A laser-ultrasound-based method for detecting internal defects in post porcelain insulators

Abstract

This invention discloses a method for detecting internal defects in post porcelain insulators based on laser ultrasound, comprising the following steps: deploying an anti-interference laser ultrasound detection device around the post porcelain insulator to create an anti-interference environment; configuring device parameters and acquiring the original signal; performing four-level preprocessing on the original signal; extracting multi-domain features to form a 6-dimensional feature vector; inputting the vector into a BP neural network for detection, and outputting the defect prediction probability. This invention utilizes an openable shielded cabin combined with a mobile platform to resist electromagnetic and vibration interference, adapting to outdoor environments; automatic scanning with an electric guide rail combined with multiple acquisitions replaces manual labor, improving efficiency; four-level preprocessing purifies the signal, providing high-fidelity data; multi-domain feature vectors combined with a BP neural network accurately distinguish minute defects, avoiding false negatives and missing defects, thus solving the pain points of outdoor detection.

Description

Technical Field

[0001] This invention relates to the field of insulator testing technology, and in particular to a laser-ultrasound-based method for detecting internal defects in post porcelain insulators, which is suitable for on-site, precise quantitative detection of internal defects such as cracks and bubbles in post porcelain insulators of substations. Background Technology

[0002] Post porcelain insulators are key components in power systems, widely used in substation facilities, and perform the dual functions of support and insulation. Their performance stability directly affects the safety of power transmission. During long-term operation, under the combined effects of electrical stress, mechanical stress, thermal stress, and complex environmental factors, post porcelain insulators are prone to internal defects such as cracks. The continued development of these defects may lead to a decline in insulation performance, a reduction in mechanical strength, and even power accidents.

[0003] Traditional contact ultrasonic testing methods require direct contact between the equipment and the surface of the post insulator, which is cumbersome and prone to damaging the insulator. Especially in high-altitude or special installation scenarios, complex testing platforms are needed, resulting in high costs and low efficiency. While laser ultrasonic testing technology offers advantages such as non-contact and high resolution, in practical applications, external electromagnetic interference and mechanical vibration can severely affect the quality of the detection signal. In unshielded outdoor environments, the signal-to-noise ratio is typically below 20dB, leading to a missed detection rate of over 40% for microcracks ≤1mm. Furthermore, existing devices are mostly fixed, making them unsuitable for mobile testing in complex outdoor environments. More importantly, existing defect identification methods often rely on single time-domain or frequency-domain features, resulting in insufficient accuracy in identifying minute defects. The false positive rate for cracks and bubbles is as high as 25%, and they cannot output defect size parameters, failing to meet the power industry's requirements for quantitative defect assessment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a laser-ultrasound-based method for detecting internal defects in post porcelain insulators, aiming to solve the problems in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting internal defects in post porcelain insulators based on laser ultrasound, comprising the following steps:

[0006] Step S1: Deploy an anti-interference laser ultrasonic testing device around the perimeter of the post porcelain insulator to create an anti-interference environment for testing the post porcelain insulator;

[0007] Step S2: In an anti-interference environment, configure the parameters of the anti-interference laser ultrasonic testing device, and obtain the original signal of the support porcelain insulator based on the configured parameters of the testing device;

[0008] Step S3: Preprocess the original signal, including Butterworth coarse denoising, wavelet threshold denoising, adaptive weight fusion, and detrending processing;

[0009] Step S4: Perform multi-domain feature extraction on the preprocessed original signal to form a 6-dimensional feature vector;

[0010] Step S5: Input the 6-dimensional feature vector into the BP neural network for defect detection and output the predicted probability of internal defects in the insulator.

[0011] Furthermore, the anti-interference laser ultrasonic testing device includes an openable shielded cabin, an electric guide rail, a high-energy pulsed laser, a filter, a laser interferometer, a retractable robotic arm, a mobile platform, and a computer. The bottom of the openable shielded cabin is fixedly connected to the end effector of the retractable robotic arm via a flange. The base of the retractable robotic arm is bolted to the table surface of the mobile platform. The electric guide rail is fixed to the top of the inner wall of the openable shielded cabin via a bracket. The high-energy pulsed laser is slidably connected to the electric guide rail via a slider to ensure rotational freedom.

