process, and get the average similarity in the m3+1-dimensional case; the approximate entropy value is the limit of the difference between the average similarity in the m3-dimensional case and the average similarity in the m3+1-dimensional case; in actual calculation, due to the limited length of data, the statistics of finite-length data are used to approximate the value of this limit; the larger the approximate entropy value, the more random, irregular, and complex the data points in the time series are; conversely, the smaller the approximate entropy value, the more regular the data points are and the lower the complexity is.

[0024] Scoring function for energy distribution ;in, is the maximum order of wavelet decomposition, For the The normalized energy of the components.

[0025] Feature consistency scoring function Ways to obtain include: The time domain features and frequency domain features of the scoring signal x(n) are extracted and constructed into a signal feature vector X (vector-level splicing); the time domain features include mean, root amplitude, peak value and peak factor; the frequency domain features include power spectrum features, frequency center and bandwidth.

[0026] The same time domain features and frequency domain features are extracted from the approximate component A and the detail component D after wavelet decomposition, and the corresponding feature vectors are constructed respectively; the feature vector corresponding to the approximate component A is recorded as the approximate feature vector A', and the feature vector corresponding to the detail component D is recorded as the detail feature vector D'.

[0027] Collect several groups of labeled vibration signals (samples) and their corresponding fault modes (fault types or operating conditions corresponding to the samples); extract three groups of vectors X, A', and D' for each vibration signal, use X as input and the fault mode as a label, and train a supervised learning model (such as an artificial neural network). The trained supervised learning model can evaluate the correlation score S(X) between X and the fault mode.

[0028] Input A' and D' into the trained supervised learning model to obtain the relevance scores respectively and ; Based on S(X), S(A') and S(D'); calculate the feature consistency score function ;in, is the fusion function, which is weighted average or product.

[0029] Through the above steps, we can obtain the feature consistency scoring function FC(A, D) based on machine learning. This function can intelligently evaluate the degree of preservation of A and D after wavelet decomposition and the original signal in key features, thereby quantifying the separation effect of the wavelet basis on signal features; The higher the value, the better the selected wavelet basis can separate the main features of the original signal into A and D without causing distortion or distortion of the features.

[0030] The wavelet basis function with the lowest comprehensive score in the wavelet basis function library is used as the best matching wavelet basis. For the best matching wavelet basis, the wavelet entropy threshold Tapen and the energy threshold en are set; the best matching wavelet basis is used to perform I-level wavelet decomposition on the equipment vibration signal to obtain the corresponding I-level approximate component A(I) and I-level detail component D(I); the wavelet entropy Apen(A(I)) of the I-level approximate component A(I) is calculated; and the energy loss rate (1 minus the energy retention rate) is calculated; the value of I is continuously adjusted until Apen(A(I)) is less than Tapen and the energy loss rate is less than en; the value of I obtained at this time is the optimal series; the optimal series is denoted as q; the best matching wavelet basis and the optimal series are used to perform q-level wavelet decomposition on the equipment vibration signal to obtain the low-frequency approximate component A(q); the original equipment vibration signal is subtracted from the low-frequency approximate component A(q) point by point to obtain a preliminary accurate signal.

[0031] It should be noted that the calculation method of wavelet entropy is: Perform wavelet decomposition on the level I approximate component A(I) to obtain a set of wavelet coefficients; calculate the probability distribution of this set of wavelet coefficients, that is, sort the wavelet coefficients by size and divide them into several intervals (such as equal spacing or logarithmic spacing, etc.); count the number of wavelet coefficients n_g in each interval g; calculate the probability of each interval p_g=n_g / N3, where N3 is the total number of wavelet coefficients; then the wavelet entropy ; Wavelet entropy reflects the degree of disorder in the distribution of wavelet coefficients.

[0032] Ways to remove noise from the preliminary accurate signal include: The preliminary accurate signal is segmented into several overlapping time windows. For each time window, its statistical characteristics are calculated, including mean, variance, peak factor, kurtosis and skewness; the statistical characteristics are concatenated into a vector to obtain a statistical feature vector.

