Multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge calculation

Abstract

The invention belongs to the field of data compression and reconstruction of a mechanical vibration wireless sensor network, and particularly relates to a multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge calculation. The method comprises the following steps: acquiring an original mechanical vibration signal through an edge computing node; adopting a sparse decomposition algorithm to obtain a sparse mechanical vibration signal; obtaining a compressed mechanical vibration signal observation value based on the dual-structure measurement matrix corresponding to the initial compression ratio; adopting a greedy algorithm to obtain a reconstructed mechanical vibration signal; judging whether the error between the reconstructed mechanical vibration signal and the original mechanical vibration signal meets the requirement of a set threshold value or not, and then dynamically adjusting the compression ratio; obtaining an optimal mechanical vibration signal observation value based on a double-structure measurement matrix corresponding to the dynamic compression ratio; and reconstructing the optimal observation value of the mechanical vibration signal by adopting a greedy algorithm through the central server to obtain a final reconstructed mechanical vibration signal. According to the invention, the edge computing technology is utilized to preprocess the mechanical vibration signal, so that the flexibility and efficiency of state monitoring are improved; the data is effectively compressed and measured at the edge computing node, so that the subsequent reconstruction precision can be optimized, the signal recovery quality is remarkably improved, and meanwhile, the computing pressure of the central server is also reduced.

Description

Technical Field

[0001] The present invention belongs to the field of data compression and reconstruction of mechanical vibration wireless sensor networks, and particularly relates to a method for reconstructing mechanical vibration signals based on multi-structure dynamic compressive sensing with edge computing. Background Art

[0002] With the continuous increase in machine complexity, the number of signals that need to be captured and processed shows an explosive growth trend, which brings huge challenges and pressures to the data center servers in wireless sensor networks. Traditional data processing methods are difficult to effectively handle such a large amount of data streams, resulting in limited efficiency in data transmission, storage, and analysis, and even possible data loss or delay.

[0003] In wireless sensor networks (WSNs), a large amount of energy consumption caused by data communication on the channel becomes a key issue restricting its development in the field of mechanical vibration. Especially in an environment with increasing machine complexity, for the increasingly demanding signal acquisition requirements, how to effectively utilize the correlation between sensor signals to compress and collect a large amount of distributed sensing data has become an important research issue.

[0004] Introducing compressive sensing as a new data acquisition and compression theory into WSNs provides an innovative solution to the challenges faced by WSNs data acquisition. By applying compressive sensing technology in the process of sensing data acquisition and making full use of the prior information of the correlation between sensing data, the edge computing nodes can perform compressive sampling while collecting data, and transfer the computational tasks of data compression and acquisition to the powerful aggregation nodes, thereby reducing the complexity of compressive coding. For example, Chinese Patent CN117412260A discloses a method for collecting data in a wireless sensor network based on compressive sensing, which considers the impact of compressive sensing on energy consumption, redefines the energy consumption model, and reduces the amount of data transmission; finally, the base station reconstructs the received data through a compressive sensing recovery algorithm to obtain the original data.

[0005] However, the structures of the observation matrices in current compressive sensing technologies are relatively single, mainly divided into two categories: random measurement matrices and deterministic measurement matrices. Among them, the random measurement matrices have better reconstruction performance, but their computational processes are extremely complex, which makes it difficult to implement in hardware systems and increases the complexity and cost of hardware design; while the deterministic measurement matrices are conducive to hardware system implementation, but their reconstruction performance is not as good as the former. At the same time, the current measurement matrix design technology does not consider dynamically adapting to the characteristics of the currently acquired signals. Therefore, the existing technologies cannot ensure the signal reconstruction accuracy while reducing the computational complexity according to the characteristics of the current signals. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention proposes a multi-structure dynamic compressive sensing mechanical vibration signal reconstruction method based on edge computing. The method is applied to an edge computing node and a central server, and the method includes:

[0007] 101. Collect the original mechanical vibration signal through the edge computing node;

