Cable lead seal welding defect classification method and system based on ultrasonic phased array

Abstract

The invention discloses a cable lead seal welding defect classification method and system based on an ultrasonic phased array. The method comprises the steps that echo signals of cable lead seals with various welding defects are collected through the ultrasonic phased array technology; filtering the echo signal by using a self-adaptive minimum mean square filter, and eliminating random noise and Gaussian noise; performing empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal; first-order and second-order feature extraction technologies are applied to extract various signal features of the echo signals after empirical wavelet transform processing; and inputting the extracted multiple signal features into a deep convolutional neural network to obtain a seal welding defect classification result. According to the invention, the cable lead seal welding defect type can be accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of cable seal defect detection and relates to a cable seal welding defect classification method and system based on ultrasonic phased array. Background Art

[0002] Ultrasonic phased array equipment can be used in virtually any research technology where standard ultrasonic flaw detectors are often used. Weld testing and fracture identification are the most relevant technologies, which are used in a variety of industries, including automotive manufacturing, power generation, petroleum, metal bolt hole and pipe products, pipeline installation and repair, structural metals, and mass production. Phased array can also be used in degradation testing systems to accurately measure remaining wall thickness.

[0003] Ultrasonic phased array simplifies the inspection of parts with complex geometries, inspection of parts in restricted areas, inspection of weld metal from a single detector at different angles, and improves the probability of identification while enhancing the signal-to-noise ratio. However, the intensity of the image signal of the defective part of the cable seal is usually distorted by various noises, making the signal collected by the ultrasonic phased array equipment have more noise, resulting in low accuracy in the classification of cable seal welding defects. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a cable seal welding defect classification method and system based on ultrasonic phased array, which utilizes an adaptive minimum mean square filter to filter the signal collected from the ultrasonic phased array device to eliminate random noise and Gaussian noise, and utilizes an improved empirical wavelet transform algorithm to adaptively capture the frequency characteristics of the signal, thereby effectively improving the accuracy of cable seal welding defect classification.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of the present invention provides a cable seal welding defect classification method based on ultrasonic phased array, comprising:

[0007] The echo signals of cable seals with various welding defects are collected using ultrasonic phased array technology;

[0008] Filtering the echo signal using an adaptive minimum mean square filter to eliminate random noise and Gaussian noise;

[0009] Performing empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal;

[0010] The first-order and second-order feature extraction techniques are used to extract various signal features of the echo signal after the empirical wavelet transform.

[0011] The extracted multiple signal features are input into the deep convolutional neural network to obtain the sealing welding defect classification results.

[0012] Preferably, the adaptive least mean square filter obtains the optimal filter weight combination by the steepest descent method, which is as follows:

[0013] The iterative update formula of the steepest descent method is:

[0014] w(k+1)=w(k)+2a(p(k)-R(k)w(k)) (2)

[0015] R(k)=x(k)x T (k) (3)

[0016] P(k)=d(k)x(k) (4)

[0017] Where w(k+1) and w(k) are the weight vectors at the k+1th and kth iterations, respectively; μ is the step size factor; R(k) is the autocorrelation matrix of the input signal x(k) at the kth iteration; P(k) is the cross-correlation vector between the input signal x(k) and the expected output d(k) at the kth iteration; T represents the vector transpose;

[0018] The steepest descent method determines the direction of weight update through the following gradient calculation formula:

[0019]

[0020] Simplify formula (5) to:

[0021]

[0022] Where e(k) is the error signal at the kth iteration;

[0023] Based on the iterative update formula (2)-(4) and the gradient calculation formula (5)-(6), the filter weight adaptive update formula is determined as:

[0024] w(k+1)=w(k)+2μe(k)x(k) (7)

[0025] The filter weights are updated based on formula (7) to obtain the filter weight combination that minimizes the mean square error as the optimal filter weight combination.

[0026] Preferably, the filtered echo signal is subjected to empirical wavelet transform processing to obtain a processed echo signal, and the formula is as follows:

[0027]

[0028] Among them, x(t)' is the echo signal after empirical wavelet transform processing;

[0029] xk (t) represents the component of the filtered echo signal x(t) in the kth frequency band, where K is the total number of frequency bands;

[0030] represents Fourier transform; represents inverse Fourier transform; H k (f) is the transfer function of the kth bandpass filter.