[0033] Based on statistical feature vectors, a noise detection model is constructed, such as a Gaussian mixture model (GMM), a one-class support vector machine (One-Class SVM), etc.; the noise detection model is trained using a large amount of labeled noise / non-noise data during the training phase; test data is set (some unseen non-noise data is prepared), and during the training process, the performance indicators of the noise detection model on the test data, such as accuracy, precision, recall, etc., are calculated; according to the performance indicators, the parameters of the noise detection model are adjusted, and the noise detection model is repeatedly trained; until the performance indicators of the noise detection model meet the requirements, a trained noise detection model is obtained.

[0034] For each time window segment, use the trained noise detection model to determine whether the corresponding time window segment is noise. If it is noise, the corresponding time window segment will be subjected to noise removal processing. If it is not noise, the corresponding time window segment will be retained.

[0035] The method for noise removal is wavelet threshold denoising, Wiener filtering or adaptive filtering; the appropriate denoising algorithm is selected according to the noise type; all the time window segments that have been processed by noise removal and the retained time window segments are spliced ​​together in the order of segmentation to obtain the preliminary input signal.

[0036] The methods for dividing the normalized vibration signal into m time periods include: Defines the length of a time window , such as 1 second, 5 seconds or 10 seconds, etc.; preset overlap rate ; ; in, and is a tuning parameter used to control the range and rate of change of the overlap rate. is the first-order derivative of the normalized vibration signal in the current time window, is the standardized vibration signal in the current time window; is a variable response parameter used to adjust the response characteristics of the overlap rate to the signal change rate; is the index of the sampling point of the normalized vibration signal.

[0037] Starting from the starting position of the standardized vibration signal, extract the first time window of length L; according to the overlap rate , calculate the starting position of the next time window; for example, if the overlap rate is 25%, the starting position of the next time period is the end position of the current time period minus 0.25×L, and repeat until the entire standardized vibration signal is covered, and m time windows are obtained, that is, m time periods.

[0038] The methods for discretizing m segments of input signals include: The input signal is defined as a one-dimensional lattice chain, and each lattice point in the one-dimensional lattice chain corresponds to a sampling point of the input signal; the spin value range of the lattice points and the interaction rules between the lattice points are defined, and the spin value range is binary (-1, 1) or multi-valued (-L, ..., 0, ..., L); the interaction rule between the lattice points is to calculate the total interaction energy and minimize the total interaction energy when simulating the evolution.

[0039] The total interaction energy is calculated as: ; in, is the total interaction energy, is the preset nearest neighbor interaction strength parameter, usually a positive value, For the The spin state value of each grid point can be binary (-1, 1) or multi-valued (-L, ..., 0, ..., L); For the The spin states of the lattice points (adjacent lattice points); is the preset interaction benchmark strength, For the and The distance between grid points (the absolute difference in the index values ​​of the grid points), is the correlation length parameter, which controls the decay rate of long-range interactions; is the strength parameter of the external magnetic field, when When , the contribution in the positive spin direction is negative, and the external field will tend to pull the positive spin; by adjusting the parameters, the weights of different interaction terms can be controlled to simulate different characteristics and correlations in the input signal.

[0040] According to the value of the input signal, the initial spin state of each lattice point is assigned, and the value of the continuous input signal is mapped to the value of the discrete spin state using a linear or nonlinear mapping relationship; the one-dimensional lattice chain is converged to the ground state through simulated evolution; specifically, an initial temperature T0 is set, and the temperature T0 is initially set to a higher value, such as several times the energy. The high temperature can make the lattice chain in a disordered state and the spin state randomly distributed.

[0041] Traverse each grid point in chronological order. grid points, calculate their spin states Energy changes before and after flipping ;in, is the total interaction energy after flipping, is the total interaction energy before flipping; For the The current spin state of each grid point The spin state after flipping; the total interaction energy before and after flipping is calculated according to the interaction rules, and flipping refers to changing the value of the spin state.