[0012] Furthermore, the specific process of step S1 is as follows: Start the mobile platform equipped with the AGV autonomous navigation system, and use the lidar to enable the mobile platform to autonomously navigate to the front of the post porcelain insulator to be tested; control the extension of the telescopic robotic arm to adjust the position of the openable shielded cabin so that the center of the openable shielded cabin door is aligned with the central axis of the post porcelain insulator; open the openable shielded cabin door, and the telescopic robotic arm drives the cabin to move forward so that the post porcelain insulator is in the central area of ​​the cabin and maintains a preset distance from the cabin wall; close the cabin door to form an anti-interference environment for testing the post porcelain insulator.

[0013] Furthermore, in step S2, the parameters of the anti-interference laser ultrasonic testing device are configured as follows: output wavelength, pulse width, single pulse energy, laser spot diameter, and repetition frequency; the electric guide rail fixed on the top support of the inner wall of the openable shielded cabin is controlled to generate a circular scanning path, the scanning step size and axial scanning range are set, and the total number of scanning points is determined; after the detection is started, the high-energy pulsed laser emits laser to excite the Lamb wave in the ceramic insulator of the support column, and the acoustic wave signal is collected simultaneously, converting the Lamb wave into an electrical signal time series, i.e., the original signal. Temporarily stored in the computer cache. Indicates time.

[0014] Furthermore, the specific process of step S3 is as follows:

[0015] Butterworth coarse denoising: An 8th-order Butterworth bandpass filter is used to denoise the original signal. Filtering is performed to obtain the filtered signal. ;

[0016] Wavelet thresholding for fine-grained noise reduction: using a db8 ​​wavelet basis for the original signal. Perform wavelet decomposition to obtain approximate coefficients. and detail coefficient , , Select detail factor Calculate the standard deviation of noise According to the original signal Number of sampling points Calculate the adaptive threshold For detail coefficients , , Soft thresholding is applied, with the following rule: when the detail coefficient > At that time, subtract the detail factor. When the detail factor is < At that time, add the detail factor. When the absolute value of the detail coefficient is ≤ When the time is right, set the detail coefficients to 0; then compare the processed detail coefficients with the approximation coefficients. Reconstruction yields the wavelet-denoised signal. ;

[0017] Adaptive weight fusion: calculating signals signal-to-noise ratio With signal signal-to-noise ratio According to the formula , Calculate separately , Weights; based on weight and weight right , The fusion is performed to obtain the fused signal. ;

[0018] Detrending processing: Multinomial fitting and signal fusion are used. The baseline trend is used to obtain the fitted curve. According to the formula From fused signal Subtracting the trend term from the original signal yields the final preprocessed original signal. .

[0019] Furthermore, the specific process of step S4 is as follows:

[0020] Temporal feature extraction: from the preprocessed raw signal Extracting peak amplitude Rise time Pulse width ;

[0021] Frequency domain feature extraction: This involves extracting features from the preprocessed original signal. Perform a Fourier transform to calculate the power spectral density. ;based on Extracting the main frequency and bandwidth ;

[0022] Time-frequency feature extraction: The db8 wavelet basis is used to extract the preprocessed original signal. Perform wavelet transform and calculate wavelet coefficients. ;

[0023] Feature standardization and weight allocation: This involves standardizing the extracted six original features, specifically the peak amplitude. Rise time Pulse width , main frequency ,bandwidth wavelet coefficients Z-score standardization is performed; the correlation between the six original features after standardization and the defect category is calculated using mutual information; weights are assigned to the six original features after standardization based on the mutual information values; the six original features after standardization are multiplied by their corresponding weights in the order of "time domain → frequency domain → time-frequency" and then concatenated to obtain a 6-dimensional feature vector. .

[0024] Furthermore, the BP neural network adopts a three-layer architecture of "6 input layers - 12 hidden layers - 3 output layers". The hidden layers use the ReLU activation function, and the output layer uses the Softmax function. During network training, the cross-entropy loss function is used to calculate the loss value, and an adaptive learning rate optimizer is introduced.

[0025] Furthermore, after the BP neural network outputs the predicted probabilities of three types of defects—no defects, cracks, and bubbles—it determines the outcome according to the following logic:

[0026] If the maximum predicted probability value is greater than or equal to the third preset threshold, it is directly determined to be the corresponding defect type; if the maximum predicted probability value is within the preset range, it is combined with the 6-dimensional feature vector. A second verification is performed; if the maximum predicted probability value is less than the fourth preset threshold, the electric guide rail is controlled to perform a dense scan of the corresponding area. After collecting the signal 5 times, steps S3-S4 are repeated to re-extract the features, which are then input into the BP neural network for classification. Finally, the defect type and size parameters are output.

