Praseodymium-neodymium alloy nondestructive testing method and system based on acoustic characteristic analysis

Abstract

The invention relates to the field of alloy defect detection, and particularly discloses a praseodymium-neodymium alloy nondestructive testing method and system based on acoustic characteristic analysis. Defect information contained in an original probe signal is comprehensively captured from two complementary physical perspectives of instantaneous dynamic characteristics and frequency band energy distribution. Furthermore, according to the scheme, simple feature splicing is abandoned, the canonical correlation analysis is used as an information decoupling tool, and two groups of original feature vectors are decomposed into a shared part for describing defect generality and a unique information part for respectively representing the unique resolution capability of the HHT and the wavelet packet on line. And finally, carrying out structured recombination on the three decoupled components to form a fusion feature vector which can effectively eliminate redundancy, amplify differences and has higher information density, and providing input with a clear structure and high refining for a subsequent classification model.

Description

Technical Field

[0001] This application relates to the field of alloy defect detection, and more specifically, to a non-destructive testing method and system for praseodymium-neodymium alloys based on acoustic feature analysis. Background Technology

[0002] Praseodymium-neodymium alloy is the core raw material for preparing high-performance neodymium-iron-boron permanent magnet materials, and its internal quality directly determines the magnetic properties, mechanical properties, and service stability of the final magnet product. During the smelting and casting processes of praseodymium-neodymium alloy production, various internal defects such as cracks and pores are inevitably generated. These microscopic defects cause discontinuities in the internal magnetic domain structure and stress concentration, severely weakening the product's performance and reliability. Therefore, precise and efficient non-destructive testing of praseodymium-neodymium alloy during production, especially accurately identifying and distinguishing different types of defects, is crucial for ensuring product quality, optimizing production processes, and reducing manufacturing costs.

[0003] Acoustic nondestructive testing technology, especially ultrasonic testing, has become the mainstream method for detecting internal defects in metallic materials due to its strong penetration, high sensitivity, and harmlessness to operators. However, applying traditional acoustic testing methods directly to praseodymium-neodymium alloys faces severe challenges. On the one hand, the polycrystalline heterostructure of praseodymium-neodymium alloys causes strong scattering and attenuation of ultrasonic waves, resulting in defect echo signals often being submerged in strong grain noise, leading to an extremely low signal-to-noise ratio. On the other hand, different types of defects (such as cracks and pores) exhibit significant nonlinear and non-stationary characteristics in their acoustic responses. The characteristic differences in their echo signals in the time domain waveform or traditional Fourier spectrum are very weak, often resulting in aliasing. This makes it difficult for traditional detection methods, which rely solely on a single or a few characteristic parameters such as echo amplitude and pulse width, to effectively distinguish defect types, leading to high false positive and false negative rates.

[0004] Therefore, an optimized nondestructive testing scheme for praseodymium-neodymium alloys based on acoustic feature analysis is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a non-destructive testing method and system for praseodymium-neodymium alloys based on acoustic feature analysis.

[0006] According to one aspect of this application, a non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis is provided, comprising: The acquired raw probe signal is subjected to adaptive bandpass filtering to obtain the filtered signal; Parallel feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector; The HHT instantaneous feature vector and the wavelet packet energy feature vector are fused to obtain the fused feature vector; The fused feature vectors are input into a pre-trained SVM model to obtain defect category labels.

[0007] In the above-mentioned non-destructive testing method for praseodymium-neodymium alloy based on acoustic feature analysis, adaptive bandpass filtering is performed on the acquired original probe signal to obtain the filtered signal. This includes: performing defect echo gating processing on the original probe signal to obtain a gated signal; estimating the center frequency of the gated signal based on fast Fourier transform to obtain the peak frequency of the energy spectrum; and performing adaptive bandpass filtering on the gated signal based on the peak frequency of the energy spectrum to obtain the filtered signal, wherein the center frequency of the adaptive bandpass filtering is the peak frequency of the energy spectrum.

[0008] In the above-mentioned non-destructive testing method for praseodymium-neodymium alloy based on acoustic feature analysis, parallel feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector, including: performing Hilbert-Huang transform instantaneous feature extraction on the filtered signal to obtain the HHT instantaneous feature vector; and performing wavelet packet transform energy feature extraction on the filtered signal to obtain the wavelet packet energy feature vector.

[0009] In the above-mentioned non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, feature fusion is performed on the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector. This includes: performing feature preprocessing on the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a shared feature vector, an HHT-specific feature vector, and a wavelet packet-specific feature vector; and performing structured feature assembly on the shared feature vector, the HHT-specific feature vector, and the wavelet packet-specific feature vector to obtain the fused feature vector.

[0010] In the aforementioned nondestructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, feature preprocessing is performed on the HHT instantaneous feature vector and wavelet packet energy feature vector to obtain shared feature vectors, HHT-specific feature vectors, and wavelet packet-specific feature vectors. This includes: obtaining the training set HHT feature matrix and the training set WPT feature matrix; performing canonical correlation analysis on the training set HHT feature matrix and the training set WPT feature matrix to obtain the canonical basis matrix; and based on the canonical basis matrix, performing online feature decoupling of the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain shared feature vectors, HHT-specific feature vectors, and wavelet packet-specific feature vectors.