[0031] Preferably, the transfer function H of the kth bandpass filter is k (f) is:

[0032]

[0033] Among them: A k is the gain coefficient of the kth bandpass filter; f is the frequency; n k is the filter order of the kth bandpass filter; f k0 is the center frequency of the kth bandpass filter, f kc is the frequency of the target signal of the kth bandpass filter; Δf k is the bandwidth of the kth bandpass filter.

[0034] Preferably, the multiple signal characteristics include the signal's average value, variance, energy average value, peak amplitude, and peak power.

[0035] A second aspect of the present invention provides a cable seal welding defect classification system based on ultrasonic phased array, the system comprising:

[0036] A signal acquisition module, used to acquire echo signals of cable seals with various welding defects using ultrasonic phased array technology;

[0037] A noise filtering module, used for filtering the echo signal using an adaptive minimum mean square filter to eliminate random noise and Gaussian noise;

[0038] A wavelet transform processing module is used to perform empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal;

[0039] A feature extraction module, used to extract various signal features of the echo signal after the empirical wavelet transform processing by applying the first-order and second-order feature extraction techniques;

[0040] The defect classification module is used to extract multiple signal features and input them into the deep convolutional neural network to obtain the sealing welding defect classification results.

[0041] Preferably, in the noise filtering module, the adaptive least mean square filter obtains the optimal filter weight combination by the steepest descent method, as follows:

[0042] The iterative update formula of the steepest descent method is:

[0043] w(k+1)=w(k)+2μ(p(k)-R(k)w(k)) (2)

[0044] R(k)=x(k)x T (k) (3)

[0045] P(k)=d(k)x(k) (4)

[0046] Where w(k+1) and w(k) are the weight vectors at the k+1th and kth iterations, respectively; μ is the step size factor; R(k) is the autocorrelation matrix of the input signal x(k) at the kth iteration; P(k) is the cross-correlation vector between the input signal x(k) and the expected output d(k) at the kth iteration; T represents the vector transpose;

[0047] The steepest descent method determines the direction of weight update through the following gradient calculation formula:

[0048]

[0049] Simplify formula (5) to:

[0050]

[0051] Where e(k) is the error signal at the kth iteration;

[0052] Based on the iterative update formula (2)-(4) and the gradient calculation formula (5)-(6), the filter weight adaptive update formula is determined as:

[0053] w(k+1)=w(k)+2μe(k)x(k) (7)

[0054] Based on formula (7), the filter weights are updated to obtain the filter weight combination W0 that minimizes the mean square error as the optimal filter weight combination.

[0055] Preferably, in the wavelet transform processing module, the filtered echo signal is subjected to empirical wavelet transform processing to obtain a processed echo signal, and the formula is as follows:

[0056]

[0057] Among them, x(t)' is the echo signal after empirical wavelet transform processing;

[0058] x k (t) represents the component of the filtered echo signal x(t) in the kth frequency band, where K is the total number of frequency bands;

[0059] represents Fourier transform; represents inverse Fourier transform; H k (f) is the frequency domain response of the kth bandpass filter.

[0060] Preferably, the transfer function H of the kth bandpass filter is k (f) is:

[0061]

[0062] Among them: A k is the gain coefficient of the kth bandpass filter; f is the frequency; n k is the filter order of the kth bandpass filter; f k0 is the center frequency of the kth bandpass filter, f kc is the frequency of the target signal of the kth bandpass filter; Δf k is the bandwidth of the kth bandpass filter.

[0063] A third aspect of the present invention provides a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0064] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0065] Compared with the prior art, the beneficial effects of the present invention include at least:

[0066] The present invention utilizes an adaptive minimum mean square filter to filter noise interference of signals collected by an ultrasonic phased array probe; the adaptive minimum mean square filter of the present invention can adaptively adjust its weight according to an input signal and an error signal to optimize filter performance, and can adjust the step size according to specific system requirements to balance convergence speed and stability, has strong flexibility, performs well in nonlinear situations, and is more suitable for real-time systems.