[0027] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a laser-ultrasound-based method for detecting internal defects in post porcelain insulators.

[0028] A non-volatile computer storage medium storing computer-executable instructions that execute a laser-ultrasound-based method for detecting internal defects in post porcelain insulators.

[0029] Compared with existing technologies, the present invention has the following advantages:

[0030] (1) This invention adopts an openable shielded cabin combined with a mobile platform, which effectively resists electromagnetic and vibration interference and is suitable for complex outdoor environments. The electric guide rail automatic scanning combined with a multiple acquisition strategy replaces manual operation and greatly improves detection efficiency. The four-level signal preprocessing process purifies the signal layer by layer, providing high-fidelity data for subsequent analysis. Multi-domain feature vectors are constructed and combined with a BP neural network to accurately distinguish various minor defects and avoid false negatives. Overall, it forms a technical advantage of strong environmental adaptability, high detection efficiency, and accurate identification, solving the industry pain points of outdoor insulator detection.

[0031] (2) This invention adopts an openable shielded cabin with a permalloy shielding layer on the inner wall and conductive foam sealing at the joints, which can effectively attenuate electromagnetic interference and mechanical vibration, creating a low-noise environment for detection and solving the problem of high missed detection rate of micro-cracks due to low signal-to-noise ratio in outdoor unshielded environments. At the same time, the mobile platform is equipped with an AGV autonomous navigation system, which can adapt to complex outdoor terrain. With the help of a telescopic robotic arm, the relative position of the shielded cabin and the support porcelain insulator is adjusted so that the insulator is in the central area of ​​the cabin, ensuring signal quality and significantly improving the adaptability of the detection device to different field environments.

[0032] (3) Regarding the detection efficiency of this invention, the electric guide rail achieves automatic circular scanning, covering the effective insulation section of the insulator. Combined with the strategy of "multiple consecutive acquisitions and averaging at each scanning point", random errors in single acquisition are avoided. Furthermore, the automated scanning replaces manual operation, significantly improving the efficiency compared to traditional contact detection. Regarding signal quality, a four-level preprocessing process is adopted, which sequentially performs filtering, wavelet decomposition for noise reduction, weighted fusion, and detrending processing to effectively filter out various types of noise and eliminate signal baseline offset, providing a high-fidelity signal for feature extraction.

[0033] (4) This invention constructs a multi-dimensional, multi-domain feature vector covering the time domain, frequency domain, and time-frequency domain. It calculates the correlation between features and defect categories through mutual information and assigns differentiated attention weights to avoid feature redundancy. It is paired with a BP neural network with a specific architecture and uses the corresponding activation function to solve the gradient vanishing problem and realize the defect probability output. An optimizer is introduced during the training process to avoid model overfitting, effectively reducing the risk of defect misjudgment and accurately distinguishing between microcracks and bubbles. Attached Figure Description

[0034] Figure 1 This is a flowchart of the method of the present invention.

[0035] Figure 2 This is a schematic diagram of the anti-interference laser ultrasonic testing device of the present invention.

[0036] Figure 2 The components include: 1. Openable shielded cabin; 2. Electric guide rail; 3. High-energy pulsed laser; 4. Filter; 5. Post porcelain insulator; 6. Laser interferometer; 7. Telescopic robotic arm; 8. Mobile platform; and 9. Computer. Detailed Implementation

[0037] like Figure 1 As shown, the present invention provides a technical solution: a method for detecting internal defects in post porcelain insulators based on laser ultrasound, comprising the following steps:

[0038] Step S1: Deploy an anti-interference laser ultrasonic testing device around the post porcelain insulator 5 to create an anti-interference environment for testing the post porcelain insulator 5;

[0039] Step S2: In an anti-interference environment, configure the parameters of the anti-interference laser ultrasonic testing device, and obtain the original signal of the support porcelain insulator based on the configured parameters of the testing device;

[0040] Step S3: Preprocess the original signal, including Butterworth coarse denoising, wavelet threshold denoising, adaptive weight fusion, and detrending processing;

[0041] Step S4: Perform multi-domain feature extraction on the preprocessed original signal to form a 6-dimensional feature vector;

[0042] Step S5: Input the 6-dimensional feature vector into the BP neural network for defect detection and output the predicted probability of internal defects in the insulator.