[0042] like , then the flip is accepted, reducing the total interaction energy; if , then with probability Accept flip, where is the temperature of the current round; repeat a round of traversal of all grid points, and reduce the temperature after each N1 round; the formula for reducing the temperature is: ;in, is the basic cooling rate, is the current number of cooling steps, is the cooling index, which can adjust the steepness of the cooling curve; , the cooling curve is steep, and the temperature drop rate is fast at first and then slow; , the cooling curve is gentle, and the temperature drop rate is slow at first and then fast; is the oscillation intensity parameter, and ; is the oscillation angular frequency, is the initial phase; adjustable and angular frequency , controlling the amplitude and frequency of temperature oscillation to make it closer to the actual system; the initial phase makes the starting point of each cooling process different, increasing the randomness and avoiding convergence to the same minimum solution.

[0043] In the process of lowering the temperature, the total interaction energy is continuously calculated. When the change value of the total interaction energy within M consecutive cooling steps is less than the preset change threshold, the one-dimensional lattice chain converges to the ground state; when the one-dimensional lattice chain converges to the ground state, the current spin states of all grid points are recorded, and the spin states are arranged in chronological order to form a signal state matrix M1; each row of the signal state matrix M1 corresponds to a time step, and each column corresponds to a grid point.

[0044] The ways to partition the signal state matrix include: The signal state matrix M1 is regarded as a two-dimensional rectangular area. Initially, M1 is divided into A small rectangular block, is the initial preset segmentation granularity; a complete binary tree Y is created, and the leaf nodes of the complete binary tree Y correspond to small rectangular blocks; each internal node represents the union of the rectangular areas covered by its two child nodes; the internal node refers to the non-leaf node in the complete binary tree Y, and each internal node has a left child node and a right child node; the rectangular area covered by the internal node is the union of the rectangular areas covered by its left child node and the right child node; the node feature vector of the internal node is obtained by fusing the node feature vectors of its two child nodes (such as taking the average); the root node of the complete binary tree Y corresponds to the entire signal state matrix M1.

[0045] For each leaf node (small rectangular block), calculate its node feature vector; the node feature vector is a joint vector composed of mean, variance, peak, wavelet energy, wavelet coefficient and Fourier coefficient; preset a similarity threshold, and for each node in the complete binary tree Y, calculate the Mahalanobis distance between it and its brother node; the left child node and the right child node of the same parent node are a pair of brother nodes.

[0046] Starting from the bottom layer of the complete binary tree Y (the layer where all leaf nodes are located), each pair of adjacent nodes in the bottom layer is a pair of sibling nodes; traverse the sibling node pairs of each layer; if the Mahalanobis distance of a pair of sibling nodes is greater than the similarity threshold, merge them into a parent node, continue to traverse to the upper layer, and repeat the merging until no further merging is possible.

[0047] Retrieve all unmerged nodes in the complete binary tree Y and use them as potential nodes. Map a rectangular area corresponding to each potential node back to the space of the original signal state matrix to obtain the corresponding potential interference segment.

[0048] Methods for determining the optimal interference rejection parameters include: A hierarchical membrane structure is constructed, the outermost layer of the membrane structure is the skin membrane, and there are several nested regional membranes inside the membrane structure; the signal characteristics of the potential interference segment are encoded as material objects and placed in the skin membrane; the signal characteristics include wavelet packet decomposition coefficients, harmonic components and spectral entropy; the encoding can be binary encoding.

[0049] Each regional membrane represents a type of interference exclusion parameter, such as filter parameters, threshold parameters, time window parameters, etc.; a material object is placed in each regional membrane, and the material object represents a candidate value of the corresponding type of interference exclusion parameter; for example, the material object in the regional membrane corresponding to the filter parameter may be {low pass, high pass, band pass...etc.}.