[0067] The present invention performs empirical wavelet transform processing on the filtered echo signal, and then applies first-order and second-order feature extraction techniques to extract multiple features in the processed signal for classification; the empirical wavelet transform algorithm of the present invention not only avoids the problem of mother wavelet selection in traditional wavelet transform, but also can adaptively capture the frequency characteristics of the signal through data-driven frequency band division, and better capture the local characteristics of the signal, which is particularly suitable for complex non-stable signals such as echo signals of ultrasonic phased arrays. And the design of the bandpass filter transfer function of the empirical wavelet transform in the present invention optimizes the frequency response of the bandpass filter, making it more flexible and accurate in signal processing with complex spectral characteristics. By adjusting the order, center frequency and bandwidth of the parameter filter, the selectivity, bandwidth and transition band steepness of the filter can be flexibly controlled, so that the filter is suitable for the analysis and processing of echo signals of ultrasonic phased arrays. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0069] Figure 2 The overall algorithm architecture diagram of the method of the present invention;

[0070] Figure 3 The basic architecture diagram of the deep convolutional neural network of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0072] Embodiment 1 of the present invention provides a cable seal welding defect classification method based on ultrasonic phased array, such as Figure 1-Figure 2 As shown, including:

[0073] S1, using ultrasonic phased array technology to collect echo signals of cable seals with various welding defects;

[0074] Further preferably, cable seals with various welding defects are synthesized first, and signal acquisition and testing are performed by using ultrasonic phased array technology. Noise interference is found in the signals acquired by the ultrasonic phased array device.

[0075] S2, filtering the echo signal using an adaptive minimum mean square filter to eliminate random noise and Gaussian noise;

[0076] Further preferably, all of the above target signals are acquired, and an adaptive minimum mean square filter is proposed to filter the signals collected from the ultrasonic phased array device to eliminate random noise and Gaussian noise mixed in the signals.

[0077] Specifically, the adaptive least mean square filter combines the characteristics of a low-pass filter, a high-pass filter, and a band-pass filter. The adaptive least mean square filter is a mathematical procedure that simplifies the processing of gradient vectors by appropriately updating the optimization problem. Due to its low computational cost, the adaptive least mean square filter and its related methods are commonly used in many adaptive enhancement implementations.

[0078] In the present invention, the optimization process of integrating dynamic enhancement parameters through a sequential controller is related to the presence of multiple control signals. When calculating the reference data d(k), information is extracted to the minimum mean square error. The optimal solution of the filter weights that minimizes the mean square error can be expressed as:

[0079] W0=R -1 P (1)

[0080] where R = E[x(k)x T (k)] and P = E[d(k)x(k)], where d(k) and x(k) are stationary.

[0081] W0 is the optimal weight vector of the filter, which represents the coefficient combination of the filter. When the weight vector W0 takes this value, the mean square error is minimized; E[·] represents the expectation (mathematical mean); d(k) represents the output of the expected signal at the kth iteration; x(k) represents the input signal vector at the kth iteration;

[0082] The approximate value R(k) of R and the approximate value p(k) of the matrix vector p can be used to solve the optimal solution of the above equation using a procedure based on the steepest descent method. The iterative update formula is as follows:

[0083] w(k+1)=w(k)+2μ(p(k)-R(k)w(k)) (2)

[0084] For k = 0, 1, 2, ..., where μ is the step size factor, a possible solution is to directly calculate R and p. For k = 0, 1, 2, ..., the estimated values ​​of the autocorrelation matrix R(k) and the cross-correlation vector P(k) are calculated as follows:

[0085] R(k)p(k)=x(k)x T (k) (3)

[0086] P(k)=d(k)x(k) (4)

[0087] The direction of weight update is determined by gradient calculation. The formula is as follows:

[0088]

[0089] Simplified as follows:

[0090]

[0091] Then the update formula of the adaptive minimum mean square filter is obtained as follows:

[0092] w(k+1)=w(k)+2μe(k)x(k) (7)

[0093] Where R(k) is the autocorrelation matrix of the input signal, which is an M×M matrix (for a filter of length M). It describes the statistical correlation between the components of the input signal.

[0094] P(k) is the cross-correlation vector between the input signal and the expected output, which represents the statistical correlation between the input signal and the target output.