[0043] like Figure 2As shown, the anti-interference laser ultrasonic testing device includes an openable shielded cabin 1, an electric guide rail 2, a high-energy pulsed laser 3, a filter 4, a laser interferometer 6, a telescopic robotic arm 7, a mobile platform 8, and a computer 9. The bottom of the openable shielded cabin 1 is fixedly connected to the end effector of the telescopic robotic arm 7 via a flange. The base of the telescopic robotic arm 7 is bolted to the table surface of the mobile platform 8. The electric guide rail 2 is fixed to the top of the inner wall of the openable shielded cabin 1 via a bracket. The high-energy pulsed laser 3 is slidably connected to the electric guide rail 2 via a slider to ensure 360° rotational freedom.

[0044] Among them, the openable shielded cabin 1 is made of high-strength aluminum alloy, with a permalloy shielding layer on the inner wall and conductive foam sealing at the seams of the cabin. It can achieve ≥60dB of electromagnetic interference attenuation and ≥25dB ​​of mechanical vibration attenuation in the 10kHz-1GHz frequency band, providing a low-noise environment for testing.

[0045] Among them, the electric guide rail 2 has a travel range of 0-1.5m and a positioning accuracy of ±0.1mm. It can drive the high-energy pulse laser 3 to complete a 360° circular scan, ensuring that the laser covers the entire surface of the support porcelain insulator 5.

[0046] Among them, the high-energy pulsed laser 3 has an output wavelength of 1064nm, an adjustable pulse width of 10-50ns, and a single pulse energy of 50-200mJ. The parameters can be adjusted according to the characteristics of the post porcelain insulator 5 to excite stable ultrasonic waves.

[0047] Among them, filter 4 is an adaptive weighted Butterworth-wavelet threshold joint filter that filters out low-frequency mechanical vibrations and high-frequency electromagnetic interference.

[0048] Among them, the laser interferometer 6 has a sampling rate of 100MHz and a measurement accuracy of ±0.1nm, and can convert acoustic signals into electrical signals in real time.

[0049] The telescopic robotic arm 7 is a 6-degree-of-freedom structure with a telescopic range of 0-3m and a repeatability of ±0.5mm. It is used to adjust the relative position of the openable shielded cabin 1 and the support porcelain insulator 5.

[0050] Among them, the mobile platform 8 is equipped with an AGV autonomous navigation system, which can adapt to moving in complex outdoor terrain.

[0051] Computer 9 is equipped with a GPU acceleration unit and runs defect identification algorithm software.

[0052] The specific process of step S1 is as follows: Start the mobile platform 8 equipped with the AGV autonomous navigation system, and use the laser radar to make the mobile platform 8 autonomously navigate to a position 3m in front of the post porcelain insulator 5 to be tested; control the extension of the telescopic robotic arm 7 to extend and adjust the position of the openable shielded cabin 1 so that the center of the door of the openable shielded cabin 1 is aligned with the central axis of the post porcelain insulator 5; open the door of the openable shielded cabin 1, and the telescopic robotic arm 7 moves the cabin forward so that the post porcelain insulator 5 is in the central area of ​​the cabin and the distance between it and the cabin wall is 35cm; close the cabin door and lock the sealing latch to form an anti-interference environment for testing the post porcelain insulator 5.

[0053] The parameters of the anti-interference laser ultrasonic testing device are as follows: output wavelength 1064nm, pulse width 20ns, single pulse energy 100mJ, laser spot diameter 0.5mm, and repetition frequency 10Hz. The electric guide rail 2, fixed to the top support of the inner wall of the openable shielded cabin 1, generates a 360° circular scanning path, with a scanning step size of 0.5mm and an axial scanning range of 0-1.2m, determining a total of 7540 scanning points. After starting the detection, the high-energy pulsed laser 3 emits laser light to excite Lamb waves (a type of ultrasonic wave) within the ceramic insulator 5 of the support column. Acquisition of acoustic signals is synchronized, employing a strategy of "continuous acquisition of 3 times per scanning point with a 50ms interval," converting the Lamb waves into an electrical signal time series, i.e., the original signal. Temporarily stored in the computer's cache area 9. Indicates time.