[0050] The membrane evolution rules are defined to include object mutation rules, object selection rules, object transfer rules and stop rules. The object mutation rules are used to perform mutation operations on material objects within the regional membrane to generate new objects; for example, under a certain probability, a low-pass filter is mutated into a high-pass filter; the object selection rules are used to perform cross operations on two material objects within the regional membrane to generate new objects; for example, the parameters of two wavelet filters of different orders are crossed; the object transfer rules are used to transfer material objects between different membranes to achieve information interaction; for example, the material objects in the regional membrane corresponding to the filter parameters are copied to the regional membrane corresponding to the threshold parameters; the stop rule is that the evolution within the regional membrane stops when the maximum number of iterations is reached; an output interface is set to read the values ​​of various interference exclusion parameters from the membrane structure after each iteration.

[0051] The evolution rule is defined as randomly perturbing the material objects of the regional membrane in each iteration, calculating the value of the objective function, and retaining the material objects with better (smallest) values ​​of the objective function; the formula of the objective function is: ; in, For the The value of the objective function at the iteration, is the potential interference segment, is the index of the signal value within the potential interference segment; In order to process the potential interference segment by using the values ​​of various interference elimination parameters read from the membrane structure during iteration, the processed potential interference segment is obtained; is the covariance calculation function; is the standard deviation of the original potential interference segment, For the The standard deviation of the potential interference segment after processing at the iteration; For the iteration Class interference exclusion parameters; For the iteration weights of class interference exclusion parameters; is a nonlinear parameter that controls the nonlinearity of the parameter term.

[0052] According to the membrane evolution rules and the evolution rules, the material objects are exchanged and evolved between the regional membranes, and each iteration generates a new combination of material objects; the value of the objective function and the corresponding material objects of each iteration are recorded; after each iteration, it is checked whether the stopping rule is satisfied. If so, a combination of material objects that minimizes the value of the objective function is selected from all the regional membranes, and the interference exclusion parameters corresponding to this group of material objects are used as the optimal interference exclusion parameters.

[0053] The method of performing interference elimination processing on the potential interference segment is: traverse all potential interference segments, extract the signal of this segment according to the time window parameter, filter the extracted signal using the filter parameter, perform wavelet threshold denoising or other threshold processing using the threshold parameter, and then obtain the processed signal of the current potential interference segment.

[0054] Initialize an empty time series to store the final interference elimination control signal. Splice the processed signal segments of each potential interference segment into the time series in chronological order. Use a smoothing window function to process the transition region between the potential interference segments to ensure the continuity of the signal. Alternatively, directly consider the transition region in the above iterative process, that is, use the transition processing parameters as a type of interference elimination parameters. After completing the merging of all processed signal segments, output the final time series, which is the interference elimination control signal.

[0055] This embodiment can intelligently adapt to different types of vibration signals and interference environments, greatly improving the accuracy and reliability of vibration detection; through adaptive preprocessing and dynamic segmentation, it can more accurately capture signal characteristics and effectively reduce the false alarm rate; the multi-parameter collaborative optimization mechanism significantly improves the interference elimination effect, so that stable and efficient performance can be maintained under complex working conditions; in addition, the self-learning ability can continuously optimize the processing strategy, and the detection effect will continue to improve with the increase of usage time; this can not only reduce maintenance costs and extend the service life of equipment, but also provide more reliable data support, thereby improving production efficiency and economic benefits.

[0056] Example 2 See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, and a method for interference elimination and control for beating detection is provided, including: S1, collecting equipment vibration signals, and preprocessing the equipment vibration signals to obtain m-segment input signals; S2, discretize the m-segment input signal to obtain a signal state matrix; S3, segmenting the signal state matrix to obtain several potential interference segments; S4. Based on several potential interference segments, determine the optimal interference elimination parameters; use the optimal interference elimination parameters to perform interference elimination processing on the potential interference segments, and output the interference elimination control signal to the control terminal.

