A torque signal denoising method based on MFO-WPT algorithm

Abstract

The application discloses a torque signal denoising method based on an MFO-WPT algorithm, and belongs to the technical field of signal processing. The method comprises the following steps: obtaining a noisy torque signal; decomposing the noisy torque signal at different decomposition scales to obtain wavelet packet coefficients of each node of the noisy torque signal at different scales; obtaining a signal signal-to-noise ratio (SNR) after wavelet packet coefficient denoising, and taking the signal SNR as a fitness function of a firefly algorithm; optimizing the wavelet packet threshold value by using the fitness function and the firefly algorithm to find an optimal wavelet packet threshold value; constructing an optimal wavelet packet threshold function according to the optimal wavelet packet threshold value, wherein the optimal wavelet packet threshold function is an optimal wavelet packet hard threshold function or an optimal wavelet packet soft threshold function; obtaining estimated wavelet packet coefficients by using the wavelet packet coefficients and the optimal wavelet packet threshold function; and performing wavelet packet inverse transformation on the estimated wavelet packet coefficients to reconstruct the estimated wavelet packet coefficients into a denoised torque signal.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a torque signal denoising method based on the MFO-WPT algorithm. Background Technology

[0002] With the continuous maturation and improvement of technology, accurate torque measurement technology has a promising future. Torque sensors are widely used in detecting the output torque and power of rotating power equipment such as torque wrenches, electric motors, and engines, as well as in production monitoring and quality control. In Europe, their application has surpassed that of force sensors. However, due to the complex measurement environment, the torque signal acquired by the torque sensor is filled with noise. To obtain a true torque signal and reduce the impact of noise on measurement accuracy, denoising the signal and improving the signal-to-noise ratio is crucial.

[0003] Currently, there are many methods for denoising torque signals, including digital filter methods, Fourier transform methods, and wavelet packet denoising methods. Digital filter methods and Fourier transform methods are mostly based on frequency or time domain signal processing and analysis, and cannot refine or separately process the local features of the torque signal, so the denoising effect is not ideal. Wavelet packet denoising is a time-frequency localization analysis method with multi-resolution characteristics. It has the feature of multi-resolution analysis, and it is also a time-frequency localization analysis method with a fixed window area but a changeable shape, making it particularly suitable for the analysis of abrupt, non-stationary signals. Therefore, wavelet packet denoising is the most ideal denoising method for torque signals.

[0004] Currently, commonly used wavelet packet denoising algorithms include modulus maxima denoising, correlation denoising, and threshold denoising. Among them, wavelet packet threshold denoising (WPT) is widely used due to its high sensitivity, strong anti-interference ability, and good denoising effect. However, wavelet packet threshold denoising has the problem that the optimal threshold cannot be accurately selected during the selection process, resulting in poor denoising effect and easy distortion of the reconstructed torque signal after denoising. Summary of the Invention

[0005] The purpose of this invention is to provide a torque signal denoising method based on the MFO-WPT algorithm, which can solve the technical problems of existing wavelet packet thresholds that cannot accurately select the optimal threshold during the selection process, resulting in poor denoising effect and easy distortion of the reconstructed torque signal after denoising.

[0006] To solve the above-mentioned technical problems, the present invention is implemented as follows:

[0007] This invention provides a torque signal denoising method based on the MFO-WPT algorithm, comprising:

[0008] S101: Acquire noisy torque signal;

[0009] S102: Decompose the noisy torque signal at different decomposition scales and obtain the wavelet packet coefficients of each node of the noisy torque signal at different scales;

[0010] S103: After denoising the wavelet packet coefficients, the signal-to-noise ratio is obtained, and the signal-to-noise ratio is used as the fitness function of the moth-to-a-flame optimization algorithm;

[0011] S104: Using the fitness function, the wavelet packet threshold is optimized through the moth-to-a-flame optimization algorithm to find the optimal wavelet packet threshold;

[0012] S105: Construct the optimal wavelet packet threshold function based on the optimal wavelet packet threshold, wherein the optimal wavelet packet threshold function is either the optimal wavelet packet hard threshold function or the optimal wavelet packet soft threshold function, and obtain the estimated wavelet packet coefficients through the wavelet packet coefficients and the optimal wavelet packet threshold function;

[0013] S106: Perform inverse wavelet packet transform on the estimated wavelet packet coefficients to reconstruct the estimated wavelet packet coefficients into a denoised torque signal.