[0054] The specific process of step S3 is as follows:

[0055] 1. Butterworth coarse noise reduction: An 8th-order Butterworth bandpass filter is used, with a passband set to 0.5-10MHz, to reduce noise in the original signal. Filtering is performed to obtain the filtered signal. It filters out low-frequency mechanical vibrations and high-frequency electromagnetic noise outside the frequency band.

[0056] 2. Wavelet thresholding for fine noise reduction: A db8 wavelet basis is used for the original signal. Perform three-level wavelet decomposition to obtain approximate coefficients. and detail coefficient , , Select detail factor Calculate the standard deviation of noise According to the original signal Number of sampling points Calculate the adaptive threshold The adaptive threshold The calculation is based on the noise standard deviation. With the number of sampling points The matching relationship, where N=1000 (corresponding to a sampling duration of 10μs and a sampling frequency of 100MHz); for detail coefficients , , Soft thresholding is applied, with the following rule: when the detail coefficient > At that time, subtract the detail factor. When the detail factor is < At that time, add the detail factor. When the absolute value of the detail coefficient is ≤ When the time is right, set the detail coefficients to 0; then compare the processed detail coefficients with the approximation coefficients. Reconstruction yields the wavelet-denoised signal. .

[0057] 3. Adaptive weight fusion: calculating signal signal-to-noise ratio With signal signal-to-noise ratio According to the formula , Calculate separately , Weights; based on weight and weight right , The fusion is performed to obtain the fused signal. .

[0058] 4. Detrending processing: The fused signal is fitted using a 5th-order polynomial. The baseline trend is used to obtain the fitted curve. According to the formula From fused signal Subtracting the trend term from the original signal yields the final preprocessed original signal. The baseline offset of the preprocessed original signal is controlled within ±0.05V.

[0059] The specific process of step S4 is as follows:

[0060] 1. Temporal feature extraction: from the preprocessed original signal Extracting peak amplitude Rise time Pulse width .

[0061] 2. Frequency domain feature extraction: This involves extracting frequency domain features from the preprocessed original signal. Perform a Fourier transform to calculate the power spectral density. :

[0062] ;

[0063] In the formula, Indicates frequency; Represents the natural constant; Represents the imaginary unit; This represents the integral.

[0064] based on Extracting the main frequency and bandwidth .

[0065] 3. Time-frequency feature extraction: The db8 wavelet basis is used to extract time-frequency features from the preprocessed original signal. Perform wavelet transform and calculate wavelet coefficients. :

[0066] ;

[0067] In the formula, Indicates the scale factor; Indicates the translation factor; This represents the conjugate function of the db8 wavelet basis functions.

[0068] 4. Feature Standardization and Weight Allocation: The six extracted original features, namely peak amplitude... Rise time Pulse width , main frequency ,bandwidth wavelet coefficients Z-score standardization is performed; the correlation between the six original features and the defect category after standardization is calculated using mutual information. The formula for calculating mutual information is:

[0069] ;

[0070] In the formula, This represents the six original features after standardization. Indicates the defect category label; Representing original features With defect category The joint probability distribution of ; Representing original features The marginal probability; Indicates the defect category The marginal probability; It represents mutual information.

[0071] Weights are assigned to the six original features after standardization based on their mutual information values. The six original features after standardization are multiplied by their corresponding weights in the order of "time domain → frequency domain → time-frequency domain" and then concatenated to obtain a 6-dimensional feature vector. :

[0072] ;

[0073] In the formula, , , , , , These represent the peak amplitude after standardization. Rise time Pulse width , main frequency ,bandwidth wavelet coefficients ; , , , , , They represent , , , , , The corresponding weights.

[0074] The BP neural network adopts a three-layer architecture of "6 input layers - 12 hidden layers - 3 output layers". The hidden layers use the ReLU activation function and the output layer uses the Softmax function. During network training, the cross-entropy loss function is used to calculate the loss value, and an adaptive learning rate optimizer is introduced. The initial learning rate is set to 0.01. When the loss function value drops below the first preset threshold for 5 consecutive iterations, the learning rate is automatically halved until it drops to the second preset threshold.

[0075] The cross-entropy loss function is expressed as: , Indicates the first A number of real labels, including "no defects", "cracks", and "bubbles"; Indicates the first The predicted probability of each output.