[0057] Example 3 This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the interference elimination and control method for beat detection provided above is implemented.

[0058] Since the electronic device introduced in this embodiment is an electronic device used to implement an interference elimination control method for beating detection in the embodiment of the present application, based on the interference elimination control method for beating detection introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as a person skilled in the art implements an electronic device used in an interference elimination control method for beating detection in the embodiment of the present application, it belongs to the scope of protection of the present application.

[0059] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0060] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An interference elimination control system for jitter detection, characterized in that: include: The signal acquisition and processing module is used to acquire the vibration signal of the equipment and pre-process the vibration signal of the equipment to obtain m-segment input signals; A discrete processing module is used to discretize the m-segment input signal to obtain a signal state matrix; A segmentation module is used to segment the signal state matrix to obtain a number of potential interference segments; The parameter fitting module determines the optimal interference elimination parameters based on several potential interference segments; uses the optimal interference elimination parameters to perform interference elimination processing on the potential interference segments, and outputs interference elimination control signals to the control terminal.

2. The interference elimination control system for jitter detection according to claim 1, characterized in that: The method of preprocessing the device vibration signal includes: Perform preliminary processing on the equipment vibration signal to obtain a preliminary accurate signal, remove noise in the preliminary accurate signal to obtain a preliminary input signal, standardize or normalize the preliminary input signal to obtain a standardized vibration signal; divide the standardized vibration signal into m time periods to obtain m input signals; m is an integer greater than 1.

3. The interference elimination control system for jitter detection according to claim 2, characterized in that: The method of performing preliminary processing on the device vibration signal includes: Collect several wavelet basis function families, construct wavelet basis functions of different orders for each wavelet basis function family to form a wavelet basis function subset, classify and encode all wavelet basis functions in the wavelet basis function subset according to the family and order to form a wavelet basis function library; Define a time period, extract a signal of a time period from any position in the equipment vibration signal as the scoring signal; traverse each wavelet basis function in the wavelet basis function library, perform wavelet decomposition on the scoring signal x(n), and obtain the approximate component A and the detail component D; where n is the index of the discrete sampling point; based on the approximate component A and the detail component D obtained by wavelet decomposition, calculate the comprehensive score of the corresponding wavelet basis function; the calculation formula for the comprehensive score is: ； in, is the approximate entropy value of the approximate component A, is the scoring function of energy distribution, is the energy distribution, i.e., the set of energy value vectors of each scale component; is the energy retention rate; is the feature consistency scoring function, is the weight parameter; is the weight of the scoring function for energy distribution, is the fast Fourier transform, represents inverse fast Fourier transform; Scoring function for energy distribution ;in, is the maximum order of wavelet decomposition, For the The normalized energy of each component; The wavelet basis function with the lowest comprehensive score in the wavelet basis function library is used as the best matching wavelet basis. For the best matching wavelet basis, the wavelet entropy threshold Tapen and the energy threshold en are set; Using the best matching wavelet basis, the equipment vibration signal is decomposed by level I wavelet to obtain the corresponding level I approximate component A(I) and level I detail component D(I); the wavelet entropy Apen(A(I)) of the level I approximate component A(I) is calculated; and the energy loss rate is calculated; the value of I is continuously adjusted until Apen(A(I)) is less than Tapen and the energy loss rate is less than en; the value of I obtained at this time is the optimal series; the optimal series is denoted as q; using the best matching wavelet basis and the optimal series, the equipment vibration signal is decomposed by level q wavelet to obtain the low-frequency approximate component A(q); the original equipment vibration signal is subtracted from the low-frequency approximate component A(q) point by point to obtain a preliminary accurate signal.