[0008] 102. Process the original mechanical vibration signal at the edge computing node using a sparse decomposition algorithm to obtain the thinned mechanical vibration signal; the thinned mechanical vibration signal is composed of the product of a sparse basis matrix and sparse coefficients;

[0009] 103. Perform compressive sensing on the thinned mechanical vibration signal at the edge computing node based on a dual-structure measurement matrix corresponding to an initial compression ratio to obtain an observed value of the compressed mechanical vibration signal; the observed value of the compressed mechanical vibration signal is composed of the product of a sensing matrix and sparse coefficients; the sensing matrix is composed of the product of the dual-structure measurement matrix and the sparse basis matrix;

[0010] 104. Reconstruct the observed value of the compressed mechanical vibration signal at the edge computing node using a greedy algorithm to obtain the reconstructed mechanical vibration signal;

[0011] 105. At the edge computing node, determine whether the error between the reconstructed mechanical vibration signal and the original mechanical vibration signal meets the set threshold requirements, and then dynamically adjust the compression ratio;

[0012] 106. Perform compressive sensing on the thinned mechanical vibration signal at the edge computing node based on a dual-structure measurement matrix corresponding to the dynamic compression ratio until the best observed value of the mechanical vibration signal is obtained;

[0013] 107. Reconstruct the best observed value of the mechanical vibration signal at the central server using a greedy algorithm to obtain the finally reconstructed mechanical vibration signal.

[0014] The beneficial effects of the present invention include:

[0015] In view of the problem of excessive transmission volume of a large amount of vibration data for mechanical vibration monitoring in a wireless sensor network, the present invention introduces the compressive sensing technology, deeply studies the linear measurement process of signals starting from the measurement matrix, combines the advantages of random measurement matrices and deterministic measurement matrices, and designs a multi-structured measurement matrix, which can not only reduce the computational complexity but also ensure the accuracy of signal reconstruction. At the same time, an adaptive dynamic measurement matrix is designed using error feedback, and by adjusting the observed values, the problem that the observed values obtained by compressing different signals with a fixed observation matrix cannot reconstruct the best effect is solved, ensuring good final signal reconstruction accuracy. The present invention uses edge computing technology to preprocess mechanical vibration signals, improving the flexibility and efficiency of condition monitoring; effectively compressing and measuring data at the edge computing node can effectively optimize the subsequent reconstruction accuracy, significantly improve the signal recovery quality, and at the same time reduce the computational pressure on the central server. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of a multi-structured dynamic compressive sensing vibration signal reconstruction method in the present invention;

[0017] Figure 2 It is a schematic diagram of original data sampling in the present invention;

[0018] Figure 3 It is a schematic diagram of a multi-structured dynamic compressive sensing vibration signal reconstruction algorithm in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] An embodiment of the present invention proposes a multi-structure dynamic compressive sensing mechanical vibration signal reconstruction method based on edge computing. The method is applied to an edge computing node and a central server. Among them, the edge computing node is located at the network edge, close to the computing resources of the data source or user terminal, and is distributed near Internet of Things devices, industrial devices, mobile terminals, etc. The edge computing node can be an edge gateway, an edge server, a router, an Internet of Things terminal device, or other fixed or mobile devices with computing capabilities. The central server is a large server located in a data center or a cloud computing platform. The edge computing node is connected to the central server and other edge computing nodes through a wired or wireless communication network. The edge computing nodes can also be connected through a gateway device or the Internet. The central server can communicate with the edge computing node through a high-speed network to obtain and process data from the edge computing node, such as the best observed value of the mechanical vibration signal obtained by the edge computing node. The central server can also be connected to other data centers and cloud computing platforms through the Internet to achieve broader resource sharing and collaboration.