[0076] The BP neural network outputs the predicted probabilities of three types of defects: no defects, cracks, and bubbles. The determination is then made according to the following logic:

[0077] If the maximum predicted probability value is ≥0.9, it is directly determined to be the corresponding defect type; if the maximum predicted probability value is between 0.7 and 0.9, it is combined with the 6-dimensional feature vector. The key parameters in the process are verified a second time; if the maximum predicted probability value is <0.7, the electric guide rail is controlled to perform a dense scan of the corresponding area (scanning step size 0.2mm). After collecting the signal 5 times, steps S3-S4 are repeated to re-extract the features, and then the data is input into the BP neural network for classification. Finally, the defect type and size parameters are output.

[0078] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program codes, and the processor is used to call the program codes stored in the memory to execute a laser-ultrasound-based method for detecting internal defects in post porcelain insulators.

[0079] A non-volatile computer storage medium storing computer-executable instructions that execute a laser-ultrasound-based method for detecting internal defects in post porcelain insulators.

[0080] To clearly present the differences between this invention and other related literature in terms of technical dimensions, a comparison is made from three key dimensions: detection object, core detection technology, and core detection target, as shown in Table 1.

[0081] Table 1. Differences between this invention and other related literature in terms of technical dimensions

[0082]

[0083] As can be seen from the comparison, this invention focuses on the precise detection of internal microscopic defects in substation post porcelain insulators. It adopts a technical approach of "laser ultrasound + shielding and anti-interference + refined signal processing + intelligent classification". It is significantly different from other patents that focus on fault identification, heat source reconstruction or stress prediction in terms of the material specificity of the detection object, the anti-interference of the technical solution and the accuracy of defect identification. It is more suitable for the in-service detection needs of microscopic internal defects in post porcelain insulators.

[0084] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting internal defects of a laser-ultrasound-based support porcelain insulator, characterized in that, It comprises the following steps: Step S1: arrange the anti-interference laser ultrasonic detection device on the periphery of the post porcelain insulator to form an anti-interference environment for detecting the post porcelain insulator; Step S2: in the anti-interference environment, configure the parameters of the anti-interference laser ultrasonic detection device, and obtain the original signal of the post porcelain insulator based on the detection device with the configured parameters; Step S3: pre-process the original signal, which includes Butterworth coarse denoising, wavelet threshold denoising, adaptive weight fusion and detrending processing; Step S4: multi-domain feature extraction is performed on the pre-processed original signal to form a 6-dimensional feature vector; Step S5: input the 6-dimensional feature vector into the BP neural network for defect detection, and output the prediction probability of the internal defect of the insulator; The anti-interference laser ultrasonic detection device comprises an open-close shielding cabin, an electric guide rail, a high-energy pulse laser, a filter, a laser interferometer, a telescopic mechanical arm, a moving platform and a computer; the bottom of the open-close shielding cabin is fixedly connected with the end effector of the telescopic mechanical arm through a flange, the base of the telescopic mechanical arm is bolted to the table top of the moving platform, the electric guide rail is fixed to the inner wall top of the open-close shielding cabin through a support, and the high-energy pulse laser is slidingly connected with the electric guide rail through a sliding block, so that the rotation degree of freedom is ensured; The specific process of step S4 is: Time domain feature extraction: peak amplitude, rise time, pulse width are extracted from the pre-processed raw signal ;​​​ Frequency domain feature extraction: Fourier transform is performed on the pre-processed original signal to calculate the power spectral density ; based on the extracted dominant frequency and bandwidth ; Time-frequency feature extraction: db8 wavelet basis is used to the pre-processed original signal Wavelet transform is performed, and wavelet coefficients are calculated ; Feature standardization and weight distribution: Z-score standardization is performed on the extracted 6 original features, i.e. peak amplitude , rising edge time , pulse width , dominant frequency , bandwidth , wavelet coefficient ; the correlation between the 6 original features after standardization and the defect category is calculated by mutual information; the 6 original features after standardization are assigned weights according to the mutual information value; the 6 original features after standardization are multiplied by the corresponding weights in the order of "time domain → frequency domain → time-frequency" and spliced to obtain a 6-dimensional feature vector ; The BP neural network adopts a three-layer architecture of "6 input layer-12 hidden layer-3 output layer", the hidden layer adopts a ReLU activation function, and the output layer adopts a Softmax function; in the network training process, a cross-entropy loss function is used to calculate the loss value, and an adaptive learning rate optimizer is introduced.