[0095] w(k) is the weight vector at the kth iteration in the steepest descent method.

[0096] When w(k) reaches W0, the mean square error is minimized. Formula (5) and Formula (6) represent the weight update strategy based on this gradient in Formula (7).

[0097] The adaptive least mean square filter algorithm of the present invention can adaptively adjust its weight according to the input signal x(k) and the error signal e(k) to optimize the filter performance. The step size μ can be adjusted according to the specific needs of the system to balance the convergence speed and stability. It has strong flexibility, performs well in nonlinear situations, and is more suitable for real-time systems.

[0098] S3, performing empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal;

[0099] Further preferably, the signal after natural noise and Gaussian noise are filtered out by the adaptive minimum mean square filter is obtained, and the target signal in the time domain is converted into a complex frequency domain signal by applying an empirical wavelet transform algorithm, so as to extract more features.

[0100] Specifically, since the echo signal of the ultrasonic phased array is an unstable signal, the traditional continuous wavelet transform method that uses a fixed mother wavelet and scale to perform signal time-frequency analysis is not applicable. However, the empirical wavelet transform can adaptively capture the frequency characteristics of the signal through data-driven frequency band division, which is particularly suitable for complex unstable signals.

[0101] The present invention uses empirical wavelet transform to transform time domain signals into complex frequency domain signals, which can better analyze and understand the characteristics of signals at different frequencies and times. The following is a specific explanation of the purpose of the transformation:

[0102] First, multi-resolution analysis: Multi-resolution analysis can be performed by converting time domain signals into frequency domain signals. This means that the behavior of the signal can be observed in different frequency ranges, thereby capturing features at different scales.

[0103] Second, feature extraction: In the frequency domain, it is easier to extract important features of the signal. These features may include the energy distribution of the signal, changes in frequency components, etc., which are crucial for subsequent signal processing and analysis.

[0104] Third, noise removal and enhancement: In the frequency domain, it is easier to identify and remove noise while enhancing the signal components of the target signal. This is because noise and signals may have different manifestations in different frequency ranges.

[0105] Fourth, pattern recognition: In frequency domain analysis, it is easier to identify patterns or anomalies in the signal, which is very useful for tasks such as fault diagnosis and signal classification.

[0106] Fifth, improve classification accuracy: Through EWT conversion, the time-frequency representation of the signal can be obtained, which is particularly helpful for classification tasks using machine learning or deep learning algorithms, because such algorithms can learn complex patterns from multi-dimensional data.

[0107] Specifically, based on S2, the ultrasonic echo signal x(t) is subjected to empirical wavelet transform:

[0108] The output after empirical wavelet transform can be expressed as the sum of each frequency band:

[0109]

[0110] Among them, x k (t) represents the component of the signal in the kth frequency band, and K is the total number of frequency bands.

[0111] For each frequency band x k (t), which can be passed through the corresponding bandpass filter h k (f) Extracted from the original signal, specifically:

[0112]

[0113] in represents Fourier transform; represents inverse Fourier transform; H k (f) is the frequency domain response of the kth bandpass filter.

[0114] The transfer function H of the kth bandpass filter k (f) is:

[0115]

[0116] Among them: A k is the gain coefficient of the kth bandpass filter; f is the frequency; n k is the filter order of the kth bandpass filter; f k0 is the center frequency of the kth bandpass filter, f kc is the frequency of the target signal of the kth bandpass filter; Δf k is the bandwidth of the kth bandpass filter.

[0117] The empirical wavelet transform algorithm of the present invention is different from the traditional continuous wavelet transform. The core idea of ​​the empirical wavelet transform is to analyze by adaptively extracting frequency bands from the signal. Therefore, the empirical wavelet transform not only avoids the problem of mother wavelet selection in the traditional wavelet transform, but also can better capture the local characteristics of the signal. And the design of the bandpass filter transfer function of the empirical wavelet transform in the present invention optimizes the frequency response of the bandpass filter, making it more flexible and accurate in the signal processing with complex spectral characteristics. By adjusting the order, center frequency and bandwidth of the parameter filter, the selectivity, bandwidth and transition band steepness of the filter can be flexibly controlled, so that the filter is suitable for the analysis and processing of various complex signals.

