A multi-modal pulsed eddy current signal collaborative denoising method and related device

Abstract

The present application belongs to the technical field of pulse eddy current detection, and particularly relates to a multi-modal pulse eddy current signal cooperative denoising method and related equipment. A signal to be processed is acquired, an improved whale swarm optimization algorithm is used to take the signal envelope entropy as an optimization target, and adaptive optimization of the variational mode decomposition parameters is performed to obtain an optimal parameter combination. Based on the combination, the signal is subjected to variational mode decomposition to obtain a plurality of intrinsic mode components. A discrimination criterion is used to distinguish between signal dominant components and noise dominant components. The noise dominant components are subjected to singular value decomposition and wavelet threshold hierarchical denoising processing in sequence to obtain modified components. The signal dominant components and the modified components are reconstructed to obtain a denoised pulse eddy current signal. The method effectively suppresses noise and improves signal quality, and is suitable for the field of industrial nondestructive testing.

Description

Technical Field

[0001] This invention belongs to the field of pulsed eddy current detection technology, specifically involving a multimodal pulsed eddy current signal collaborative denoising method and related equipment. Background Technology

[0002] Pulsed eddy current testing technology, as an advanced non-destructive testing method, is widely used in industries such as petrochemicals, nuclear power, and shipbuilding. It can realize non-contact detection of defects in metal components. The quality of its sensor signal directly determines the defect detection accuracy. However, multi-source mixed noise in industrial environments can easily interfere with the detection signal, leading to a decrease in detection accuracy. Therefore, signal denoising technology has become a research focus in this field.

[0003] Currently, the denoising methods for pulsed eddy current sensor signals mainly rely on traditional signal decomposition techniques, among which Variational Mode Decomposition (VMD) is widely used. Parameters are usually set manually based on experience or optimized using a single algorithm. Related techniques largely depend on a single denoising model or a simple combination of models, and rarely incorporate the characteristics of noise components for accurate identification and classification.

[0004] The above methods do not systematically optimize the overall signal processing architecture and denoising link, and have obvious defects when dealing with complex multi-source mixed noise in industrial environments: First, VMD parameter optimization is prone to getting trapped in local optima, resulting in incomplete mode decomposition or aliasing; second, using a uniform denoising strategy for all modal components makes it difficult to distinguish between signal and noise-dominant components, which can easily cause signal distortion or noise residue; third, denoising performance is limited in strong noise environments, with insufficient improvement in signal-to-noise ratio, poor robustness, and difficulty in balancing noise suppression and signal feature preservation. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a multimodal pulsed eddy current signal collaborative denoising method and related equipment to address the shortcomings of the prior art, thereby solving the technical problem of incomplete noise suppression and signal distortion in traditional methods.

[0006] The objective of this invention is achieved through the following technical solutions: In a first aspect, the present invention provides a method for collaborative denoising of multimodal pulsed eddy current signals, comprising: Acquire the pulsed eddy current detection signal to be processed; An improved whale swarm optimization algorithm is used, with signal envelope entropy as the optimization objective, to adaptively optimize the parameters of the variational mode decomposition of the pulse eddy current detection signal to be processed, and obtain the optimal parameter combination. Based on the optimal parameter combination, variational mode decomposition is performed on the pulse eddy current detection signal to be processed to obtain several intrinsic mode components; Distinguish between the signal-dominant component and the noise-dominant component in each of the intrinsic mode components; The noise-dominant components are sequentially subjected to singular value decomposition and hierarchical denoising based on wavelet thresholds to obtain the noise-suppressed corrected components. The dominant component of the signal and all the modified components are reconstructed to obtain the denoised pulsed eddy current signal.

[0007] As a further improvement of the present invention, the improved whale swarm optimization algorithm includes: applying dynamic neighborhood perturbation to the current optimal solution during the search process, and using an equidistant spiral structure as the position update trajectory of the search individuals; the dynamic neighborhood perturbation calculation expression is:

[0008] In the formula, For the first t At the +1st iteration, the... d The location of the searched individual; For the first t During the nth iteration d The location of the searched individual; This is the disturbance amplitude control coefficient; The scale factor for normalizing the perturbation vector; The perturbation step size; Rand (-1,1) is a uniform random variable in the interval (-1,1). For the current number t The globally optimal search position at the next iteration.

[0009] As a further improvement of the present invention, the step of using signal envelope entropy as the optimization objective includes calculating the average envelope entropy of the modal components obtained by variational mode decomposition under different candidate parameters, and performing iterative optimization with the goal of minimizing the average envelope entropy; the expression for the average envelope entropy is:

[0010] In the formula, The average envelope entropy of the modal components obtained from variational mode decomposition; For the first i The probability distribution of the envelope of each modality; This represents the number of current candidate modes.

[0011] As a further improvement of the present invention, variational mode decomposition is performed on the pulsed eddy current detection signal to obtain several intrinsic mode components, including:

[0012] In the formula, or The first after VMD decomposition One IMF component; For the first The instantaneous frequency of each IMF component; δ ( t ) is the Dirac function; To The Hilbert transform performed; The pulsed eddy current detection signal to be processed; For Lagrange-constrained inner product terms; or For Lagrange multipliers; It is the imaginary unit.