[0011] In the aforementioned nondestructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, online feature decoupling of the HHT instantaneous feature vector and wavelet packet energy feature vector is performed based on the canonical basis matrix to obtain a shared feature vector, an HHT-specific feature vector, and a wavelet packet-specific feature vector. This includes: projecting the HHT instantaneous feature vector and wavelet packet energy feature vector onto the feature space of the canonical basis matrix to obtain the HHT projected feature vector and the wavelet packet projected feature vector; calculating the mean vector of the HHT projected feature vector and the wavelet packet projected feature vector as the shared feature vector; calculating the residual vector between the HHT instantaneous feature vector and the HHT projected feature vector as the HHT-specific feature vector; and calculating the residual vector between the wavelet packet energy feature vector and the wavelet packet projected feature vector as the wavelet packet-specific feature vector.

[0012] In the above-mentioned non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, the defect category labels include crack labels, porosity labels, and defect-free labels.

[0013] According to another aspect of this application, a non-destructive testing system for praseodymium-neodymium alloys based on acoustic feature analysis is provided, comprising: The bandpass filter module is used to perform adaptive bandpass filtering on the acquired raw probe signal to obtain the filtered signal; The parallel feature extraction module is used to perform parallel feature extraction on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector; The feature fusion module is used to fuse the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector. The defect detection module is used to input the fused feature vector into the pre-trained SVM model to obtain defect category labels.

[0014] Compared with existing technologies, this application provides a non-destructive testing method and system for praseodymium-neodymium alloys based on acoustic feature analysis. It comprehensively captures the defect information contained in the original probe signal from two complementary physical perspectives: instantaneous dynamic characteristics and frequency band energy distribution, by employing Hilbert-Huang transform and wavelet packet transform in parallel. Furthermore, this scheme abandons simple feature splicing and instead utilizes canonical correlation analysis as an information decoupling tool to decompose the two sets of original feature vectors online into a shared part describing the commonalities of defects, and unique information parts representing the unique resolution capabilities of HHT and wavelet packets, respectively. Finally, these three decoupled components are structurally recombined to form a fused feature vector that effectively eliminates redundancy, amplifies differences, and has a higher information density. This provides a clearly structured and highly refined input for subsequent classification models, which are then used for defect category identification, fundamentally improving the identification accuracy and system stability for similar defects such as cracks and porosity. Attached Figure Description

[0015] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 This is a flowchart of a non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of a non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to an embodiment of this application; Figure 3 This is a flowchart illustrating the parallel feature extraction of the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector according to the non-destructive testing method for praseodymium-neodymium alloy based on acoustic feature analysis according to an embodiment of this application. Figure 4 The flowchart illustrates the process of fusing the HHT instantaneous feature vector and wavelet packet energy feature vector to obtain a fused feature vector in the praseodymium-neodymium alloy nondestructive testing method based on acoustic feature analysis according to an embodiment of this application. Figure 5 This is a block diagram of a praseodymium-neodymium alloy nondestructive testing system based on acoustic feature analysis according to an embodiment of this application. Detailed Implementation

[0017] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0018] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0019] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0020] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0021] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0022] To address the technical problems of existing technologies in processing acoustic signals for nondestructive testing of praseodymium-neodymium alloys, such as the overlapping of defect features like cracks and pores, and the introduction of information redundancy and obscuring of key classification criteria when using traditional feature fusion methods, this application proposes a nondestructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis. Specifically, firstly, the acquired raw probe signal is subjected to adaptive bandpass filtering to suppress noise and enhance effective echoes. Then, the filtered signal is subjected to Hilbert-Huang Transform (HHT) and wavelet packet transform in parallel. From the complementary perspectives of instantaneous dynamic characteristics and frequency band energy distribution, the HHT instantaneous feature vector and wavelet packet energy feature vector are extracted, respectively. Furthermore, this scheme abandons the traditional approach of directly splicing features and instead introduces a feature decoupling mechanism based on canonical correlation analysis. This mechanism utilizes a canonical basis matrix learned beforehand on the training set to project and calculate the residuals of two sets of feature vectors acquired online, thereby precisely decomposing them into three parts: a shared feature vector that condenses common information from the two features (such as macroscopic energy of defects), and two unique feature vectors that capture HHT-specific transient information (possibly corresponding to nonlinear response at the crack tip) and wavelet packet-specific frequency band information (possibly corresponding to porosity resonance), respectively. Subsequently, these three decoupled and semantically clear feature components are structurally assembled to form a final fused feature vector with information redundancy eliminated and key differences significantly amplified. Finally, this high-quality fused feature vector is input into a pre-trained support vector machine (SVM) model to obtain accurate defect category labels. In this way, the present invention effectively solves the problems of feature redundancy and confusion, enabling the classifier to focus on distinguishing the essential differences in defects, thereby significantly improving the accuracy and reliability of detection.

[0023] The technical solution of this application proposes a non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis. Figure 1 This is a flowchart of a non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, according to an embodiment of this application. Figure 2This is a schematic diagram of the data flow in a non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, according to an embodiment of this application. Figure 1 and Figure 2 As shown, the non-destructive testing method for praseodymium-neodymium alloy based on acoustic feature analysis according to an embodiment of this application includes the following steps: S100, performing adaptive bandpass filtering on the acquired original probe signal to obtain a filtered signal; S200, performing parallel feature extraction on the filtered signal to obtain an HHT instantaneous feature vector and a wavelet packet energy feature vector; S300, performing feature fusion on the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector; S400, inputting the fused feature vector into a pre-trained SVM model to obtain a defect category label.