[0118] S4, applying first-order and second-order feature extraction techniques to extract various signal features of the echo signal after empirical wavelet transform processing;

[0119] Further preferably, all the above target signals are obtained, and in the complex frequency domain signals, first-order and second-order feature extraction techniques are applied to extract multiple features for further classification.

[0120] The second-order feature extraction technology focuses on the changing characteristics of the signal. Specifically, the first-order feature extraction technology mainly focuses on the basic properties of the target signal. The first-order features usually include statistics such as the amplitude, mean, and standard deviation of the signal. These features directly reflect some basic characteristics of the signal. Compared with the first-order features, the second-order features are more complex and usually involve the rate of change and correlation of the signal. Second-order feature extraction usually involves the derivative and autocorrelation function of the signal. The feature signals that the present invention focuses on for the target signal are mean, variance, energy mean, peak amplitude, low amplitude, and peak power.

[0121] S5, the extracted multiple signal features are input into the deep convolutional neural network to obtain the sealing welding defect classification results.

[0122] Further preferably, all the above signal features are obtained, and the features are further classified using a deep convolutional neural network to determine whether the cable seal welding signal has defects. A confusion matrix can also be used to estimate classification performance, including calculating accuracy, sensitivity and specificity.

[0123] Specifically, in the present invention, a deep convolutional neural network is used to classify the welding defect signals of cable seals. Figure 3 As shown, the network architecture of the deep convolutional neural network of the present invention is:

[0124] Input layer: The input layer has q nodes, which means that the input signal is represented as a q-dimensional vector. q is taken according to the number of features extracted from the original signal by the feature extraction method (such as empirical wavelet transform) in the specific implementation;

[0125] Hidden layer: The first hidden layer contains multiple nodes. These nodes are connected to the input layer through a series of weights and biases to extract higher-level features. The second hidden layer contains multiple nodes. This layer further processes the information from the first hidden layer to extract more complex features.

[0126] Output layer: The output layer contains multiple nodes that are used to identify multiple different signal groups.

[0127] The network training and training objectives of the deep convolutional neural network of the present invention are: the training goal is to make the network error in the classification task reach 0.01 (i.e. 1e-2) or lower.

[0128] Training process: The neural network will be trained multiple times, and each training will adjust the neuron weights of the hidden layer until the desired error target is achieved.

[0129] Embodiment 2 of the present invention provides a cable seal welding defect classification system based on ultrasonic phased array, comprising:

[0130] A signal acquisition module, used to acquire echo signals of cable seals with various welding defects using ultrasonic phased array technology;

[0131] A noise filtering module, used for filtering the echo signal using an adaptive minimum mean square filter to eliminate random noise and Gaussian noise;

[0132] Further preferably, in the noise filtering module, the adaptive least mean square filter obtains the optimal filter weight combination by the steepest descent method, as follows:

[0133] The iterative update formula of the steepest descent method is:

[0134] w(k+1)=w(k)+2μ(p(k)-R(k)w(k)) (2)

[0135] R(k)=x(k)x T (k) (3)

[0136] P(k)=d(k)x(k) (4)

[0137] Where w(k+1) and w(k) are the weight vectors at the k+1th and kth iterations, respectively; μ is the step size factor; R(k) is the autocorrelation matrix of the input signal x(k) at the kth iteration; P(k) is the cross-correlation vector between the input signal x(k) and the expected output d(k) at the kth iteration; T represents the vector transpose;

[0138] The steepest descent method determines the direction of weight update through the following gradient calculation formula:

[0139]

[0140] Simplify formula (5) to:

[0141]

[0142] Where e(k) is the error signal at the kth iteration;

[0143] Based on the iterative update formula (2)-(4) and the gradient calculation formula (5)-(6), the filter weight adaptive update formula is determined as:

[0144] w(k+1)=w(k)+2μe(k)x(k) (7)

[0145] Based on formula (7), the filter weights are updated to obtain the filter weight combination W0 that minimizes the mean square error as the optimal filter weight combination.