[0013] As a further improvement of the present invention, the method of distinguishing between the signal-dominant component and the noise-dominant component in each of the intrinsic mode components includes: Calculate the correlation coefficient between any two intrinsic mode components to serve as a measure of modal correlation. Intrinsic mode components with modal correlation metrics higher than a preset threshold are identified as signal-dominant components, while intrinsic mode components with modal correlation metrics lower than the preset threshold are identified as noise-dominant components.

[0014] As a further improvement of the present invention, singular value decomposition is performed on the dominant noise component, including: The dominant noise components are constructed as Hankel matrices; Singular value decomposition is performed on the Hankel matrix to obtain a sequence of singular values; A first threshold is determined based on the singular value sequence, and singular values ​​smaller than the first threshold are reset to zero. The matrix is ​​reconstructed using the processed singular value sequence to obtain the corrected components after initial noise suppression.

[0015] As a further improvement of the present invention, the hierarchical denoising process based on wavelet thresholding involves performing wavelet transform on the signal after singular value decomposition and then performing threshold quantization and reconstruction on the obtained wavelet coefficients to further suppress local transient noise.

[0016] Secondly, the present invention provides a multimodal pulsed eddy current signal collaborative denoising system, comprising: The signal acquisition module is used to acquire the pulsed eddy current detection signal to be processed; The parameter optimization module is used to adaptively optimize the parameters of the variational mode decomposition of the pulse eddy current detection signal to be processed, using an improved whale swarm optimization algorithm with signal envelope entropy as the optimization target, to obtain the optimal parameter combination. The mode decomposition module is used to perform variational mode decomposition on the pulse eddy current detection signal to be processed based on the optimal parameter combination to obtain several intrinsic mode components. The noise identification module is used to distinguish between the signal-dominant component and the noise-dominant component in each of the intrinsic mode components. The noise reduction module is used to sequentially perform singular value decomposition and wavelet threshold-based hierarchical noise reduction on the noise-dominant components to obtain the noise-suppressed corrected components. The signal reconstruction module is used to reconstruct the dominant component of the signal and all the modified components to obtain the denoised pulsed eddy current signal.

[0017] Thirdly, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the multimodal pulsed eddy current signal collaborative denoising method as described above.

[0018] Fourthly, the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed as described above for the multimodal pulsed eddy current signal collaborative denoising method.

[0019] The beneficial effects of this invention are as follows: This invention provides a collaborative denoising method for multimodal pulsed eddy current signals. Through an improved whale swarm optimization algorithm, it adaptively optimizes variational mode decomposition parameters using signal envelope entropy as the optimization objective. This overcomes the subjectivity of traditional parameter setting, which relies on experience, achieving precise parameter matching of signal characteristics and improving decomposition efficiency and accuracy. Based on the optimal parameter combination, variational mode decomposition is performed to obtain intrinsic mode components, which better reflects the actual characteristics of pulsed eddy current signals compared to fixed parameter decomposition, effectively avoiding mode aliasing. A discrimination criterion distinguishes between the dominant signal component and the dominant noise component, targeting the noise component rather than processing the entire signal. Local noise reduction preserves more effective signal details, solving the problem of excessive filtering of effective signals in traditional methods. Singular value decomposition is used to suppress global noise interference on the noise-dominant component, and wavelet thresholding is used to finely filter out noise of different frequencies. The two work together to achieve multi-level noise suppression, which is more thorough than single noise reduction methods. Finally, the dominant and modified components of the signal are reconstructed. The reconstructed signal has both high signal-to-noise ratio and low distortion, which solves the technical problems of incomplete noise suppression and signal distortion in traditional methods. It significantly improves the quality of pulsed eddy current signals in industrial non-destructive testing scenarios and meets the requirements of high-precision testing. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the multimodal pulsed eddy current signal collaborative denoising method in an embodiment of the present invention; Figure 2 This is a comparison diagram of search paths in embodiments of the present invention; Figure 3 This is a performance comparison chart of IWOA with other optimization algorithms in an embodiment of the present invention; Figure 4 These are frequency domain diagrams of different modal components after VMD decomposition in an embodiment of the present invention; Figure 5 These are comparison results of different models under weak noise conditions in the embodiments of the present invention; Figure 6 These are comparison results of different models under moderate noise conditions in the embodiments of the present invention; Figure 7 These are comparison results of different models under strong noise conditions in the embodiments of the present invention; Figure 8 This is a schematic diagram of the internal structure of a computer device in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives and technical solutions of this invention clearer and easier to understand, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0023] Terminology Explanation: VMD, Variational Mode Decomposition.

[0024] IWOA, Improved Whale Optimization Algorithm, is a metaheuristic optimization algorithm that improves upon the classic Whale Optimization Algorithm (WOA) by introducing improved strategies such as dynamic neighborhood perturbation and equidistant spiral position updates.

[0025] IMF, Intrinsic Mode Function: A finite-bandwidth, single-component oscillating signal obtained after a signal is decomposed by VMD.

[0026] WOA, Whale Optimization Algorithm.

[0027] DBO, Dung Beetle Optimizer: Dung Beetle Optimizer.