[0024] Specifically, in step S100, the acquired raw probe signal is subjected to adaptive bandpass filtering to obtain a filtered signal. It should be understood that during the acoustic testing of praseodymium-neodymium alloys, the acquired raw probe signal not only contains signal components unrelated to defects, such as the initial wave and the bottom wave, but also contains a large amount of high-frequency and low-frequency noise originating from material grain scattering and electronic systems. Simultaneously, the attenuation effect of sound waves propagating within the material causes the actual center frequency of the defect echo to deviate from the probe's nominal frequency. Therefore, in the technical solution of this application, after acquiring the raw probe signal, an adaptive bandpass filter is further applied to obtain a filtered signal. This precisely isolates the defect echo signal, and the filtering parameters are dynamically adjusted according to the echo's spectral characteristics, thereby filtering out noise interference from irrelevant frequency bands to the greatest extent possible. This significantly improves the signal-to-noise ratio of the defect echo signal, laying a high-quality data foundation for subsequent accurate HHT instantaneous feature and wavelet packet energy feature extraction, and avoiding interference from noise components in feature analysis.

[0025] More specifically, in the embodiments of this application, adaptive bandpass filtering is performed on the acquired original probe signal to obtain a filtered signal, including: performing defect echo gating processing on the original probe signal to obtain a gated signal; estimating the center frequency of the gated signal based on fast Fourier transform to obtain the peak frequency of the energy spectrum; and performing adaptive bandpass filtering on the gated signal based on the peak frequency of the energy spectrum to obtain the filtered signal, wherein the center frequency of the adaptive bandpass filtering is the peak frequency of the energy spectrum.

[0026] Accordingly, the original probe signal is subjected to defect echo gating processing to obtain a gated signal. It should be understood that since the acquired original probe signal is a complete time series, it not only contains the defect echo reflecting defect information, but also necessarily the strong initial pulse emitted by the probe (i.e., the initial wave) and the bottom echo formed by the reflection of the sound wave onto the bottom surface of the workpiece (i.e., the bottom wave). Therefore, in the technical solution of this application, the original probe signal is further subjected to defect echo gating processing to obtain a gated signal, thereby precisely focusing the analysis range on the effective signal segment containing defect information and eliminating interference from irrelevant signal components such as the initial wave and the bottom wave. This ensures that all subsequent analysis steps, such as center frequency estimation and feature extraction, are performed on the pure defect echo signal, thus avoiding the misguidance of the defect spectral characteristics by the strong energy initial wave or bottom wave, and guaranteeing the accuracy and effectiveness of subsequent processing.

[0027] Specifically, in one example of this application, the defect echo gating process uses a time window based on the acoustic path principle to capture the effective signal. First, the effective time range for sound wave propagation within the workpiece is calculated based on the known thickness of the praseodymium-neodymium alloy workpiece, the material's sound velocity, and the incident position of the ultrasonic probe. Specifically, the starting time point of the gating is set to avoid the region of the initial wave signal emitted by the probe, and the ending time point is set before the bottom echo arrives. Then, this time gate with defined start and end times is applied to the acquired original probe signal. Finally, only the signal data segments falling within the time gate are extracted and retained, while all signal data outside the gate are discarded, thereby generating a gated signal containing only defect echo information for subsequent spectrum analysis and filtering.

[0028] Accordingly, the center frequency of the gated signal is estimated based on Fast Fourier Transform (FFT) to obtain the peak frequency of the energy spectrum. It should be understood that due to frequency-dependent attenuation of ultrasound propagation within praseodymium-neodymium alloys, high-frequency components attenuate faster than low-frequency components, and different types and depths of defects exhibit varying scattering responses to sound waves. This causes the center frequency of the actually received defect echo signal to deviate from the probe's nominal frequency. Therefore, in the technical solution of this application, the center frequency of the gated signal is further estimated based on FFT to obtain the peak frequency of the energy spectrum, thereby accurately and data-drivenly determining the frequency point where the energy of the specific defect echo signal is most concentrated. This provides an accurate and adaptive center frequency parameter for subsequent bandpass filtering steps, ensuring that the filter accurately matches the spectral characteristics of the current signal, rather than relying on a fixed, potentially inaccurate, preset frequency.

[0029] Specifically, in a concrete example of this application, the center frequency estimation process involves transforming the time-domain gated signal to the frequency domain for analysis. First, a Fast Fourier Transform (FFT) algorithm is applied to the input gated signal, i.e., a time-domain waveform representing a defect echo. This transform converts the signal from the time domain to the frequency domain, generating a complex array where each element corresponds to the amplitude and phase of a specific frequency component. Next, to obtain energy information, the square of the modulus of each element in the complex array is calculated, thus obtaining the energy spectrum of the signal. This energy spectrum visually displays the distribution of signal energy at different frequencies. Finally, a peak search is performed on the energy spectrum, i.e., all frequency points are traversed to find the point with the highest energy value, and the frequency value corresponding to this highest energy point is determined as the peak frequency of the energy spectrum. This frequency value is then used as the center frequency of the adaptive bandpass filter.