[0146] A wavelet transform processing module is used to perform empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal;

[0147] Further preferably, in the wavelet transform processing module, the filtered echo signal is subjected to empirical wavelet transform processing to obtain a processed echo signal, and the formula is as follows:

[0148]

[0149] Among them, x(t)' is the echo signal after empirical wavelet transform processing;

[0150] x k (t) represents the component of the filtered echo signal x(t) in the kth frequency band, where K is the total number of frequency bands;

[0151] represents Fourier transform; represents inverse Fourier transform; H k (f) is the frequency domain response of the kth bandpass filter.

[0152] The transfer function H of the kth bandpass filter k (f) is:

[0153]

[0154] Among them: A k is the gain coefficient of the kth bandpass filter; f is the frequency; n k is the filter order of the kth bandpass filter; f k0 is the center frequency of the kth bandpass filter, f kc is the frequency of the target signal of the kth bandpass filter; Δf k is the bandwidth of the kth bandpass filter.

[0155] A feature extraction module, used to extract various signal features of the echo signal after the empirical wavelet transform processing by applying the first-order and second-order feature extraction techniques;

[0156] The defect classification module is used to extract multiple signal features and input them into the deep convolutional neural network to obtain the sealing welding defect classification results.

[0157] Embodiment 3 of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0158] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0159] Compared with the prior art, the beneficial effects of the present invention include at least:

[0160] The present invention utilizes an adaptive minimum mean square filter to filter noise interference of signals collected by an ultrasonic phased array probe; the adaptive minimum mean square filter of the present invention can adaptively adjust its weight according to an input signal and an error signal to optimize filter performance, and can adjust the step size according to specific system requirements to balance convergence speed and stability, has strong flexibility, performs well in nonlinear situations, and is more suitable for real-time systems.

[0161] The present invention performs empirical wavelet transform processing on the filtered echo signal, and then applies first-order and second-order feature extraction techniques to extract multiple features in the processed signal for classification; the empirical wavelet transform algorithm of the present invention not only avoids the problem of mother wavelet selection in traditional wavelet transform, but also can adaptively capture the frequency characteristics of the signal through data-driven frequency band division, and better capture the local characteristics of the signal, which is particularly suitable for complex non-stable signals such as echo signals of ultrasonic phased arrays. And the design of the bandpass filter transfer function of the empirical wavelet transform in the present invention optimizes the frequency response of the bandpass filter, making it more flexible and accurate in signal processing with complex spectral characteristics. By adjusting the order, center frequency and bandwidth of the parameter filter, the selectivity, bandwidth and transition band steepness of the filter can be flexibly controlled, so that the filter is suitable for the analysis and processing of echo signals of ultrasonic phased arrays.

[0162] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0163] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0164] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0165] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A cable seal welding defect classification method based on ultrasonic phased array, characterized in that: include: The echo signals of cable seals with various welding defects are collected using ultrasonic phased array technology; Filtering the echo signal using an adaptive minimum mean square filter to eliminate random noise and Gaussian noise; Performing empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal; The first-order and second-order feature extraction techniques are used to extract various signal features of the echo signal after the empirical wavelet transform. The extracted multiple signal features are input into the deep convolutional neural network to obtain the sealing welding defect classification results.

2. The method for classifying cable seal welding defects based on ultrasonic phased array according to claim 1 is characterized in that: The adaptive least mean square filter obtains the optimal filter weight combination through the steepest descent method, which is as follows: The iterative update formula of the steepest descent method is: w(k+1)=w(k)+2μ(p(k)-R(k)w(k)) (2) R(k)=x(k)x T (k) (3) P(k)=d(k)x(k) (4) Where w(k+1) and w(k) are the weight vectors at the k+1th and kth iterations, respectively; μ is the step size factor; R(k) is the autocorrelation matrix of the input signal x(k) at the kth iteration; P(k) is the cross-correlation vector between the input signal x(k) and the expected output d(k) at the kth iteration; T represents the vector transpose; The steepest descent method determines the direction of weight update through the following gradient calculation formula: Simplify formula (5) to: Where e(k) is the error signal at the kth iteration; Based on the iterative update formula (2)-(4) and the gradient calculation formula (5)-(6), the filter weight adaptive update formula is determined as: w(k+1)=w(k)+2μe(k)x(k) (7) The filter weights are updated based on formula (7) to obtain the filter weight combination that minimizes the mean square error as the optimal filter weight combination.