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. The described embodiments are only some embodiments of the present invention, and not all embodiments.

[0029] Example 1 Existing pulsed eddy current signal denoising techniques mostly employ traditional variational mode decomposition, relying on manual experience to set parameters or a single optimization algorithm for optimization, without constructing a systematic denoising architecture for multi-source mixed noise. Firstly, traditional optimization algorithms are prone to getting trapped in local optima when optimizing variational mode decomposition parameters, leading to mode aliasing and incomplete decomposition. Secondly, existing joint denoising methods use a uniform processing strategy for all modal components, failing to accurately distinguish between signal and noise-dominant components, easily causing signal distortion or noise residue. Thirdly, denoising performance bottlenecks are significant in strong noise environments, with limited improvement in signal-to-noise ratio and poor robustness, making it difficult to achieve complete noise suppression while preserving key signal characteristics, and thus failing to meet the high-precision requirements of non-destructive testing of metal grounding bodies in complex industrial electromagnetic environments.

[0030] In summary, existing pulsed eddy current signal denoising techniques suffer from the following technical problems: denoising algorithms lack adaptive capabilities and are difficult to automatically adjust processing strategies based on signal characteristics; parameter selection relies on human experience, making it impossible to achieve globally optimal parameter combinations; it is difficult to achieve a good balance between preserving signal feature information and effectively suppressing noise during the denoising process; and there is a lack of effective coordination mechanisms among various denoising methods, making it impossible to fully leverage the advantages of different methods.

[0031] To address the above problems, this invention provides a collaborative denoising method for multimodal pulsed eddy current signals, such as... Figure 1 The illustrated process organically combines various techniques, including an improved whale swarm optimization algorithm, variational mode decomposition, singular value decomposition, and wavelet thresholding, to construct a complete collaborative denoising framework for pulsed eddy current signals. This method adaptively determines the optimal parameters for variational mode decomposition, accurately distinguishes between the dominant signal components and noise-dominant components, and employs different denoising strategies accordingly, ultimately achieving high-quality denoising of pulsed eddy current detection signals.

[0032] S1: Acquire the pulsed eddy current detection signal to be processed.

[0033] Specifically, a pulsed eddy current testing system is used to scan and inspect the metal component under test, acquiring raw detection signals containing defect information. The raw detection signal is the response signal obtained by the receiving coil after the pulsed magnetic field generated by the excitation coil acts on the metal component under test.

[0034] In some embodiments of this application, the pulsed eddy current detection system includes a pulse excitation source, an excitation coil, a receiving coil, and a signal acquisition module. The pulse excitation source generates a pulsed current with a specific frequency and amplitude, and the excitation coil generates a pulsed magnetic field in space after being energized. When the pulsed magnetic field acts on the metal component under test, an induced current, i.e., eddy current, is generated inside the metal component according to the principle of electromagnetic induction. The magnitude and distribution of the eddy current are closely related to the conductivity, permeability, and internal defect conditions of the metal component. The receiving coil senses the secondary magnetic field generated by the eddy current and converts it into an electrical signal output. The signal acquisition module performs preprocessing such as amplification, filtering, and analog-to-digital conversion on the electrical signal to obtain a digitized pulsed eddy current detection signal.

[0035] The pulsed eddy current detection signal to be processed is typically a one-dimensional time-domain signal sequence, represented as follows: ,in This indicates the sampling time point. In actual data acquisition, the sampling frequency must satisfy the Nyquist sampling theorem to ensure that high-frequency components in the signal are not aliased. Typically, the sampling frequency of the pulsed eddy current detection signal is set to at least ten times the excitation pulse frequency to ensure the signal's time resolution and spectral integrity.

[0036] The acquired pulsed eddy current detection signals mainly contain useful signal components such as defect echo signals, edge response signals, and characteristic signals related to material properties. Noise components include white noise, powder noise, and 1 / Noise and various types of interference noise. These noises often overlap with the useful signal in both the time and frequency domains.

[0037] S2: An improved whale swarm optimization algorithm is adopted, with the signal envelope entropy as the optimization target, to adaptively optimize the parameters of the variational mode decomposition of the pulse eddy current detection signal to be processed, and obtain the optimal parameter combination.

[0038] After obtaining the pulsed eddy current detection signal to be processed, this application adopts an improved whale swarm optimization algorithm (IWOA) to adaptively optimize the key parameters of variational mode decomposition with the signal envelope entropy as the optimization objective.

[0039] It should be noted that the whale pod optimization algorithm is a metaheuristic optimization algorithm that simulates the hunting behavior of humpback whales.

[0040] In some embodiments of this application, the improved whale swarm optimization algorithm enhances its global search capability and local search accuracy by applying dynamic neighborhood perturbation to the current optimal solution during the search process and employing an equidistant spiral structure as the position update trajectory of individual search entities. Specifically, the calculation expression for the dynamic neighborhood perturbation is as follows:

[0041] In the formula, For the first t At the +1st iteration, the... d The location of the searched individual; For the first t During the nth iteration d The location of the searched individual; This is the disturbance amplitude control coefficient, used to adjust the strength of the disturbance. The scale factor for normalizing the perturbation vector; The perturbation step size; Rand (-1,1) is a uniform random variable in the interval (-1,1), which is the source of uncertainty in the direction of the disturbance; For the first t The globally optimal search position at the next iteration.