[0030] Accordingly, based on the peak frequency of the energy spectrum, an adaptive bandpass filter is applied to the gated signal to obtain the filtered signal, wherein the center frequency of the adaptive bandpass filter is the peak frequency of the energy spectrum. It should be understood that, in addition to containing the main energy components of the defect echo, the gated signal still contains random noise introduced by material grain scattering and electronic systems distributed throughout the entire frequency band. This noise can severely interfere with the accuracy of subsequent feature extraction. Therefore, in the technical solution of this application, the gated signal is further subjected to an adaptive bandpass filter based on the peak frequency of the energy spectrum to obtain the filtered signal, wherein the center frequency of the adaptive bandpass filter is the peak frequency of the energy spectrum, thereby constructing a filter whose passband accurately covers the core energy region of the current defect echo. This allows for the maximum retention of the effective information components of the defect echo while filtering out noise interference outside the frequency band, thereby significantly improving the signal-to-noise ratio and providing clean data input for subsequent high-precision feature extraction.

[0031] Specifically, in a concrete example of this application, the adaptive bandpass filtering process involves dynamically designing and applying a digital filter. First, the peak frequency of the energy spectrum calculated in the previous step is used as the center frequency of the adaptive bandpass filter. Simultaneously, the filter bandwidth is set; for example, the upper and lower cutoff frequencies are set to 0.5 and 1.5 times the center frequency, respectively, to ensure complete coverage of the main energy region of the signal. Next, based on these dynamically determined center frequency and bandwidth parameters, a digital Butterworth bandpass filter of a specific order is designed and generated, and the coefficients of its transfer function are calculated. Finally, the designed filter is applied to a gated signal, and each data point in the gated signal is processed through convolution operations. The output signal sequence after the operation is completed is the filtered signal, whose energy is concentrated near the peak frequency, and out-of-band noise components have been effectively suppressed.

[0032] Specifically, in step S200, parallel feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector. It should be understood that different types of defects (such as cracks and pores) have fundamentally different physical mechanisms when interacting with sound waves, and these differences are reflected in the echo signal in a complex and complementary form. Specifically, the diffraction and nonlinear effects at the crack tip manifest as drastic changes in the instantaneous characteristics of the signal, while the geometry of pores may cause energy resonance in a specific frequency band. Therefore, in the technical solution of this application, parallel feature extraction is further performed on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector, thereby comprehensively and deeply characterizing the defect echo signal from two mutually orthogonal dimensions: instantaneous dynamic characteristics and frequency band energy distribution. This allows the construction of a more discriminative, multi-dimensional feature description than a single feature set, providing a rich and complementary information foundation for accurately distinguishing similar defect signals with different physical causes.

[0033] Figure 3 This is a flowchart illustrating the parallel feature extraction of a filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector, according to an embodiment of the nondestructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis. Figure 3 As shown, step S200 includes: S210, performing Hilbert-Huang transform instantaneous feature extraction on the filtered signal to obtain the HHT instantaneous feature vector; S220, performing wavelet packet transform energy feature extraction on the filtered signal to obtain the wavelet packet energy feature vector.

[0034] Accordingly, in step S210, the Hilbert-Huang transform instantaneous feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector. It should be understood that due to the crack-like defects within praseodymium-neodymium alloys, their sharp edges and complex geometry produce diffraction, scattering, and nonlinear effects when interacting with sound waves, resulting in significant non-stationary and nonlinear characteristics in the echo signal; that is, the frequency and amplitude of the signal change drastically within a very short time. Therefore, in the technical solution of this application, the Hilbert-Huang transform instantaneous feature extraction is further performed on the filtered signal to obtain the HHT instantaneous feature vector, thereby adaptively decomposing the complex defect echo into a series of intrinsic vibration modes and accurately tracking the instantaneous trajectory of energy and frequency changes over time in each mode. This effectively captures and quantifies transient dynamic information closely related to the physical nature of defects (such as the dynamic opening and closing of cracks) that cannot be revealed by traditional Fourier analysis, thereby generating a feature description with high sensitivity and specificity to this type of defect.

[0035] Specifically, in a specific example of this application, the Hilbert-Huang Transform (HHT) instantaneous feature extraction process first processes the input filtered signal using an Empirical Mode Decomposition (EMD) algorithm. This algorithm iteratively decomposes the filtered signal into a finite number of Intrinsic Mode Function (IMF) components, each satisfying a specific time-domain oscillation condition. Next, the top few IMF components with the largest energy contribution are selected, and a Hilbert transform is applied to each selected IMF component, transforming it into an analytic signal. Then, based on this analytic signal, its instantaneous amplitude sequence and instantaneous frequency sequence over time are calculated. Finally, to form a fixed-dimensional feature vector, statistical characteristic parameters, such as mean, standard deviation, variance, kurtosis, and kurtosis, are calculated for the instantaneous amplitude and frequency sequences of each IMF component. All calculated statistical parameters are combined into a one-dimensional vector, which is the HHT instantaneous feature vector.

[0036] Accordingly, in step S220, wavelet packet transform energy feature extraction is performed on the filtered signal to obtain the wavelet packet energy feature vector. It should be understood that due to the volumetric defects such as pores and porosity within praseodymium-neodymium alloys, their specific geometric structure and size distribution will acoustically manifest as resonant absorption or scattering of sound waves in specific frequency bands, resulting in a unique energy distribution pattern in the frequency domain of the defect echo signal. Therefore, in the technical solution of this application, wavelet packet transform energy feature extraction is further performed on the filtered signal to obtain the wavelet packet energy feature vector, thereby performing a fine decomposition of the signal across the entire frequency band and accurately quantifying its energy distribution in each orthogonal sub-band. This allows the construction of a feature fingerprint that reflects the defect's geometric structure or resonant characteristics, providing a key basis for identifying defect types that primarily manifest their existence by changing the signal's spectral energy structure, and forming an effective supplement to the instantaneous characteristics of HHT.