[0021] First, refer to Figure 1 and Figure 3 , as Figure 1 A multi-structure dynamic compressive sensing mechanical vibration signal reconstruction method based on edge computing proposed by the present invention, the method includes the following contents:

[0022] 101. Collect the original mechanical vibration signal through the edge computing node;

[0023] In the embodiment of the present invention, the acquisition of the original mechanical vibration signal can be derived from the experimental data of a dynamic drive system simulator (DDS). The edge computing node collects experimental data of 8 different fault states, and the sample length of each sampling is 2048.

[0024] In the embodiment of the present invention, the acquisition of the original mechanical vibration signal can also be derived from a real sensor device. As Figure 2 shown, for example, an inertial acceleration sensor is installed on a mechanical device to facilitate the measurement of vibration acceleration data from multiple directions. Ensure that the sensor is firmly fixed on the device to accurately capture the vibration characteristics. Make the correct physical and electrical connections between the inertial acceleration sensor, the data acquisition system, and the computer to ensure that the signal transmission is interference-free and stable. Set the total sampling duration to T seconds and select an appropriate sampling frequency fs Hz. According to these settings, calculate the corresponding number of sampling points N, that is, N = fs × T. Through multiple samplings, the sample length of the sampling can also be set to 2048, so that a time-domain vibration acceleration signal matrix X can be obtained.

[0025] In the embodiments of the present invention, the acquisition of the original mechanical vibration signal is performed at the edge computing node, which can reduce the amount of data transmission, has strong real-time performance, reduces the load of the central server, and protects data security and privacy, etc.

[0026] 102. At the edge computing node, the original mechanical vibration signal is processed using a sparse decomposition algorithm to obtain a sparse mechanical vibration signal; the sparse mechanical vibration signal is composed of the product of a sparse basis matrix and sparse coefficients;

[0027] In the embodiments of the present invention, processing the original mechanical vibration signal using a sparse decomposition algorithm to obtain a sparse mechanical vibration signal includes:

[0028] Select an initial dictionary; the size of the initial dictionary is larger than the dimension of the original mechanical vibration signal;

[0029] Use a sparse coding method to find the sparse coefficients of each original mechanical vibration signal sample on the dictionary;

[0030] Update the dictionary according to the obtained sparse coefficients and the original mechanical vibration signal;

[0031] Through the iteration of sparse coding and dictionary update, when the convergence condition is met, a sparse mechanical vibration signal is obtained.

[0032] In some embodiments of the present invention, the K-SVD algorithm can also be used to update the algorithm by alternately performing sparse representation and constrained singular value decomposition to obtain an adaptive redundant dictionary D, and its objective function can be expressed as:

[0033]

[0034] Among them, is the target signal, D ∈ R N×K (K << N) is the sparse representation dictionary; is the sparse representation coefficient; ‖·‖ F is the Frobenius norm error; ‖·‖ 0 is the l 0 norm, that is, the number of non-zero elements; K is the maximum value of the number of non-zero elements in the sparse coefficients.

[0035] In some embodiments of the present invention, the SAMP algorithm is used in the sparse coding part of the K-SVD algorithm. The SAMP algorithm estimates the true sparsity of the signal by gradually selecting the dictionary atoms with the maximum correlation. The sparse coding process is further refined, that is, first fix the dictionary D and optimize the sparse matrix X. The following sparse representation model is adopted:

[0036] S test= DΨ test + E k

[0037] where D is the dictionary obtained through dictionary learning in the K-SVD algorithm, and E k represents the error matrix. By minimizing the residual matrix, the dictionary D and the sparse matrix Ψ are gradually improved, thereby improving the accuracy of sparse representation to obtain a better sparse mechanical vibration signal.

[0038] 103. At the edge computing node, compressive sensing is performed on the sparse mechanical vibration signal based on the dual-structure measurement matrix corresponding to the initial compression ratio to obtain an observed value of the compressed mechanical vibration signal; the observed value of the compressed mechanical vibration signal is composed of the product of the sensing matrix and the sparse coefficient; the sensing matrix is composed of the product of the dual-structure measurement matrix and the sparse basis matrix;

[0039] In the embodiment of the present invention, a new dual-structure measurement matrix can be designed according to the initial compression ratio (CR). The dual-structure measurement matrix is formed by combining an M-dimensional identity matrix and an M×(N - M) random Bernoulli matrix, and the specific size can be determined according to the actual requirements of the application scenario.