[0042] Through the aforementioned dynamic neighborhood perturbation mechanism, the algorithm can introduce a certain degree of random perturbation while maintaining the guiding effect of the current optimal solution, effectively preventing the search process from getting trapped in local optima. (Perturbation amplitude control coefficient) and perturbation step size The introduction of this feature allows the perturbation intensity to be adaptively adjusted according to the search progress. Larger perturbations are used in the early stages of the search to enhance global exploration capabilities, while smaller perturbations are used in the later stages to improve local mining accuracy. (Perturbation amplitude control coefficient) and perturbation step size The value can be adjusted according to the specific signal characteristics. As an optional implementation method, The value can be set from 0.5 to 2.0. The value of u can be set from 0.1 to 1.0. Larger u values ​​and A larger value enhances the algorithm's global exploration capability, while a smaller value improves local mining accuracy. The algorithm can adaptively adjust these two parameters based on the current iteration number, using larger values ​​in the early stages of the search to fully explore the solution space, and smaller values ​​in the later stages to refine the search of the region near the optimal solution.

[0043] To ensure effective global exploration, the traditional logarithmic spiral structure with decreasing scale is abandoned in favor of an equidistant Archimedean spiral as the overall trajectory update, as expressed below:

[0044] In the formula, Represents a random variable in the interval [-1, 1]. θ This is the current offset angle. a This is the initial position offset coefficient. b A constant that limits the size of the spiral. δ To control the parameters of the correction intensity, To correct the period coefficient, To search for the x-coordinate component of an individual's two-dimensional position on a spiral trajectory, To search for the ordinate component of the two-dimensional position of an individual on a spiral trajectory.

[0045] When updating the position of the search individual using an equidistant spiral structure, the algorithm simulates the spiral hunting path of a humpback whale, approximating the optimal solution region in a spiral shape. The parameters of the spiral can be adjusted according to the actual problem to balance the convergence speed and the quality of the solution. This position update strategy allows the search individual to explore favorable regions in the solution space more effectively, increasing the probability of the algorithm finding the global optimum. By comparing the equidistant Archimedean spiral with three other search paths—logarithmic curve (logarithmic spiral), rose curve, and Fermat curve (Fermat spiral)—and testing them using a test function, the results are as follows: Figure 2 As shown, compared with logarithmic curves, Fermat curves, etc., the equidistant Archimedes spiral curve can escape local optima faster in the early stage of iteration and has higher convergence accuracy in the later stage, verifying the reliability of the Archimedes spiral curve for parameter optimization.

[0046] In some embodiments of this application, signal envelope entropy is used as the objective function for iterative optimization. Signal envelope entropy is an important indicator for measuring signal complexity and stationarity. When the parameters of variational mode decomposition are chosen reasonably, the resulting modal components have good physical meaning and low envelope entropy values. When the parameters are not chosen properly, the decomposition results may have problems such as mode aliasing or over-decomposition, leading to an increase in envelope entropy values.

[0047] Specifically, the average envelope entropy of the modal components obtained from variational mode decomposition under different candidate parameters is calculated, and iterative optimization is performed with the goal of minimizing the average envelope entropy. The expression for the average envelope entropy is:

[0048] In the formula, The average envelope entropy of the modal components obtained from variational mode decomposition; For the first i The probability distribution of the envelope of each modality; This represents the current number of candidate modes. A smaller objective function value indicates a higher concentration of modal energy and better decomposition quality. Through this optimization process, the VMD modal components best suited to the current noise environment can be obtained. K and penalty factor α . Figure 3 This is a performance comparison chart of IWOA with other optimization algorithms. Figure 3 This indicates that the IWOA algorithm converges to a lower average envelope entropy with fewer iterations, demonstrating that its optimization speed and parameter matching ability are significantly better than those of the comparison algorithms such as WOA and DBO.

[0049] The improved whale swarm optimization algorithm uses the average envelope entropy as the fitness function. It iteratively updates the positions of search individuals, gradually searching for the parameter combination that minimizes the average envelope entropy. In each iteration, the algorithm evaluates the search individuals based on their fitness values ​​and generates new candidate solutions according to the improved position update rule. After a predetermined number of iterations, the algorithm converges and outputs the optimal parameter combination, including the number of decomposition levels. And key parameters such as the penalty factor α.

[0050] As an optional implementation, the improved whale swarm optimization algorithm also incorporates a premature convergence detection mechanism. When the optimal fitness value fails to improve for several consecutive generations, a mutation operation is triggered to increase population diversity and avoid getting trapped in local optima. Specifically, the premature convergence detection mechanism works as follows: during iterative optimization, if the difference in optimal fitness values ​​between adjacent generations continuously reaches a preset number of generations and is always less than the set improvement tolerance, the algorithm is considered to have prematurely converged and a mutation operation is triggered. At this point, the algorithm applies a specific mathematical perturbation (such as Cauchy mutation perturbation) to the currently trapped global optimum to generate a new individual. Based on a greedy strategy, individuals with better fitness are retained for the next iteration. This utilizes the mutation perturbation to give stagnant individuals a step size to escape the current local extreme region, effectively preventing variational mode decomposition parameter optimization from getting trapped in local optima and further improving the accuracy of the global search.