4. The interference elimination control system for jitter detection according to claim 3, characterized in that: The method for obtaining the feature consistency scoring function includes: Extract time domain features and frequency domain features from the scoring signal x(n) and construct a signal feature vector X; the time domain features include mean, root amplitude, peak value and peak factor; the frequency domain features include power spectrum features, frequency center and bandwidth; extract the same time domain features and frequency domain features from the approximate component A and detail component D after wavelet decomposition, and construct corresponding feature vectors respectively; the feature vector corresponding to the approximate component A is recorded as the approximate feature vector A', and the feature vector corresponding to the detail component D is recorded as the detail feature vector D'; Collect several groups of labeled vibration signals and their corresponding fault modes; extract three groups of vectors X, A', and D' from each vibration signal, use X as input and the fault mode as a label, train a supervised learning model, and evaluate the correlation score S(X) between X and the fault mode by the trained supervised learning model; Input A' and D' into the trained supervised learning model to obtain the relevance scores respectively and ; Calculate feature consistency scoring function based on S(X), S(A') and S(D') ;in, is the fusion function, which is weighted average or product.

5. The interference elimination control system for jitter detection according to claim 4, characterized in that: The method of dividing the standardized vibration signal into m time periods includes: Defines the length of a time window ; Preset overlap ratio ; ； in, and is the adjustment parameter, is the first-order derivative of the normalized vibration signal in the current time window, is the standardized vibration signal in the current time window; is the change response parameter; is the index of the sampling point of the normalized vibration signal; Starting from the starting position of the standardized vibration signal, extract the first time window of length L; according to the overlap rate , calculate the starting position of the next time window, and repeat until the entire standardized vibration signal is covered, and obtain m time windows, that is, m time periods.

6. The interference elimination control system for jitter detection according to claim 5, characterized in that: The method of discretizing the m-segment input signal includes: The input signal is defined as a one-dimensional lattice chain, and each lattice point in the one-dimensional lattice chain corresponds to a sampling point of the input signal; the spin value range of the lattice points and the interaction rules between the lattice points are defined; the interaction rules between the lattice points are to calculate the total interaction energy and minimize the total interaction energy when simulating evolution; The total interaction energy is calculated as: ； in, is the total interaction energy, is the preset nearest neighbor interaction strength parameter, For the The spin state value of each grid point; For the The spin state of each lattice point; is the preset interaction benchmark strength, For the and The distance between the grid points, is the correlation length parameter; is the intensity parameter of the external magnetic field; According to the value of the input signal, the initial spin state of each grid point is assigned a value, and a linear or nonlinear mapping relationship is used to map the value of the continuous input signal to the value of the discrete spin state; through simulated evolution, the one-dimensional lattice chain converges to the ground state; When the one-dimensional lattice chain converges to the ground state, the current spin states of all lattice points are recorded, and the spin states are arranged in chronological order to form a signal state matrix M1; each row of the signal state matrix M1 corresponds to a time step, and each column corresponds to a lattice point.

7. The interference elimination control system for jitter detection according to claim 6, characterized in that: The method of making the one-dimensional lattice chain converge to the ground state by simulating evolution includes: Set an initial temperature T0; traverse each grid point in chronological order. grid points, calculate their spin states Energy changes before and after flipping ;in, is the total interaction energy after flipping, is the total interaction energy before flipping; For the The current spin state of each grid point The spin state after flipping; the total interaction energy before and after flipping is calculated and obtained according to the interaction rules; like , then accept the flip; if , then with probability Accept flip, where is the temperature of the current round; repeat a round of traversal of all grid points, and reduce the temperature after each N1 round; the formula for reducing the temperature is: ;in, is the basic cooling rate, is the current number of cooling steps, is the cooling index; is the oscillation intensity parameter, and ; is the oscillation angular frequency, is the initial phase; In the process of lowering the temperature, the total interaction energy is continuously calculated. When the change in the total interaction energy within M consecutive cooling steps is less than the preset change threshold, the one-dimensional lattice chain has converged to the ground state.