[0040] In some embodiments of the present invention, the set initial compression ratio CR is 0.6, the number of original signal samples and the signal length are (200, 512), that is, N = 512. The calculation formula for the number of rows of the dual-structure measurement matrix is:

[0041] M = N - CR×N

[0042] where CR represents the current compression ratio, and N represents the length of the original mechanical vibration signal. Therefore, the number of rows M of the obtained measurement matrix is 204.

[0043] Further, the dual-structure measurement matrix is composed of an M×M-dimensional identity matrix Φ 1 and an M×(N - M)-dimensional random Bernoulli matrix Φ 2 ; the number of rows M of the dual-structure measurement matrix is determined by the compression ratio CR and the length N of the original mechanical vibration signal, and the number of columns N of the dual-structure measurement matrix is determined by the length N of the original mechanical vibration signal. Its formula can be expressed as:

[0044]

[0045] where Φ 1 is an M-dimensional identity matrix, Φ 2 is an M×(N - M) random Bernoulli matrix, Y is the observed value obtained after compressive measurement, and X is the original signal sample of length N.

[0046] In a preferred embodiment of the present invention, the method further includes determining whether the double-structured measurement matrix satisfies the condition based on the restricted isometry property formula. If the condition is satisfied, the double-structured measurement matrix is used for compressive sensing. If the condition is not satisfied, the number of rows of the double-structured measurement matrix is adjusted until the condition is satisfied.

[0047] Specifically, let the double-structured measurement matrix Φ be an M×N matrix in the form of [I, B], where the non-zero terms of the Bernoulli matrix B conform to random numbers. The Bernoulli distribution is a discrete probability distribution, and the random variable has only two possible values, generally 0 or 1. In a random Bernoulli matrix, each element takes the value 1 with probability p and the value 0 with probability 1 - p. When the measurement value satisfies then [I, B] satisfies the K-order RIP property, and the proof process is as follows:

[0048]

[0049] Since the random Bernoulli matrix B satisfies the RIP property, the bound of the second term on the right side is:

[0050]

[0051] Since the elements in the random Bernoulli matrix B satisfy therefore, it can be obtained that:

[0052]

[0053] where A 1 represents the coefficient for determining the minimum measurement value M, log represents the logarithm, which can be base 10, or the irrational number e, or other common bases. The present invention does not make any limitations in this regard. K represents the sparsity of the signal. Among them, x is an N-dimensional sparse vector, x 1 has a length of M, and x 2 has a length of N - M.

[0054] In the embodiment of the present invention, by adjusting the number of rows of the double-structured measurement matrix to make the double-structured measurement matrix satisfy the condition, it can ensure the accurate reconstruction of the signal in compressive sensing, enabling the original signal to be effectively reconstructed from a small amount of measurement data, thereby realizing data compression and efficient transmission.

[0055] In some embodiments of the present invention, after steps 101-102, the size of the obtained observation matrix is (204, 512). The observed value of each signal sample obtained by passing the signal with a sparse sample number of 200 and a length of 512 through this observation matrix is 204. This step significantly reduces the amount of mechanical vibration signal data required, reduces the data transmission pressure and data storage pressure, and at the same time can also ensure the accurate reconstruction of the original signal in the subsequent process.

[0056] 104. At the edge computing node, use the greedy algorithm to reconstruct the observed values of the compressed mechanical vibration signals to obtain the reconstructed mechanical vibration signals.

[0057] In the embodiments of the present invention, the sparse coefficient estimation value can be estimated locally at the edge node using the SAMP algorithm. According to this sparse coefficient estimation value, the mechanical vibration signal is reconstructed, and the details of the signal are restored through sparse representation to obtain the reconstructed signal.