[0037] Specifically, in a specific example of this application, the wavelet packet transform energy feature extraction process first involves parameter setting and signal decomposition. First, a preset wavelet basis function, such as the Debage wavelet, is selected, and a decomposition level is determined, for example, three levels. This level determines the fineness of the frequency band division. Next, the wavelet packet decomposition is performed on the filtered input signal. This process decomposes the entire frequency band of the signal layer by layer, equally, into low-frequency and high-frequency parts, until the preset decomposition depth of three levels is reached, ultimately generating 2^3, or eight, mutually orthogonal sub-frequency bands covering the entire spectrum. Then, for each final sub-frequency band obtained after decomposition, the energy of its wavelet packet coefficients is calculated, i.e., the sum of the squares of all coefficients within that sub-frequency band is calculated. Finally, the energy values ​​calculated for all eight sub-frequency bands are normalized (e.g., their sum is made equal to 1), and these normalized energy values ​​are arranged in ascending order of frequency band to form an 8-dimensional vector, which is the wavelet packet energy feature vector.

[0038] Specifically, in step S300, the HHT instantaneous feature vector and the wavelet packet energy feature vector are fused to obtain a fused feature vector. It should be understood that since the HHT instantaneous feature vector and the wavelet packet energy feature vector originate from the same physical signal, they inevitably share information and collinearity describing the macroscopic characteristics of the defect, while each also contains unique information reflecting different physical mechanisms. If a simple splicing fusion is used, the redundant shared information will dilute and overwhelm those unique features crucial for accurate classification, thereby reducing the performance of the classification model. Therefore, in the technical solution of this application, the HHT instantaneous feature vector and the wavelet packet energy feature vector are further fused to obtain a fused feature vector, thereby actively decoupling the two feature sets into shared information components, HHT unique information components, and wavelet packet unique information components, and then structurally reorganizing these components with clear physical meaning. In this way, a fused feature vector can be constructed with internal redundancy removed, information density increased, and key differential features significantly enhanced, providing a clear and more discriminative input for subsequent classifiers, thereby fundamentally solving the feature confusion problem.

[0039] Figure 4 This is a flowchart illustrating the feature fusion of the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector in the praseodymium-neodymium alloy nondestructive testing method based on acoustic feature analysis according to an embodiment of this application. Figure 4As shown, step S300 includes: S310, performing feature preprocessing on the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a shared feature vector, an HHT-specific feature vector, and a wavelet packet-specific feature vector; S320, performing structured feature assembly on the shared feature vector, the HHT-specific feature vector, and the wavelet packet-specific feature vector to obtain the fused feature vector.

[0040] Specifically, in step S310, the HHT instantaneous feature vector and wavelet packet energy feature vector are preprocessed to obtain shared feature vectors, HHT-specific feature vectors, and wavelet packet-specific feature vectors. It should be understood that when processing acoustic signals for nondestructive testing of praseodymium-neodymium alloys, the Hilbert-Huang transform instantaneous features and wavelet packet transform energy features extracted from the same original echo are essentially descriptions of the same physical process from different mathematical perspectives. Mathematically, they inevitably contain common information describing the macroscopic physical properties of defects (such as overall energy and size), leading to high correlation and information redundancy between features. At the same time, they each contain key unique information characterizing different physical mechanisms. That is, there is inherent information coupling and collinearity between them. Using a simple feature splicing strategy is tantamount to repeatedly introducing redundant information describing the same physical phenomenon (such as the macroscopic size of the defect) into the model. This not only increases the dimensionality of the feature space but also dilutes the more refined and unique information truly used to distinguish the nature of defects. Furthermore, the computational noise of the two transformation methods (such as the endpoint effect of HHT and the low-energy band noise of WPT) is indiscriminately mixed, forming a high-dimensional, high-noise, and structurally ambiguous feature vector. When such a contaminated and redundant vector is fed into a subsequent classifier, it will seriously interfere with its learning process, making it difficult to focus on subtle but crucial acoustic feature differences such as splitting patterns and pores, thus leading to a decline in classification performance and a lack of model stability. Therefore, in the technical solution of this application, the HHT instantaneous feature vector and wavelet packet energy feature vector are further preprocessed to obtain shared feature vectors, HHT-specific feature vectors, and wavelet packet-specific feature vectors. In this way, through a projection and residual calculation mechanism based on canonical correlation analysis, the original, mutually coupled feature space is decomposed into a shared subspace representing commonalities and two mutually orthogonal unique subspaces representing their respective characteristics. This enables effective decoupling of multi-source feature information, laying the foundation for constructing a fusion feature vector with a clear structure, minimized information redundancy, and highlighted key differences, thus providing the classifier with semantically clear and highly refined input.

[0041] More specifically, in this embodiment, feature preprocessing is performed on the HHT instantaneous feature vector and wavelet packet energy feature vector to obtain shared feature vector, HHT-specific feature vector, and wavelet packet-specific feature vector, including: obtaining the training set HHT feature matrix and the training set WPT feature matrix; performing canonical correlation analysis on the training set HHT feature matrix and the training set WPT feature matrix to obtain canonical basis matrix; and based on the canonical basis matrix, performing online feature decoupling of the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain shared feature vector, HHT-specific feature vector, and wavelet packet-specific feature vector.