[0014] In this embodiment of the invention, the acquired noisy torque signal is decomposed using wavelet packet thresholding. The Moth-Flame Optimization Algorithm (MFO) is used to optimize the selection of the wavelet packet threshold, accurately selecting the optimal threshold. A wavelet packet threshold function is constructed using the selected optimal threshold, and then the estimated wavelet packet coefficients are obtained using the wavelet packet threshold function. The wavelet packet coefficients are then inversely transformed to reconstruct the denoised torque signal. This method effectively denoises the noisy torque signal, resulting in high fidelity and distortion-free reconstruction, thereby improving the accuracy of torque signal characteristic waveforms in clinical diagnosis. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a torque signal denoising method based on the MFO-WPT algorithm provided in an embodiment of the present invention.

[0016] The realization of the objective, functional characteristics and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0018] The following detailed description, in conjunction with the accompanying drawings, of the method for denoising dissolved organic carbon torque signals using ultraviolet-visible spectra provided by the present invention through specific embodiments and application scenarios, will illustrate this invention in detail.

[0019] Reference Figure 1 The diagram shows a flowchart of a torque signal denoising method based on the MFO-WPT algorithm provided by an embodiment of the present invention.

[0020] This invention provides a torque signal denoising method based on the MFO-WPT algorithm, comprising:

[0021] S101: Acquire noisy torque signal.

[0022] In one possible implementation, S101 specifically includes:

[0023] S1011: Acquires torque signals via a torque sensor;

[0024] S1012: Amplify the torque signal to obtain a noisy torque signal.

[0025] S102: Decompose the noisy torque signal at different decomposition scales to obtain the wavelet packet coefficients of each node of the noisy torque signal at different scales.

[0026] In one possible implementation, S102 specifically includes:

[0027] S1021: Select the decomposition level and wavelet packet basis functions;

[0028] S1022: Decompose the noisy torque signal according to the decomposition level and wavelet packet basis function:

[0029] in, and They represent the first j Layer 2 n The node and the first j Layer 2 n +1 wavelet packet coefficients corresponding to the nodes, h and g Represents the filter coefficients. n This represents the number of decomposition layers.

[0030] The number of decomposition layers is n, where n is an integer greater than or equal to 3 and less than or equal to 5.

[0031] Optionally, the wavelet packet basis functions belong to the db series and sym series basis functions.

[0032] It should be noted that using the WPT algorithm to decompose noisy torque signals can effectively alleviate the problem that wavelet transform (WT) can only decompose the low-frequency part of noisy torque signals. WPT can further decompose the high-frequency part of noisy torque signals, has higher resolution, and is suitable for analyzing non-stationary signals.

[0033] S103: After denoising the wavelet packet coefficients, the signal-to-noise ratio is obtained, and the signal-to-noise ratio is used as the fitness function of the moth-to-a-flame optimization algorithm.

[0034] In one possible implementation, S103 specifically includes:

[0035] S1031: Obtain the signal-to-noise ratio after denoising the wavelet packet coefficients. :

[0036] in, This indicates a noisy torque signal. This represents the torque signal obtained after noise reduction. Indicates the length of the torque signal sequence;

[0037] S1032: Use the signal-to-noise ratio as the fitness function of the moth-to-a-flame optimization algorithm.

[0038] The fitness function measures the spatial quality of the moth and the flame; the larger the fitness function, the better the location of the moth and the flame.

[0039] S104: Using the fitness function, the wavelet packet threshold is optimized through the moth-to-a-flame optimization algorithm to find the optimal wavelet packet threshold.