8. The interference elimination control system for jitter detection according to claim 7, characterized in that: The method of segmenting the signal state matrix includes: The signal state matrix M1 is regarded as a two-dimensional rectangular area. Initially, M1 is divided into A small rectangular block, is the initial preset segmentation granularity; Create a complete binary tree Y, where the leaf nodes of the complete binary tree Y correspond to small rectangular blocks; each internal node represents the union of the rectangular areas covered by its two child nodes; the root node of the complete binary tree Y corresponds to the entire signal state matrix M1; For each leaf node, calculate its node feature vector; the node feature vector is a joint vector composed of mean, variance, peak, wavelet energy, wavelet coefficient and Fourier coefficient; Preset the similarity threshold and calculate the Mahalanobis distance between each node in the complete binary tree Y and its sibling nodes; Starting from the bottom layer of the complete binary tree Y, traverse the pairs of sibling nodes at each layer; if the Mahalanobis distance of a pair of sibling nodes is greater than the similarity threshold, merge them into a parent node, continue to traverse to the upper layer, and repeat the merging until no further merging is possible; retrieve all unmerged nodes in the complete binary tree Y, use them as potential nodes, and map a rectangular area corresponding to each potential node back to the space of the original signal state matrix to obtain the corresponding potential interference segment.

9. The interference elimination control system for jitter detection according to claim 8, characterized in that: The method of determining the optimal interference elimination parameter includes: A hierarchical membrane structure is constructed, the outermost layer of the membrane structure is the skin membrane, and the inside of the membrane structure has several nested regional membranes; the signal characteristics of the potential interference segment are encoded as material objects and placed in the skin membrane; the signal characteristics include wavelet packet decomposition coefficients, harmonic components and spectral entropy; the encoding adopts binary encoding; Each regional membrane represents a type of interference exclusion parameter; a material object is placed in each regional membrane, and the material object represents a candidate value of the corresponding type of interference exclusion parameter; The membrane evolution rules are defined to include object mutation rules, object selection rules, object transfer rules and stop rules. The object mutation rules are used to perform mutation operations on material objects in the regional membrane to generate new objects; the object selection rules are used to perform cross operations on two material objects in the regional membrane to generate new objects; the object transfer rules are used to transfer material objects between different membranes; the stop rule is that the evolution in the regional membrane stops when the maximum number of iterations is reached; an output interface is set to read the values ​​of various interference exclusion parameters from the membrane structure after each iteration; The evolution rule is defined as randomly perturbing the material objects of the regional membrane in each iteration, calculating the value of the objective function, and retaining the material object with the smallest value of the objective function; The formula of the objective function is: ； in, For the The value of the objective function at the iteration, is the potential interference segment, is the index of the signal value within the potential interference segment; In order to process the potential interference segment by using the values ​​of various interference elimination parameters read from the membrane structure during iteration, the processed potential interference segment is obtained; is the covariance calculation function; is the standard deviation of the original potential interference segment, For the The standard deviation of the potential interference segment after processing at the iteration; For the iteration Class interference exclusion parameters; For the iteration weights of class interference exclusion parameters; is a nonlinear parameter; According to the membrane evolution rules and the evolution rules, the material objects are exchanged and evolved between the regional membranes, and each iteration generates a new combination of material objects; the value of the objective function and the corresponding material objects of each iteration are recorded; after each iteration, it is checked whether the stopping rule is satisfied. If so, a combination of material objects that minimizes the value of the objective function is selected from all the regional membranes, and the interference exclusion parameters corresponding to this group of material objects are used as the optimal interference exclusion parameters.

10. An interference elimination control method for jitter detection, which is implemented based on an interference elimination control system for jitter detection according to any one of claims 1 to 9, characterized in that: include: S1, collecting equipment vibration signals, and preprocessing the equipment vibration signals to obtain m-segment input signals; S2, discretize the m-segment input signal to obtain a signal state matrix; S3, segmenting the signal state matrix to obtain several potential interference segments; S4. Based on several potential interference segments, determine the optimal interference elimination parameters; use the optimal interference elimination parameters to perform interference elimination processing on the potential interference segments, and output the interference elimination control signal to the control terminal.

Images