[0042] In other words, specifically, the first step involves offline learning using a shared-unique information base to obtain the training set HHT feature matrix and training set WPT feature matrix. This offline learning phase of the entire intelligent detection system lays the data foundation for the subsequent establishment of an information decoupling model. Specifically, this step requires ultrasonic testing of a large number of praseodymium-neodymium alloy standard test blocks containing known defect types (such as clearly defined cracks and porosity) and defect-free areas during the offline phase. A complete signal processing procedure is performed on each acquired signal sample until its HHT instantaneous feature vector and wavelet packet energy feature vector are generated. Subsequently, the HHT feature vectors of all samples are stacked row-wise to form the training set HHT feature matrix; similarly, the WPT energy feature vectors of all samples are stacked row-wise to form the training set WPT feature matrix. These two matrices completely record the mathematical expressions of different defects under two feature perspectives in this specific detection scenario.

[0043] Next, canonical correlation analysis is performed on the HHT feature matrix and WPT feature matrix of the training set to obtain the canonical basis matrix. The purpose is to learn and solidify the inherent correlation between the two features from massive historical data. In other words, considering that any online processing would be blind without first establishing a mathematical description of this common information, it is necessary to learn from historical data to find and solidify the strongest correlation axis between the two feature spaces. Since both HHT and WPT features originate from the same physical signal, they must have common information. Canonical correlation analysis, as a statistical tool, is applied to these two feature matrices. Its core task is to solve for a transformation basis that maximizes the correlation after projection of the two sets of features, i.e., the canonical basis matrix. These two bases constitute a mapping bridge from the original feature space to the shared information subspace, expressed by the following formula: , , , in, and Let A and B represent a transformed basis vector or projection vector for the HHT feature space and a transformed basis vector or projection vector for the WPT feature space, respectively. These are column vectors of the transformation matrices A and B, which are also called canonical basis matrices. and These are the HHT feature matrix and the WPT feature matrix of the training set, respectively. Represents the HHT feature matrix of the training set The internal covariance matrix, Represents the WPT feature matrix of the training set The internal covariance matrix, Represents the HHT feature matrix of the training set With WPT characteristic matrix The cross-covariance matrix between them Represents the WPT feature matrix of the training set With HHT feature matrix The cross-covariance matrix between them The first term obtained by solving the characteristic equation is... The nth canonical basis vector, which is the nth vector that constitutes the canonical basis matrix A. column vectors, The first term obtained by solving the characteristic equation is... The nth canonical basis vector, which is the nth vector that constitutes the canonical basis matrix B. column vectors, Indicates the relationship with the first For canonical basis vectors and Corresponding eigenvalues, That is, the first One canonical correlation coefficient, Represents the canonical correlation coefficient, i.e., through vector transformation. and For the characteristic matrix respectively and After performing linear projection, the Pearson correlation coefficient between the two new variables is obtained. Indicates return The maximum value.

[0044] In a physical sense, this canonical basis matrix is ​​equivalent to an information converter in the scenario of praseodymium-neodymium alloy defect detection. It defines a common feature subspace. Once any HHT feature and WPT feature are projected into this space, the common information contained therein will be extracted and aligned to the maximum extent, so as to efficiently extract the common and most stable information components from any set of HHT and WPT features input later.

[0045] Finally, based on the canonical basis matrix, online feature decoupling is performed on the HHT instantaneous feature vector and the wavelet packet energy feature vector. This involves projection and residual calculation. As can be understood, when a new acoustic signal of a test sample is processed, this step utilizes the canonical basis matrix learned in the first step to perform online information decomposition on the newly generated HHT and WPT feature vectors. This is the core step in real-time purification and decomposition of newly acquired unknown defect signal features during actual detection. In other words, the new feature vectors also carry mixed and entangled information. To achieve accurate classification, real-time purification and decomposition must be performed before the classifier intervenes, separating shared information from unique information. When a test signal from a field workpiece is processed to obtain its HHT and WPT feature vectors, the system calls the canonical basis matrix trained offline to decompose these two entangled feature vectors into three mutually orthogonal components with clear physical meaning: a shared feature vector, an HHT-specific feature vector, and a wavelet packet-specific feature vector. This achieves a transformation from chaos to order. The two input feature vectors are effectively decoupled. The shared eigenvectors condense consensus information from both perspectives, such as the approximate energy response of defects. More importantly, the unique eigenvectors of HHT and wavelet packet, as residuals, amplify those weak signals that exist only from a single perspective, such as the nonlinear response at the crack tip captured in HHT, or the specific frequency band resonance of the porosity group reflected in WPT. This information is easily submerged in the original mixed vectors.