[0058] It can be understood that performing locally at the edge node can reduce the amount of data transmission, and has the advantages of strong real-time performance, reducing the load of the central server, protecting data security and privacy, etc. This embodiment can ensure the security of data to a certain extent.

[0059] 105. At the edge computing node, determine whether the error between the reconstructed mechanical vibration signal and the original mechanical vibration signal meets the set threshold requirements, and then dynamically adjust the compression ratio.

[0060] In the embodiments of the present invention, the dynamic compression ratio is determined by the error between the original mechanical vibration signal and the reconstructed mechanical vibration signal; the error can use the mean square error, signal-to-noise ratio, and peak signal-to-noise ratio.

[0061] Exemplarily, the calculation formula of the error can be:

[0062]

[0063] where error is the relative error between the calculated restored signal and the original signal, which measures the difference degree between the restored signal and the original signal, and x or,i represents the i-th sample value of the original signal (original mechanical vibration signal), and x rec,i represents the i-th sample value of the restored signal (reconstructed mechanical vibration signal), and n is the number of observed values, that is, the number of samples. By calculating the square root of the sum of the squares of the differences between the corresponding samples of the original signal and the restored signal, and the square root of the sum of the squares of the original signal samples, and then dividing the two, the relative error is obtained, so as to quantitatively evaluate the reduction degree of the restored signal to the original signal.

[0064] When the error exceeds the set threshold, the dynamic compression ratio is adjusted to:

[0065] CR new = max[CR(1 - factor), CR min

[0066] When the error does not exceed the set threshold, the dynamic compression ratio is adjusted to:

[0067] CR new = min[CR(1 + factor), CR max

[0068] Among them, CR new represents the updated compression ratio, CR represents the value of the previous round of iteration, that is, the current compression ratio, factor represents the proportionality factor controlling the amplitude of adjusting the compression ratio, CR min represents the minimum compression ratio, and CR max represents the maximum compression ratio.

[0069] In the embodiment of the present invention, by calculating and judging whether the recovery error between the reconstructed signal and the original signal meets the set threshold requirement, the compression ratio (CR) is dynamically adjusted, and the parameters of the double - structure measurement matrix are adjusted in reverse to optimize the reconstruction effect and ensure the best observed value.

[0070] 106. At the edge computing node, perform compressive sensing on the sparse mechanical vibration signal based on the double - structure measurement matrix corresponding to the dynamic compression ratio until the best observed value of the mechanical vibration signal is obtained;

[0071] Since the compression ratio affects the row dimension of the double - structure measurement matrix, once the compression ratio is changed, the double - structure measurement matrix will be dynamically adjusted. Through this adaptive dynamic compression ratio in this embodiment, the double - structure measurement matrix can be dynamically changed, and by adjusting the observed value, the problem that the observed value obtained by compressing different signals with a fixed observation matrix cannot reconstruct the best effect is solved, ensuring good final signal reconstruction accuracy.

[0072] The processing process of dynamically adjusting the double - structure measurement matrix according to the error feedback is expressed as:

[0073] Y = ΦX = ΦΨS = ΘS

[0074]

[0075] Among them, Y is the observed value obtained after compressive measurement, n is the number of observed values, Φ is the double - structure measurement matrix, Ψ is the N - dimensional sparse basis matrix, Θ is the sensing matrix, and S is the sparse coefficient; in the iterative process, Lasso regression is used to make the recovered signal as close as possible to the original signal, ‖·‖ 2 is l​​2 Norm, ‖·‖ 1 is the l 1 norm, and λ is a regularization parameter that controls the model complexity and sparsity.

[0076] Through the above formula, an adaptive dynamic measurement matrix is designed using error feedback. By adjusting the observed values, the problem that the observed values obtained by compressing different signals with a fixed observation matrix cannot reconstruct the best effect is solved, ensuring good final signal reconstruction accuracy. This embodiment can not only reduce the computational complexity but also ensure the accuracy of signal reconstruction.