[0040] It should be noted that the moth-to-flame optimization algorithm is a swarm intelligence optimization algorithm. It involves two types of individuals: moths and flames. The moths select and search for the optimal wavelet packet threshold around the flame using a spiral search strategy. After the search is complete, the flame moves to its optimal position. The moths are candidate solutions, and the flame is the current optimal solution. Initially, the number of moths and flames is the same. The optimization variable is the position of the moths. Moths can change their position vectors to fly in one-dimensional, two-dimensional, three-dimensional, and other multi-dimensional spaces, progressively searching and updating their positions. An adaptive mechanism is employed to reduce the number of flames and obtain the optimal wavelet packet threshold.

[0041] In one optional implementation, S104 specifically includes:

[0042] S1041: Initialize the moth-to-a-flame optimization algorithm, set the current iteration number to 1, and determine the wavelet packet threshold of the parameter to be optimized. d Moth population size n ;

[0043] S1042: Determine the upper and lower limits for wavelet packet threshold optimization:

[0044] in, for j Wavelet packet thresholding at scale for j The length of the wavelet packet detail coefficients at different scales. For noise variance;

[0045] In one possible implementation, S1042 specifically includes:

[0046] S1042A: Introducing the noise variance equation :

[0047] in, This represents the median function. for j Scale k The wavelet packet coefficients of each node, q It is a constant, selected between 0.1 and 1.0;

[0048] S1042B: Get parameters q The values ​​are 0.1 and 1, and the maximum wavelet packet threshold is calculated. Sum and minimum wavelet packet threshold .

[0049] S1043: Randomly generate moth positions in the search space as the initial moth population matrix. And calculate the fitness value vector corresponding to the initial population matrix of moths. ;

[0050] S1044: Construct a flame position variable matrix based on the initial moth population size. In this process, the initial number of flame iterations is the same as the initial number of moths, and the fitness value vector corresponding to the flame position variable matrix is ​​calculated. :

[0051] S1045: Sort the moth positions in ascending order of fitness value, and assign the sorted moth positions to the flame as the flame's initial position. ;

[0052] S1046: Update the current moth position;

[0053] in, Indicates the first i A moth, Indicates the first j A flame, This represents the path function of a moth flying towards a flame, where the path function is a spiral function starting from the moth's position and ending at the flame's position. Indicates the first i Only the moth and the first j The distance between the flames b The constant representing the spiral, t This indicates how close the moth's next position is to the flame;

[0054] S1047: Reorder the updated moth positions and flame positions according to their fitness values, and select the position with the best fitness value as the position of the next generation flame.

[0055] S1048: Adopts an adaptive mechanism to reduce the number of flames;

[0056] S1049: Determine whether the current iteration number has reached the maximum iteration number. If the current iteration number is not less than the maximum iteration number, output the optimal wavelet packet threshold; otherwise, return to S1045.

[0057] It should be noted that the moth-to-flame optimization algorithm requires each moth to update its own position using the unique flame corresponding to it. Therefore, in the search space, the number of flames in the initial iteration is consistent with the size of the moth population.

[0058] It should be noted that by introducing the "moth to a flame" algorithm to select the optimal wavelet packet threshold, it has better continuity and smaller coefficient deviation compared to the four traditional selection rules for the optimal wavelet packet threshold.

[0059] S105: Construct the optimal wavelet packet threshold function based on the optimal wavelet packet threshold, where the optimal wavelet packet threshold function is either the optimal wavelet packet hard threshold function or the optimal wavelet packet soft threshold function, and obtain the estimated wavelet packet coefficients through the wavelet packet coefficients and the optimal wavelet packet threshold function.

[0060] In one possible implementation, S105 specifically includes:

[0061] The optimal wavelet packet hard thresholding function is:

[0062] in, This indicates the estimation of wavelet packet coefficients. Represents the wavelet packet coefficients. Represents a symbolic function. Indicates the optimal wavelet packet threshold;

[0063] The optimal wavelet packet soft thresholding function is:

[0064]

[0065] It should be noted that there are two types of optimal wavelet packet thresholding functions: the optimal wavelet packet hard thresholding function and the optimal wavelet packet soft thresholding function. The optimal wavelet packet hard thresholding function sets all wavelet packet coefficients below the optimal wavelet packet threshold to zero, while the optimal wavelet packet soft thresholding function sets all wavelet packet coefficients below the optimal wavelet packet threshold to zero, and reduces the number of wavelet packet coefficients above the optimal wavelet packet threshold accordingly.