2. The method for detecting internal defects of a laser-ultrasound-based post porcelain insulator according to claim 1, characterized in that: The specific process of step S1 is: start the moving platform equipped with an AGV autonomous navigation system, use the laser radar to make the moving platform autonomously navigate to the front of the post porcelain insulator to be detected; control the telescopic mechanical arm to extend, adjust the position of the open-close shielding cabin, and align the center of the cabin door with the central axis of the post porcelain insulator; Open the cabin door of the open-close shielding cabin, drive the cabin body forward by the telescopic mechanical arm, so that the post porcelain insulator is in the central area of the cabin and keeps a preset distance from the cabin wall, close the cabin door, and form an anti-interference environment for detecting the post porcelain insulator.

3. The method for detecting internal defects of a laser-ultrasound-based pillar porcelain insulator according to claim 2, characterized in that: The parameters of the anti-interference laser ultrasonic detection device are configured in step S2: output wavelength, pulse width, single pulse energy, laser spot diameter, and repetition frequency; the electric guide rail fixed to the top bracket of the inner wall of the open-close shielding cabin is controlled to generate a ring-shaped scanning path, the scanning step, the axial scanning range, and the total number of scanning points are set; after starting detection, the high-energy pulse laser emits laser to excite the Lamb wave in the pillar porcelain insulator, and the sound wave signal is synchronously collected, and the Lamb wave is converted into an electric signal time sequence, that is, an original signal temporarily stored in a computer cache area, represents time.

4. The method for detecting internal defects of a laser-ultrasound-based post porcelain insulator according to claim 3, characterized in that: The specific process of step S3 is: Butterworth coarse denoising: an 8th order Butterworth band-pass filter is used to filter the original signal to obtain a filtered signal ; Wavelet threshold fine denoising: select db8 wavelet base for original signal Wavelet decomposition is carried out to obtain approximation coefficients And detail coefficients 、 、 Select the detail coefficient Calculate the noise standard deviation , according to the sampling point number of the original signal Calculate the adaptive threshold 、 、 Soft threshold processing is carried out on the detail coefficient , the processing rule is: when the detail coefficient is greater than , the detail coefficient is subtracted ; When the detail coefficient is less than , the detail coefficient is added ; When the absolute value of the detail coefficient is less than or equal to , the detail coefficient is set to 0; The processed detail coefficient is reconstructed with the approximation coefficient , and the wavelet denoising signal is obtained Adaptive weight fusion: compute signal to noise ratio of signal to noise ratio ; compute weights of , respectively by formula , ; Based on the weight of and the weight of , , , , , fuse to get the fusion signal Detrending processing: Multinomial fitting and signal fusion are used. The baseline trend is used to obtain the fitted curve. According to the formula From fused signal Subtracting the trend term from the original signal yields the final preprocessed original signal. .

5. The method for detecting internal defects of a laser-ultrasound-based post porcelain insulator according to claim 4, characterized in that: After the BP neural network outputs the prediction probabilities of the three types of defects of no defect, crack and bubble, the following logic is used for judgment: If the maximum predicted probability value is greater than or equal to a third preset threshold value, the corresponding defect type is directly determined; if the maximum predicted probability value is between the preset range, the 6-dimensional feature vector is combined A secondary verification is performed; if the maximum predicted probability value is less than a fourth preset threshold value, the corresponding region is subjected to an encrypted scan by the electric guide rail, 5 signals are collected, and then the feature is re-extracted from steps S3 to S4, and finally the BP neural network is inputted for classification, and the defect type and size parameters are outputted.

6. An electronic device, comprising: The computer comprises a processor, a memory and a bus, the processor and the memory are connected through the bus, wherein the memory is used to store a group of program codes, and the processor is used to call the program codes stored in the memory to execute the post porcelain insulator internal defect detection method based on laser ultrasonic as claimed in any one of claims 1-5.

7. A non-transitory computer storage medium storing computer-executable instructions, the computer-executable instructions comprising instructions for: The computer executable instructions execute the post porcelain insulator internal defect detection method based on laser ultrasonic as claimed in any one of claims 1-5.

Images