[0077] 107. At the central server, the optimal mechanical vibration signal observed values are reconstructed using the greedy algorithm to obtain the finally reconstructed mechanical vibration signal.

[0078] In the embodiment of the present invention, the optimized optimal observed values and the new double-structure measurement matrix can be transmitted to the central server. The central server reconstructs the mechanical vibration signal again to obtain the finally reconstructed signal. The powerful computing power of the central server can quickly complete these large amounts of matrix operations and iterative solution calculations, greatly shortening the signal reconstruction time and improving the processing efficiency.

[0079] It can be understood that in the embodiment of the present invention, steps 101-106 are implemented through the edge computing node, and step 107 is implemented through the central server. Between step 106 and step 107, the edge computing node transmits the optimal mechanical vibration signal observed values to the central server through a wired or wireless communication network, and the central server then processes the data calculated by each edge computing node. This method uses edge computing technology to preprocess the mechanical vibration signal, improving the flexibility and efficiency of condition monitoring; at the edge computing node, the reconstruction accuracy can be effectively optimized, significantly improving the signal recovery quality, and at the same time reducing the computational pressure on the central server.

[0080] In some embodiments of the present invention, a power transmission comprehensive test bench (DDS) is used to collect fault signals of a parallel gearbox. The test bench mainly includes a driving motor, a two-stage planetary gearbox, a two-stage parallel gearbox, and a programmable heavy-duty brake. An acceleration sensor is installed on the right intermediate shaft in the vertical direction, and the monitoring object is the outer ring fault bearing on the right side of the intermediate shaft of the parallel gear. This experiment uses a NI9234 signal acquisition card and a vibration acceleration sensor (such as model: PCB352C03) to collect the outer ring bearing fault signal at a sampling frequency of 25600Hz and a motor speed of 20Hz. The acquisition length is set to 2048 points. In order to quantitatively and intuitively evaluate the reconstruction performance of the model, the Mean Absolute Error (MAE), Mean Square Error (MSE), Normalized Mean Square Error (NMSE), and Pearson Correlation Coefficient (PCC) are introduced to evaluate the reconstruction accuracy of the model. The calculation formulas for these four evaluation indicators are as follows:

[0081]

[0082] where x i represents the original signal, represents the reconstructed signal, and N represents the length of the original signal samples.

[0083] The results obtained by experimentally comparing the method proposed in the present invention with several other different reconstruction methods are shown in Table 1:

[0084] Table 1 Comparison table of key parameters of the present invention on the DDS dataset

[0085]

[0086] As can be seen from Table 1, the multi-structure dynamic compressive sensing vibration signal reconstruction method proposed in the present invention has better reconstruction accuracy for signals than several other algorithms. At the same time, the present invention uses edge computing technology to preprocess mechanical vibration signals, enhancing the flexibility and efficiency of condition monitoring; it can effectively optimize the reconstruction accuracy at the edge computing node, significantly improving the recovery quality of mechanical vibration signals, and at the same time reducing the computing pressure on the central server.

[0087] The above-described embodiments have further elaborated on the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made to the present invention within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge computing, characterized in that: The method is applied to an edge computing node and a central server, and the method comprises:

101. Collecting original mechanical vibration signals through edge computing nodes; 102. Processing the original mechanical vibration signal at the edge computing node using a sparse decomposition algorithm to obtain a sparse mechanical vibration signal; the sparse mechanical vibration signal is composed of the product of a sparse basis matrix and a sparse coefficient; 103. Performing compressed sensing on the sparse mechanical vibration signal based on the dual-structure measurement matrix corresponding to the initial compression ratio at the edge computing node to obtain a compressed mechanical vibration signal observation value; the compressed mechanical vibration signal observation value is composed of the product of the sensing matrix and the sparse coefficient; the sensing matrix is ​​composed of the product of the dual-structure measurement matrix and the sparse basis matrix; 104. Reconstructing the compressed mechanical vibration signal observation value at the edge computing node by using a greedy algorithm to obtain a reconstructed mechanical vibration signal; 105. When determining at the edge computing node whether an error between the reconstructed mechanical vibration signal and the original mechanical vibration signal meets a set threshold requirement, the compression ratio is dynamically adjusted; 106. Performing compressed sensing on the sparse mechanical vibration signal based on the dual-structure measurement matrix corresponding to the dynamic compression ratio at the edge computing node until an optimal mechanical vibration signal observation value is obtained; 107. Reconstruct the optimal mechanical vibration signal observation value at the central server by using a greedy algorithm to obtain a final reconstructed mechanical vibration signal.