[0051] In addition, the algorithm can also adopt an adaptive adjustment strategy to dynamically adjust the algorithm's control parameters based on the changing trend of the current iteration number and the optimal fitness value.

[0052] S3: Based on the optimal parameter combination, variational mode decomposition is performed on the pulsed eddy current detection signal to obtain several intrinsic mode components.

[0053] Variational mode decomposition (VMD) is a non-recursive signal decomposition method. Its core idea is to represent the signal to be decomposed as the sum of several eigenmode components with finite bandwidths, and then solve for each eigenmode component and its corresponding center frequency using a variational optimization framework. Compared with empirical mode decomposition (EMD), VMD effectively avoids mode aliasing and provides better physical interpretability of the decomposition results.

[0054] In some embodiments of this application, the variational problem of variational mode decomposition can be expressed as the following constrained optimization form:

[0055] In the formula, or The first after VMD decomposition K One IMF component; For the first The instantaneous frequency of each IMF component; δ ( t ) is the Dirac function; To The Hilbert transform performed; The pulsed eddy current detection signal to be processed.

[0056] The variational optimization problem described above can be solved using the alternating direction multiplier algorithm. The specific steps include: initializing each eigenmode component and its center frequency; alternately updating each eigenmode component and its center frequency; updating the Lagrange multipliers; and repeating the above steps until convergence.

[0057] In the formula, u k The first after VMD decomposition One IMF component; ω k For the first The instantaneous frequency of each IMF component; δ ( t ) is the Dirac function; It is the Hilbert transform; The pulsed eddy current detection signal to be processed; For Lagrange-constrained inner product terms; or For Lagrange multipliers; The imaginary unit. This decomposition yields the following: Figure 4 The modal components shown are well isolated in the frequency domain after decomposition (IMF1~IMF5), with no mode aliasing. The dominant signal components are effectively preserved in the low-frequency band, and noise is successfully removed in the high-frequency band.

[0058] S4: Distinguish between the signal-dominant component and the noise-dominant component in each intrinsic mode component.

[0059] The intrinsic mode components obtained after variational mode decomposition contain both useful defect echo signal components and various noise components. In order to achieve targeted denoising, it is necessary to distinguish between signal dominance and noise dominance for each component.

[0060] In some embodiments of this application, the process of distinguishing between signal-dominant components and noise-dominant components in each intrinsic mode component includes the following steps: First, calculating the average value of the correlation coefficients between each intrinsic mode component and the remaining components to serve as a modal correlation measure; then, determining intrinsic mode components with modal correlation measures higher than a preset threshold as signal-dominant components, and determining intrinsic mode components with modal correlation measures lower than the preset threshold as noise-dominant components.

[0061] In determining the dominant signal component, defect echo signals often exhibit strong energy continuity and phase consistency among adjacent modal components, resulting in typically high correlation coefficients between these components. In contrast, noise components are randomly distributed across different modal components, leading to weaker correlations between adjacent components. Based on this characteristic, setting a correlation coefficient threshold can effectively distinguish between dominant signal and noise components.

[0062] The threshold can be adjusted according to the specific characteristics of the pulsed eddy current detection signal and the requirements of engineering applications. As a preferred implementation, the preset threshold can be set to 0.5. When the average correlation coefficient between a certain intrinsic mode component and the other components is higher than 0.5, the component is determined to be the dominant signal component; conversely, when the correlation coefficient is lower than 0.5, it is determined to be the dominant noise component.

[0063] As an alternative implementation, the energy ratio criterion can be used for auxiliary determination. The dominant signal component typically has a higher energy proportion, while the dominant noise component has a lower energy proportion. By comprehensively considering both the correlation coefficient and the energy ratio, the reliability of the determination result can be further improved.

[0064] S5: The noise-dominant components are sequentially subjected to singular value decomposition and hierarchical denoising based on wavelet thresholds to obtain the noise-suppressed corrected components.

[0065] In some embodiments of this application, singular value decomposition of the noise-dominant component includes the following steps: constructing the noise-dominant component into a Hankel matrix; performing singular value decomposition on the Hankel matrix to obtain a sequence of singular values; determining a first threshold based on the distribution characteristics of the singular value sequence, and zeroing out singular values ​​smaller than the first threshold; and reconstructing the matrix using the processed singular value sequence to obtain the corrected component after initial noise suppression.

[0066] The specific method for constructing the noise dominant component into a Hankel matrix is ​​as follows: Let the noise dominant component be a one-dimensional signal sequence x, and construct the noise dominant component of the one-dimensional signal sequence x into a Hankel matrix form. The Hankel matrix is ​​an (N-m+1)×m matrix, where m is a construction parameter, typically around N / 2. The elements on the diagonal of the Hankel matrix are identical.

[0067] In an ideal scenario, a sequence of singular values ​​in a noise-free signal exhibits a distinct step characteristic, where the first few larger singular values ​​represent the main energy components of the signal, while the remaining smaller singular values ​​approach zero. When the signal is contaminated with noise, all singular values ​​are affected, and the noise component causes the smaller singular values ​​to increase.