[0066] S106: Perform inverse wavelet packet transform on the estimated wavelet packet coefficients to reconstruct the estimated wavelet packet coefficients into a denoised torque signal.

[0067] In one possible implementation, S106 specifically includes:

[0068] S1061: By performing the estimated wavelet packet coefficients according to the number of decomposition levels... n The denoised torque signal is obtained by performing an inverse wavelet packet transform.

[0069] It should be noted that the method of using the MFO-WPT algorithm to denoise the noisy torque signal can achieve the highest signal-to-noise ratio (SNR) and the lowest root mean square error (RMSE).

[0070] For example, the noisy torque signal collected by the torque sensor in the Atlas ETV STR61-70-13 torque wrench was used as the torque signal for verification of the MFO-WPT algorithm. The motor power was 1500W, the torque measurement range was 15-80 N·m, the sampling frequency was 1000Hz, and 1500 sampling points were selected for analysis.

[0071] A 30dB Gaussian white noise perturbation was added to the torque signal, and then comparative experiments were conducted in Matlab (Matrix Laboratory) using four traditional threshold selection rules and the WFO-WPT method of this invention. In the experiments, the number of iterations of the moth-to-flame optimization algorithm was set to 30, the population size to 20, the wavelet packet decomposition level to 4, the wavelet packet basis function to dB4, and a soft thresholding function was used to process the wavelet packet threshold. The experimental results are shown in Table 1.

[0072]

[0073] As can be seen from Table 1, the denoising effect of the WFO-WPT algorithm is better than the other methods in terms of both signal-to-noise ratio and root mean square error.

[0074] Among them, a larger signal-to-noise ratio (SNR) indicates a better denoising effect, and a smaller root mean square error (RMSE) indicates a smaller deviation between the denoised torque signal and the real torque signal.

[0075] In this embodiment of the invention, the acquired noisy torque signal is decomposed using wavelet packet thresholding. The Moth-Flame Optimization Algorithm (MFO) is used to optimize the selection of the wavelet packet threshold, accurately selecting the optimal threshold. A wavelet packet threshold function is constructed using the selected optimal threshold, and then the estimated wavelet packet coefficients are obtained using the wavelet packet threshold function. The wavelet packet coefficients are then inversely transformed to reconstruct the denoised torque signal. This method effectively denoises the noisy torque signal, resulting in high fidelity and distortion-free reconstruction, thereby improving the accuracy of torque signal characteristic waveforms in clinical diagnosis.