2. According to the edge computing-based multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method of claim 1, it is characterized in that: The original mechanical vibration signal is processed by a sparse decomposition algorithm to obtain a sparse mechanical vibration signal including: Selecting an initial dictionary; the size of the initial dictionary is greater than the dimension of the original mechanical vibration signal; Using sparse coding method to find the sparse coefficient of each original mechanical vibration signal sample on the dictionary; updating the dictionary according to the obtained sparse coefficients and the original mechanical vibration signal; Through the iteration of sparse coding and dictionary updating, when the convergence condition is met, the sparse mechanical vibration signal is obtained.

3. According to the edge computing-based multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method of claim 1, it is characterized in that: The dual-structure measurement matrix is ​​composed of an M×M-dimensional unit matrix Φ1 and an M×(NM)-dimensional random Bernoulli matrix Φ2; the number of rows M of the dual-structure measurement matrix is ​​determined by the compression ratio CR and the length N of the original mechanical vibration signal, and the number of columns N of the dual-structure measurement matrix is ​​determined by the length N of the original mechanical vibration signal.

4. According to the edge computing-based multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method of claim 3, it is characterized in that: The calculation formula for the number of rows of the dual structure measurement matrix is: M=N-CR×N Wherein, CR represents the current compression ratio, and N represents the length of the original mechanical vibration signal.

5. A multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge computing according to claim 3 or 4, characterized in that: The method also includes judging whether the dual structure measurement matrix meets conditions based on the finite isometry property formula; if the conditions are met, the dual structure measurement matrix is ​​used for compressed sensing; if the conditions are not met, the number of rows of the dual structure measurement matrix is ​​adjusted.

6. The multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge computing according to claim 5 is characterized in that: The conditions are Wherein, A1 represents the coefficient for determining the minimum measurement value M, log represents the logarithm, and K represents the sparsity of the signal.

7. The multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge computing according to claim 1 is characterized in that: The optimal mechanical vibration signal observation value is obtained by multiplying the dual structure measurement matrix and the optimal original mechanical vibration signal; the optimal original mechanical vibration signal is determined by a sparse basis matrix and an optimal sparse coefficient.

8. The multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge computing according to claim 7 is characterized in that: The solution formula of the optimal sparse coefficient S is expressed as: Where S is the sparsity coefficient, n is the number of observations, Y is the observed value of the mechanical vibration signal obtained after compression measurement, Θ is the sensing matrix, ‖·‖2 is the l2 norm, ‖·‖1 is the l1 norm, and λ is the regularization parameter that controls the complexity and sparsity of the model.

9. The multi-structure dynamic compressed sensing mechanical vibration signal reconstruction method based on edge computing according to claim 1 is characterized in that: The dynamic compression ratio is determined by the error between the original mechanical vibration signal and the reconstructed mechanical vibration signal; When the error exceeds the set threshold, the dynamic compression ratio is adjusted to: CR new =max[CR(1-factor),CR min ] When the error does not exceed the set threshold, the dynamic compression ratio is adjusted to: CR new =min[CR(1+factor),CR max ] Among them, CR new Indicates the updated compression ratio, CR indicates the value of the previous iteration, that is, the current compression ratio, and factor indicates the proportional factor that controls the adjustment of the compression ratio. min Indicates the minimum compression ratio, CR max Indicates the maximum compression ratio.

Images