[0046] More specifically, in this embodiment, based on the canonical basis matrix, online feature decoupling of the HHT instantaneous feature vector and the wavelet packet energy feature vector is performed to obtain a shared feature vector, an HHT-specific feature vector, and a wavelet packet-specific feature vector. This includes: projecting the HHT instantaneous feature vector and the wavelet packet energy feature vector onto the feature space of the canonical basis matrix to obtain the HHT projected feature vector and the wavelet packet projected feature vector. This step maps the new feature vectors to the common subspace that maximizes commonality through the canonical basis matrix, and the two projected vectors obtained are the mathematical expressions of the common information part in the original features. Subsequently, the mean vector of the HHT projected feature vector and the wavelet packet projected feature vector is calculated as the shared feature vector, aiming to fuse the two projection results to obtain a more stable and robust unified description of common information (such as the macroscopic size or energy of defects). Finally, the residual vector between the HHT instantaneous eigenvector and the HHT projected eigenvector is calculated as the HHT-unique eigenvector, and the residual vector between the wavelet packet energy eigenvector and the wavelet packet projected eigenvector is calculated as the wavelet packet-unique eigenvector. The essence of this residual calculation is to remove known common information from the total information, leaving only the unique characteristic information of each component. For example, the HHT-unique eigenvector will highlight transient details such as the nonlinear effect at the crack tip, while the wavelet packet-unique eigenvector will amplify the resonance effect of pores on specific frequency bands. This purified unique information is crucial for distinguishing similar defects.

[0047] In summary, this structured representation enables downstream classifiers to learn more refined and robust decision rules. For example, it can learn to give higher weight to the HHT-specific feature vector when judging cracks, and pay more attention to the pattern of wavelet packet-specific feature vectors when judging pores, thereby greatly improving the accuracy and interpretability of classification.

[0048] Specifically, in step S320, the shared feature vector, the HHT-specific feature vector, and the wavelet packet-specific feature vector are structurally assembled to obtain the fused feature vector. It should be understood that since the preceding feature preprocessing steps have successfully decomposed the original features into three physically orthogonal and semantically clear independent components: the shared feature vector, the HHT-specific feature vector, and the wavelet packet-specific feature vector, the technical solution of this application further performs structural feature assembly on the shared feature vector, the HHT-specific feature vector, and the wavelet packet-specific feature vector to obtain the fused feature vector. This integrates these purified and decoupled, complementary information components into a unified, high-dimensional feature representation for use by the subsequent classification model. This ensures that the feature vector ultimately input into the classifier retains both the common information characterizing the general attributes of defects and highlights the key specific information distinguishing different defect categories, thus providing the most complete and optimized information foundation for achieving high-precision classification.

[0049] More specifically, in a concrete example of this application, the structured feature assembly process is a deterministic vector concatenation operation. First, the three independent feature vectors generated in the previous step—a shared feature vector, an HHT-specific feature vector, and a wavelet packet-specific feature vector—are acquired and concatenated in a pre-defined fixed order, for example, first the shared feature vector, then the HHT-specific feature vector, and finally the wavelet packet-specific feature vector. The resulting vector is the fused feature vector. This fused feature vector fully preserves all decoupled information and presents it in a structured manner. It is then directly used as input to a pre-trained support vector machine model to perform the final defect classification decision.

[0050] Specifically, in step S400, the fused feature vector is input into a pre-trained SVM model to obtain defect category labels. It should be understood that since the fused feature vector generated in the previous steps is a high-dimensional numerical representation, it does not directly provide defect category information, and the distribution boundaries of different defect categories (such as cracks and porosity) in the feature space are complex and nonlinear. Therefore, in the technical solution of this application, the fused feature vector is further input into a pre-trained SVM model to obtain defect category labels. This utilizes the Support Vector Machine (SVM), a machine learning model with superior performance in handling high-dimensional, nonlinear classification problems, to automatically learn and construct a decision boundary that can optimally distinguish different defect category features. It is worth mentioning that the defect category labels include crack labels, porosity labels, and no-defect labels. In this way, complex numerical feature vectors can be mapped to clear, physically meaningful defect category judgments, thereby achieving automated, intelligent, and highly accurate classification and identification of internal defects in praseodymium-neodymium alloys.

[0051] More specifically, in a concrete example of this application, the classification process relies on a pre-trained classification model. First, in an offline phase, a large number of known types of praseodymium-neodymium alloy defect samples (including cracks, porosity, and defect-free samples) are collected. A complete signal processing and feature fusion process is performed on the signal of each sample to obtain its corresponding fused feature vector, and each vector is labeled with its known true class label. Then, this labeled fused feature vector dataset is used to train a support vector machine (SVM) classifier. This training process selects an appropriate kernel function (e.g., a radial basis function kernel) and optimizes its parameters, ultimately learning a hyperplane that can maximally separate samples of different classes. This trained model is then permanently saved. In the subsequent online detection phase, the fused feature vector generated from the current sample after all the aforementioned steps is directly provided as input to this pre-trained SVM model. The model then applies its internally determined decision function to calculate the input vector, determine its position in the feature space, and finally output a unique classification result, which is the defect class label, specifically one of a crack label, a porosity label, or a defect-free label.

[0052] In summary, the non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to the embodiments of this application is elucidated. It comprehensively captures the defect information contained in the original probe signal from two complementary physical perspectives: instantaneous dynamic characteristics and frequency band energy distribution, by employing Hilbert-Huang transform and wavelet packet transform in parallel. Furthermore, this scheme abandons simple feature splicing and instead utilizes canonical correlation analysis as an information decoupling tool to decompose the two sets of original feature vectors online into a shared part describing the commonalities of defects, and unique information parts representing the unique resolution capabilities of HHT and wavelet packets, respectively. Finally, these three decoupled components are structurally recombined to form a fused feature vector that effectively eliminates redundancy, amplifies differences, and has higher information density. This provides a clearly structured and highly refined input for subsequent classification models, which are then used for defect category identification, thereby fundamentally improving the identification accuracy and system stability for similar defects such as cracks and porosity.