[0068] Based on the above analysis, this application employs a singular value thresholding method for denoising. Specifically, a first threshold is determined based on the distribution characteristics of the singular value sequence. When a singular value is less than the first threshold, it is set to zero. The first threshold can be determined using the following methods: finding the inflection point position based on the difference curve of the singular value sequence, retaining the singular values ​​before the inflection point, and setting the singular values ​​after the inflection point to zero; or determining the threshold based on the singular value energy contribution rate, where the singular value corresponding to the cumulative energy contribution reaching a certain proportion is the threshold. This embodiment uses the singular value difference spectrum method, setting the singular value corresponding to the maximum abrupt change point of the difference curve as the first threshold. The processed singular value sequence and singular vector matrix are reconstructed to obtain the preliminarily denoised Hankel matrix. By averaging the anti-diagonal elements of the Hankel matrix, the one-dimensional denoised signal can be restored.

[0069] The Hankel matrix is ​​decomposed as follows:

[0070] In the formula, It is a Hankel matrix; It is a left singular matrix; V It is a right singular matrix. Singular value matrix The sequence formed.

[0071] The singular value matrix is:

[0072] The reconstructed Hankel matrix after initial denoising is as follows:

[0073] In the formula, Singular value matrix The sequence formed.

[0074] The signal after singular value decomposition is further subjected to hierarchical denoising based on wavelet threshold to further suppress local transient noise.

[0075] While singular value decomposition (SVD) can effectively remove most of the noise components from the dominant noise components, its effect on suppressing certain local transient noises is limited. Therefore, this embodiment further employs wavelet thresholding denoising to process the signal after SVD, in order to achieve precise suppression of local transient noise.

[0076] In some embodiments of this application, the hierarchical denoising process based on wavelet thresholds includes the following steps: performing wavelet transform on the signal after singular value decomposition to obtain wavelet decomposition coefficients; performing threshold quantization on the wavelet coefficients using an adaptive threshold strategy; and performing inverse wavelet transform on the processed wavelet coefficients to reconstruct the further denoised corrected components.

[0077] Wavelet transform can analyze signals at different scales and locations, making it particularly suitable for capturing local transient features of signals. Commonly used wavelet basis functions include the Daubechies wavelet, Symlet wavelet, and Coiflet wavelet. In the specific implementation of this invention, a suitable wavelet basis function can be selected based on the characteristics of the pulse eddy current detection signal. As a preferred embodiment, the Daubechies wavelet db4 is selected as the wavelet basis function. This wavelet has good regularity and compact support characteristics, making it suitable for processing pulse-like signals.

[0078] The key to wavelet thresholding denoising lies in the selection of the threshold and the quantization strategy. The hierarchical thresholding strategies employed in this invention include the following: the first is a hard thresholding strategy, which keeps the wavelet coefficients unchanged when their absolute values ​​are greater than the threshold, otherwise sets them to zero; the second is a soft thresholding strategy, which performs shrinkage processing when the absolute values ​​of the wavelet coefficients are greater than the threshold; and the third is an adaptive thresholding strategy, which dynamically adjusts the threshold based on the number of decomposition levels and the noise level. In some embodiments of this application, an improved adaptive thresholding strategy is used. This improved adaptive thresholding strategy means that the threshold is adaptively reduced layer by layer according to the noise variance of different decomposition levels, rather than using a globally uniform threshold.

[0079] By setting hierarchical thresholds, higher-level decomposition uses larger thresholds to preserve the main features of the signal, while lower-level decomposition uses smaller thresholds to protect the detailed components of the signal. This hierarchical strategy effectively balances the trade-off between denoising performance and signal detail preservation.

[0080] S6: Reconstruct the dominant component of the signal with all the modified components to obtain the denoised pulsed eddy current signal.

[0081] After the aforementioned steps, both the dominant signal component and the dominant noise component have been denoised. At this point, the processed components need to be reconstructed to obtain the complete pulsed eddy current signal.

[0082] In some embodiments of this application, the reconstruction process includes: keeping the determined dominant signal component unchanged; using the noise dominant component after singular value decomposition and wavelet threshold denoising as the correction component; and summing all the dominant signal components and the correction component to reconstruct the denoised pulse eddy current signal.