[0076] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method of denoising a torque signal, the method comprising: The method comprises the following steps: S101: acquiring a noisy torque signal; S102: decomposing the noisy torque signal at different decomposition scales to obtain wavelet packet coefficients of each node of the noisy torque signal at different scales; S103: obtaining a signal-to-noise ratio of the signal after denoising the wavelet packet coefficients, and taking the signal-to-noise ratio as a fitness function of a firefly optimization algorithm; S104: optimizing the wavelet packet threshold value by using the fitness function and the firefly optimization algorithm to find an optimal wavelet packet threshold value; S105: constructing an optimal wavelet packet threshold function according to the optimal wavelet packet threshold value, wherein the optimal wavelet packet threshold function is an optimal wavelet packet hard threshold function or an optimal wavelet packet soft threshold function, and obtaining estimated wavelet packet coefficients by using the wavelet packet coefficients and the optimal wavelet packet threshold function; S106: performing wavelet packet inverse transformation on the estimated wavelet packet coefficients to reconstruct the estimated wavelet packet coefficients into a denoised torque signal; The S103 specifically comprises: S1031: obtaining signal-to-noise ratio of the signal after denoising the wavelet packet coefficients : wherein, denotes the noisy torque signal, denotes the denoised torque signal, denotes the length of the torque signal sequence; S1032: taking the signal-to-noise ratio as the fitness function of the firefly optimization algorithm; The S104 specifically comprises: S1041: initialize the firefly optimization algorithm, set the current iteration number to 1, determine the wavelet packet threshold to be optimized parameters d , firefly population size r ; S1042: determining the upper and lower limits of the optimization of the wavelet packet threshold value; wherein, is j the wavelet packet threshold at scale, is j the length of the wavelet packet detail coefficients at scale, is the noise variance; S1043: Randomly generate firefly positions in the search space as a firefly initial population matrix , and calculate the fitness value vector corresponding to the firefly initial population matrix ; S1044: Constructing a flame position variable matrix according to the number of the moth initial population wherein the preliminary iteration number of the flame is same as the number of the moth initial population, and a fitness value vector corresponding to the flame position variable matrix is calculated : S1045: rank the moth positions in ascending order according to the size of the fitness values, and assign the ranked moth positions to the flame as initial positions of the flame; S1046: updating the current firefly position; wherein, represents the i flies only moth, represents the j flame, represents a path function of the moth flying towards the flame, wherein the path function is a spiral function with the moth position as the starting point and the flame position as the ending point, represents the i distance between only the moth and the j branch of the flame, b represents a constant of the spiral line, t represents the closeness of the next position of the moth to the flame; S1047: reordering the updated firefly position and the flame position according to the size of the fitness value, and selecting the position with the optimal fitness value as the position of the next generation of the flame; S1048: adopting an adaptive mechanism to reduce the number of the flames; S1049: determining whether the current iteration number reaches a maximum iteration number, and outputting the optimal wavelet packet threshold value when the current iteration number is not less than the maximum iteration number, otherwise, returning to S1045.

2. The method of claim 1, wherein, The S101 specifically comprises: S1011: collecting a torque signal by a torque sensor; S1012: amplifying the torque signal to obtain the noisy torque signal.

3. The method of claim 1, wherein, The S102 specifically comprises: S1021: selecting a decomposition level and a wavelet packet basis function; S1022: decomposing the noisy torque signal according to the decomposition level and the wavelet packet basis function; wherein, and denote the wavelet packet coefficients corresponding to the 2nd node of the 1st layer and the 2nd node of the (l+1)th layer, respectively, j wherein, n denote the wavelet packet coefficients corresponding to the 2nd node of the 1st layer and the 2nd node of the (l+1)th layer, respectively, j wherein, n denote the wavelet packet coefficients corresponding to the 2nd node of the 1st layer and the 2nd node of the (l+1)th layer, respectively, h and g denote the filter coefficients, n is the decomposition level.

4. The method of claim 3, wherein, The number of decomposition layers is n wherein, n is an integer and n is greater than or equal to 3 and less than or equal to 5.

5. The method of claim 3, wherein, The wavelet packet basis function belongs to db series and sym series basis functions.

6. The method of claim 1, wherein, The S1042 specifically comprises: S1042A: Introduce noise variance equation : wherein, denotes the median function, is the k-th j wavelet packet coefficient of the node at scale k, k q is a constant, chosen between 0.1 and 1.0.​ S1042B: take the value of the constant q with values of 0.1 and 1, the maximum wavelet packet threshold and the minimum wavelet packet threshold is calculated.

7. The method of claim 6, wherein, The S105 specifically comprises: The optimal wavelet packet hard threshold function is: The optimal wavelet packet soft threshold function is: wherein, denotes the estimated wavelet packet coefficient, denotes the wavelet packet coefficient, denotes the sign function, denotes the optimal wavelet packet threshold.

8. The method of claim 7, wherein, The S106 specifically comprises: S1061: obtaining a denoised torque signal by inverse wavelet packet transform of the estimated wavelet packet coefficients by decomposition level. n S1061: obtaining a denoised torque signal by inverse wavelet packet transform of the estimated wavelet packet coefficients by decomposition level.

Images