3. The cable seal welding defect classification method based on ultrasonic phased array according to claim 1 is characterized by: The filtered echo signal is subjected to empirical wavelet transform processing to obtain a processed echo signal, and the formula is as follows: Among them, x(t)' is the echo signal after empirical wavelet transform processing; x k (t) represents the component of the filtered echo signal x(t) in the kth frequency band, where K is the total number of frequency bands; represents Fourier transform; represents inverse Fourier transform; H k (f) is the transfer function of the kth bandpass filter.

4. The method for classifying cable seal welding defects based on ultrasonic phased array according to claim 3 is characterized in that: The transfer function H of the kth bandpass filter k (f) is: Among them: A k is the gain coefficient of the kth bandpass filter; f is the frequency; n k is the filter order of the kth bandpass filter; f k0 is the center frequency of the kth bandpass filter, f kc is the frequency of the target signal of the kth bandpass filter; Δf k is the bandwidth of the kth bandpass filter.

5. The method for classifying cable seal welding defects based on ultrasonic phased array according to claim 1, characterized in that: The various signal characteristics include the average value, variance, energy average value, peak amplitude, and peak power of the signal.

6. A cable seal welding defect classification system based on ultrasonic phased array, characterized in that: The system comprises: A signal acquisition module, used to acquire echo signals of cable seals with various welding defects using ultrasonic phased array technology; A noise filtering module, used for filtering the echo signal using an adaptive minimum mean square filter to eliminate random noise and Gaussian noise; A wavelet transform processing module is used to perform empirical wavelet transform processing on the filtered echo signal to obtain a processed echo signal; A feature extraction module, used to extract various signal features of the echo signal after the empirical wavelet transform processing by applying the first-order and second-order feature extraction techniques; The defect classification module is used to extract multiple signal features and input them into the deep convolutional neural network to obtain the sealing welding defect classification results.

7. The cable seal welding defect classification system based on ultrasonic phased array according to claim 6 is characterized by: In the noise filtering module, the adaptive least mean square filter obtains the optimal filter weight combination through the steepest descent method, which is as follows: The iterative update formula of the steepest descent method is: w(k+1)=w(k)+2μ(p(k)-R(k)w(k)) (2) R(k)=x(k)x T (k) (3) P(k)=d(k)x(k) (4) Where w(k+1) and w(k) are the weight vectors at the k+1th and kth iterations, respectively; μ is the step size factor; R(k) is the autocorrelation matrix of the input signal x(k) at the kth iteration; P(k) is the cross-correlation vector between the input signal x(k) and the expected output d(k) at the kth iteration; T represents the vector transpose; The steepest descent method determines the direction of weight update through the following gradient calculation formula: Simplify formula (5) to: Where e(k) is the error signal at the kth iteration; Based on the iterative update formula (2)-(4) and the gradient calculation formula (5)-(6), the filter weight adaptive update formula is determined as: w(k+1)=w(k)+2μe(k)x(k) (7) The filter weights are updated based on formula (7) to obtain the filter weight combination that minimizes the mean square error as the optimal filter weight combination.

8. The cable seal welding defect classification system based on ultrasonic phased array according to claim 6 is characterized by: In the wavelet transform processing module, the filtered echo signal is subjected to empirical wavelet transform processing to obtain the processed echo signal. The formula is as follows: Among them, x(t)' is the echo signal after empirical wavelet transform processing; x k (t) represents the component of the filtered echo signal x(t) in the kth frequency band, where K is the total number of frequency bands; represents Fourier transform; represents inverse Fourier transform; H k (f) is the frequency domain response of the kth bandpass filter.

9. The cable seal welding defect classification system based on ultrasonic phased array according to claim 6 is characterized by: The transfer function H of the kth bandpass filter k (f) is: Among them: A k is the gain coefficient of the kth bandpass filter; f is the frequency; n k is the filter order of the kth bandpass filter; f k0 is the center frequency of the kth bandpass filter, f kc is the frequency of the target signal of the kth bandpass filter; Δf k is the bandwidth of the kth bandpass filter.

10. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-5.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images