[0083] The reconstructed signal not only effectively suppresses various noise interferences but also retains useful signal components such as defect echoes relatively well. Evaluation metrics for denoising performance can include signal-to-noise ratio improvement, root mean square error reduction rate, and defect feature preservation. To verify the performance of the method in this embodiment under different signal-to-noise ratio conditions, this embodiment also sets three different intensities of random noise in the verification phase and compares this method with other methods. The three different intensities of random noise include: weak noise (signal-to-noise ratio of 5dB, such as...) Figure 5 As shown), moderate noise (signal-to-noise ratio of -10dB, such as Figure 6 (as shown) and strong noise (signal-to-noise ratio of -25dB, such as) Figure 7 (As shown). The software model formed using the method of this embodiment was tested using simulation software to evaluate the denoising effect of the three noise levels on the pulse eddy current signal under different noise environments. The results are shown below. Figure 5 , Figure 6 , Figure 7 As shown, under any noisy environment, the denoised signal of this application is closer to the ideal signal, and the denoising effect is superior to other denoising methods. By comparing the denoised signal generated by the method of this embodiment with the initial noisy signal, the ideal noiseless signal, and the denoising results of other comparative algorithms such as WOA-VMD and DBO-VMD-WTD, the results show that under any noisy environment, the denoised signal curve reconstructed by the method of this application is closest to the ideal signal. This indicates that this method effectively suppresses multi-source mixed noise while minimizing the distortion of the useful signal, and its denoising effect and robustness are significantly better than other comparative methods. Specifically, Figure 5 As shown, when the interference is small, the differences in the performance of each algorithm are relatively small, and they can all roughly restore the signal outline. Figure 6 As shown, the differences begin to emerge, and the curves of the comparison algorithms (such as WOA-VMD) begin to become coarser and gradually deviate from the ideal dotted line. Figure 7As shown, under extremely harsh and strong noise conditions, the curves of conventional comparison algorithms will exhibit severe distortion or significant spikes, while the curves of this invention (IWOA-VMD-SVD-WTD) still fit the ideal dotted line, which intuitively demonstrates the strong robustness and high fidelity of this method in dealing with complex industrial electromagnetic environments.

[0084] The multimodal pulsed eddy current signal collaborative denoising method proposed in this application organically combines various techniques such as improved whale swarm optimization algorithm, variational mode decomposition, singular value decomposition, and wavelet threshold denoising to form a complete framework for pulsed eddy current signal collaborative denoising.

[0085] An improved whale swarm optimization algorithm is employed to adaptively optimize the variational mode decomposition parameters, effectively addressing the problems of traditional methods that rely on manual experience for parameter selection and lack systematic optimization strategies. Through dynamic neighborhood perturbation and equidistant spiral position update strategies, the algorithm's global search capability and local search accuracy are enhanced, enabling reliable convergence to the optimal parameter combination that minimizes the signal envelope entropy.

[0086] Variational mode decomposition (VMD) decomposes the original signal into several physically meaningful eigenmode components, creating favorable conditions for subsequent separation of signal and noise components. The variational optimization framework of VMD can adaptively separate different frequency components in the signal into different mode components, effectively avoiding the mode aliasing problem present in empirical mode decomposition.

[0087] The method distinguishes between the dominant signal component and the dominant noise component by calculating the correlation coefficient between each modal component. This method has clear physical meaning and high accuracy. The dominant signal component retains the characteristic information and energy distribution of useful signals such as defect echoes, while the dominant noise component contains various noise components, laying the foundation for targeted denoising.

[0088] Finally, a two-stage denoising strategy combining singular value decomposition (SVD) and wavelet thresholding is employed for the dominant noise components. SVD effectively removes most background and random noise, while wavelet thresholding further suppresses local transient and residual noise. This organic combination of the two denoising methods leverages their respective advantages while mitigating the shortcomings of a single method, achieving effective noise suppression and good preservation of signal characteristics.

[0089] Example 2 This embodiment provides a multimodal pulsed eddy current signal collaborative denoising system, including: The signal acquisition module is used to acquire the pulsed eddy current detection signal to be processed; The parameter optimization module is used to adaptively optimize the parameters of variational mode decomposition using an improved whale swarm optimization algorithm with signal envelope entropy as the optimization objective, so as to obtain the optimal parameter combination. The mode decomposition module is used to perform variational mode decomposition on the pulsed eddy current detection signal based on the optimal parameter combination to obtain several intrinsic mode components. The noise identification module is used to distinguish between the dominant signal component and the dominant noise component in each intrinsic mode component. The noise reduction module is used to sequentially perform singular value decomposition and wavelet threshold-based hierarchical noise reduction on the noise-dominant components to obtain the noise-suppressed corrected components. The signal reconstruction module is used to reconstruct the dominant component of the signal with all the modified components to obtain the denoised pulsed eddy current signal.

[0090] Specific limitations regarding the multimodal pulsed eddy current signal collaborative denoising system can be found in the limitations of the multimodal pulsed eddy current signal collaborative denoising method described above; the corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned multimodal pulsed eddy current signal collaborative denoising system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0091] Example 3 Figure 8 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 8 As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement a multimodal pulse eddy current signal collaborative denoising method. The display screen can be an LCD screen or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device casing, or an external keyboard, touchpad, or mouse.

[0092] As will be understood by those skilled in the art, computer equipment Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0093] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0094] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0095] In summary, the multimodal pulsed eddy current signal collaborative denoising method, system, computer device, and storage medium provided in this application overcome the limitations of traditional single denoising methods. It innovatively integrates multiple techniques, including an improved whale swarm optimization algorithm, variational mode decomposition (VMD), singular value decomposition (SVD), and wavelet thresholding denoising. Furthermore, it creatively applies the improved whale swarm optimization algorithm to the optimization process of VMD parameters, using signal envelope entropy as the objective function to achieve intelligent selection of decomposition parameters. VMD can automatically adjust parameter settings according to different signal characteristics, significantly improving the adaptability and versatility of the denoising algorithm. Secondly, singular value decomposition and wavelet thresholding are sequentially applied to the noise-dominant components, forming an efficient two-stage denoising process. Singular value decomposition effectively removes diffuse background noise, while wavelet thresholding finely suppresses local transient noise; the combination of the two achieves comprehensive suppression of various types of noise.