[0053] Furthermore, a non-destructive testing system for praseodymium-neodymium alloys based on acoustic feature analysis is also provided.

[0054] Figure 5 This is a block diagram of a praseodymium-neodymium alloy nondestructive testing system based on acoustic feature analysis according to an embodiment of this application. Figure 5As shown, the praseodymium-neodymium alloy nondestructive testing system 100 based on acoustic feature analysis according to an embodiment of this application includes: a bandpass filtering module 110, used to perform adaptive bandpass filtering on the acquired original probe signal to obtain a filtered signal; a parallel feature extraction module 120, used to perform parallel feature extraction on the filtered signal to obtain an HHT instantaneous feature vector and a wavelet packet energy feature vector; a feature fusion module 130, used to perform feature fusion on the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector; and a defect detection module 140, used to input the fused feature vector into a pre-trained SVM model to obtain a defect category label.

[0055] As described above, the praseodymium-neodymium alloy nondestructive testing system 100 based on acoustic feature analysis according to the embodiments of this application can be implemented in various wireless terminals, such as servers with praseodymium-neodymium alloy nondestructive testing algorithms based on acoustic feature analysis. In one possible implementation, the praseodymium-neodymium alloy nondestructive testing system 100 based on acoustic feature analysis according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the praseodymium-neodymium alloy nondestructive testing system 100 based on acoustic feature analysis can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the praseodymium-neodymium alloy nondestructive testing system 100 based on acoustic feature analysis can also be one of many hardware modules of the wireless terminal.

[0056] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis, characterized in that, include: The acquired raw probe signal is subjected to adaptive bandpass filtering to obtain the filtered signal; Parallel feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector; The HHT instantaneous feature vector and the wavelet packet energy feature vector are fused to obtain the fused feature vector; The fused feature vectors are input into a pre-trained SVM model to obtain defect category labels.

2. The non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to claim 1, characterized in that, The acquired raw probe signal is subjected to adaptive bandpass filtering to obtain the filtered signal, including: Defect echo gating processing is performed on the original probe signal to obtain the gated signal; The center frequency of the gated signal is estimated based on the fast Fourier transform to obtain the peak frequency of the energy spectrum; Based on the peak frequency of the energy spectrum, the gated signal is subjected to adaptive bandpass filtering to obtain the filtered signal, wherein the center frequency of the adaptive bandpass filter is the peak frequency of the energy spectrum.

3. The non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to claim 1, characterized in that, Parallel feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector, including: The Hilbert-Huang transform instantaneous feature extraction is performed on the filtered signal to obtain the HHT instantaneous feature vector; Wavelet packet transform is performed on the filtered signal to extract energy features and obtain the wavelet packet energy feature vector.

4. The non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to claim 3, characterized in that, Feature fusion is performed on the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector, including: Feature preprocessing is performed on the HHT instantaneous feature vector and wavelet packet energy feature vector to obtain shared feature vector, HHT-specific feature vector, and wavelet packet-specific feature vector; The shared feature vector, the HHT-specific feature vector, and the wavelet packet-specific feature vector are structurally assembled to obtain the fused feature vector.

5. The non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to claim 4, characterized in that, Feature preprocessing is performed on the HHT instantaneous feature vector and wavelet packet energy feature vector to obtain shared feature vectors, HHT-specific feature vectors, and wavelet packet-specific feature vectors, including: Obtain the HHT feature matrix and WPT feature matrix of the training set; Canonical correlation analysis was performed on the HHT feature matrix and WPT feature matrix of the training set to obtain the canonical basis matrix. Based on the canonical basis matrix, the instantaneous eigenvectors of HHT and the energy eigenvectors of wavelet packets are decoupled online to obtain shared eigenvectors, HHT-specific eigenvectors, and wavelet packet-specific eigenvectors.

6. The non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to claim 5, characterized in that, Based on the canonical basis matrix, online feature decoupling of the HHT instantaneous eigenvector and wavelet packet energy eigenvector is performed to obtain shared eigenvectors, HHT-specific eigenvectors, and wavelet packet-specific eigenvectors, including: The HHT instantaneous eigenvector and wavelet packet energy eigenvector are projected onto the eigenspace of the canonical basis matrix to obtain the HHT projected eigenvector and wavelet packet projected eigenvector. The mean vector of the HHT projection feature vector and the wavelet packet projection feature vector is calculated as the shared feature vector; The residual vector between the instantaneous feature vector of HHT and the projected feature vector of HHT is calculated as the unique feature vector of HHT. The residual vector between the wavelet packet energy eigenvector and the wavelet packet projection eigenvector is calculated as the unique eigenvector of the wavelet packet.

7. The non-destructive testing method for praseodymium-neodymium alloys based on acoustic feature analysis according to claim 6, characterized in that, The defect category labels include crack labels, porosity labels, and no-defect labels.

8. A non-destructive testing system for praseodymium-neodymium alloys based on acoustic feature analysis, characterized in that, include: The bandpass filter module is used to perform adaptive bandpass filtering on the acquired raw probe signal to obtain the filtered signal; The parallel feature extraction module is used to perform parallel feature extraction on the filtered signal to obtain the HHT instantaneous feature vector and wavelet packet energy feature vector; The feature fusion module is used to fuse the HHT instantaneous feature vector and the wavelet packet energy feature vector to obtain a fused feature vector. The defect detection module is used to input the fused feature vector into the pre-trained SVM model to obtain defect category labels.

Images