[0096] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0097] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for collaborative denoising of multimodal pulsed eddy current signals, characterized in that, include: Acquire the pulsed eddy current detection signal to be processed; An improved whale swarm optimization algorithm is used, with signal envelope entropy as the optimization objective, to adaptively optimize the parameters of the variational mode decomposition of the pulse eddy current detection signal to be processed, and obtain the optimal parameter combination. Based on the optimal parameter combination, variational mode decomposition is performed on the pulse eddy current detection signal to be processed to obtain several intrinsic mode components; Distinguish between the signal-dominant component and the noise-dominant component in each of the intrinsic mode components; The noise-dominant components are sequentially subjected to singular value decomposition and hierarchical denoising based on wavelet thresholds to obtain the noise-suppressed corrected components. The dominant component of the signal and all the modified components are reconstructed to obtain the denoised pulsed eddy current signal.

2. The multimodal pulsed eddy current signal collaborative denoising method according to claim 1, characterized in that, The improved whale swarm optimization algorithm includes: applying dynamic neighborhood perturbation to the current optimal solution during the search process, and using an equidistant spiral structure as the position update trajectory of the search individuals; the dynamic neighborhood perturbation calculation expression is: In the formula, For the first t At the +1st iteration, the... d The location of the searched individual; For the first t During the nth iteration d The location of the searched individual; This is the disturbance amplitude control coefficient; The scale factor for normalizing the perturbation vector; The perturbation step size; Rand (-1,1) is a uniform random variable in the interval (-1,1). For the first t The globally optimal search position at the next iteration.

3. The multimodal pulsed eddy current signal collaborative denoising method according to claim 1, characterized in that, The optimization objective, which uses signal envelope entropy, includes calculating the average envelope entropy of the modal components obtained from variational mode decomposition under different candidate parameters, and performing iterative optimization with the goal of minimizing the average envelope entropy; the expression for the average envelope entropy is: In the formula, The average envelope entropy of the modal components obtained from variational mode decomposition; For the first i The probability distribution of the envelope of each modality; This represents the number of current candidate modes.

4. The multimodal pulsed eddy current signal collaborative denoising method according to claim 1, characterized in that, Variational mode decomposition is performed on the pulsed eddy current detection signal to obtain several intrinsic mode components, including: In the formula, or The first after VMD decomposition One IMF component; For the first The instantaneous frequency of each IMF component; δ ( t ) is the Dirac function; To The Hilbert transform performed; The pulsed eddy current detection signal to be processed; For Lagrange-constrained inner product terms; or For Lagrange multipliers; It is the imaginary unit.

5. The multimodal pulsed eddy current signal collaborative denoising method according to claim 4, characterized in that, Distinguishing between the signal-dominant component and the noise-dominant component in each of the intrinsic mode components includes: Calculate the correlation coefficient between any two intrinsic mode components to serve as a measure of modal correlation. Intrinsic mode components with modal correlation metrics higher than a preset threshold are identified as signal-dominant components, while intrinsic mode components with modal correlation metrics lower than the preset threshold are identified as noise-dominant components.

6. The multimodal pulsed eddy current signal collaborative denoising method according to claim 5, characterized in that, Singular value decomposition is performed on the dominant noise components, including: The dominant noise components are constructed as Hankel matrices; Singular value decomposition is performed on the Hankel matrix to obtain a sequence of singular values; A first threshold is determined based on the singular value sequence, and singular values ​​smaller than the first threshold are reset to zero. The matrix is ​​reconstructed using the processed singular value sequence to obtain the corrected components after initial noise suppression.

7. The multimodal pulsed eddy current signal collaborative denoising method according to claim 6, characterized in that, The wavelet threshold-based hierarchical denoising process involves performing wavelet transform on the signal after singular value decomposition, and then threshold quantizing and reconstructing the obtained wavelet coefficients to further suppress local transient noise.

8. A multimodal pulsed eddy current signal collaborative denoising system, characterized in that, include: The signal acquisition module is used to acquire the pulsed eddy current detection signal to be processed; The parameter optimization module is used to adaptively optimize the parameters of the variational mode decomposition of the pulse eddy current detection signal to be processed, using an improved whale swarm optimization algorithm with signal envelope entropy as the optimization target, to obtain the optimal parameter combination. The mode decomposition module is used to perform variational mode decomposition on the pulse eddy current detection signal to be processed based on the optimal parameter combination to obtain several intrinsic mode components. The noise identification module is used to distinguish between the signal-dominant component and the noise-dominant component in each of the intrinsic mode components. The noise reduction module is used to sequentially perform singular value decomposition and wavelet threshold-based hierarchical noise reduction on the noise-dominant components to obtain the noise-suppressed corrected components. The signal reconstruction module is used to reconstruct the dominant component of the signal and all the modified components to obtain the denoised pulsed eddy current signal.

9. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the multimodal pulsed eddy current signal collaborative denoising method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 7, the method for collaborative denoising of multimodal pulsed eddy current signals.

Images