Ultrasonic intelligent detection method and device for diagnosing steel structure defects of railway station canopies

Abstract

The present invention provides an ultrasonic intelligent detection method and device for diagnosing defects in steel structures of railway station canopies. The method comprises: multi-channel ultrasonic scanning and data acquisition; ultrasonic data processing and image generation; steel structure defect feature information extraction; classification of the extracted defect feature information to determine a score for each defect; and comprehensive diagnosis of the degree of defect in the canopy steel structure based on the score value and weight of each defect. The present invention uses multi-channel ultrasound to scan and collect data on the canopy steel structure, then processes the data and generates an image. Through intelligent recognition of the image, defect feature information is extracted, and a statistical machine learning method is used to classify and predict the output score value of each defect. The total defect score is further calculated based on the weight relationship of each defect, and then the degree of defect in the canopy steel structure is diagnosed based on the total defect score. The diagnostic result is accurate, the diagnostic method is advanced, and accurate and effective evaluation of the service safety of the canopy steel structure can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic non-destructive testing, and in particular to an ultrasonic intelligent detection method and device for diagnosing steel structure defects of a railway station canopy. Background Art

[0002] In recent years, the use of steel structure canopies in railway stations has become more and more widespread. Compared with reinforced concrete structure canopies, steel structure canopies have many advantages such as high strength, light weight, flexible structure, large coverage area, strong seismic performance, and reusability. In addition, the quality of steel structure canopies is easy to control during the construction process. Canopies are mostly prefabricated in factories, with a high degree of on-site assembly, convenient maintenance and installation and disassembly, reducing high-altitude operations and shortening the construction period.

[0003] However, in actual use, steel canopies inevitably develop defects, posing a potential threat to railway transportation safety. The structure of a railway station steel canopy primarily consists of a top steel truss, top steel beams, platform steel columns, and external components. The top steel truss and top steel beams are welded on-site at high altitude. Due to construction issues or stress corrosion from prolonged exposure to the open air, weld cracking and brittle failure can occur at these welds. These weld cracks allow rainwater to flow into the platform during the rainy season, flowing along the top truss and beams into the platform steel columns. Since the bottom of the platform steel columns is sealed from the platform floor, the accumulated water is difficult to drain, accelerating corrosion of the inner walls of the columns and causing defects such as rust and thinning. If water accumulation within the platform steel columns is not detected promptly, it could lead to further frost heave in winter, compromising the bearing capacity of the columns or causing safety hazards such as column top support instability. In addition, in order to improve the compressive resistance of the platform steel pipe columns, concrete needs to be poured into some platform steel pipe columns. However, due to construction problems at the source or long-term immersion in accumulated water, the concrete may detach from the inner wall of the platform steel pipe column and cause holes inside the concrete, among other defects.

[0004] Therefore, the steel structure defects of railway station canopies mainly include: wall thickness reduction, inner wall rust, internal water accumulation, weld cracks and concrete defects. These defects seriously threaten the safety and durability of railway station canopies.

[0005] In 2022, Tian Tian et al. published a paper titled "Research on Ultrasonic Detection Methods for Water Accumulation in Steel Columns of Railway Station Canopies" in the Journal of Construction Machinery Technology and Management. This paper analyzes the proportion of ultrasonic reflection energy when water accumulates within a steel column, detecting the presence and depth of water accumulation within the column. However, this method can only detect water accumulation on the inner wall of the column and cannot diagnose any defects in the steel columns of a canopy. In 2023, Hebei Tieda Technology Co., Ltd. applied for an invention patent titled "Method, Terminal, and Storage Medium for Evaluating the Health Status of Steel Columns." This patent conducts a comprehensive, multi-dimensional evaluation of the health of steel columns based on the values ​​of multiple indicators affecting their health, and then determines the health status of the columns based on the health status. However, the method and apparatus disclosed in this patent do not involve image generation and intelligent recognition of ultrasonic signals. Therefore, the evaluation index values ​​obtained are significantly affected by external interference, which affects the accuracy of the final steel column health assessment. Furthermore, this patent does not consider the important indicator of concrete defects within the column, which limits its evaluation.

[0006] It can be seen that the existing technology can only perform single or non-intelligent identification detection and evaluation of the defects of the steel structure of the railway station canopy. There are no methods and devices for comprehensive and integrated intelligent detection and diagnosis of these defects, which makes the railway station canopy a safety hazard. Summary of the Invention

[0007] The embodiments of the present invention provide an ultrasonic intelligent detection method and device for diagnosing defects in steel structures of railway station canopies, so as to solve the problem that the existing evaluation methods for steel structures of railway station canopies are single and incomplete.

[0008] In a first aspect, an embodiment of the present invention provides an ultrasonic intelligent detection method for diagnosing steel structure defects of a railway station canopy, comprising:

[0009] Multi-channel ultrasound scanning data acquisition;

[0010] Ultrasound data processing and image generation;

[0011] Extraction of steel structure defect feature information;

[0012] Classify the extracted disease characteristic information and determine the score value of each disease;

[0013] Based on the score value and weight of each defect, the degree of defect of the canopy steel structure is comprehensively diagnosed.

[0014] Optional, multi-channel ultrasound includes:

[0015] The ultrasonic longitudinal wave straight probe, shear wave oblique probe and guided wave probe excite three ultrasonic waves, namely longitudinal wave, shear wave and guided wave, in the steel structure of the canopy;

[0016] Among them, ultrasonic longitudinal waves and shear waves are high-frequency ultrasonic signals with frequencies between 2MHz and 10MHz;

[0017] Ultrasonic guided waves are low-frequency ultrasonic signals with a frequency between 20kHz and 500kHz.

[0018] Optional, multi-channel ultrasound scan data acquisition, including:

[0019] The ultrasonic transmitter module generates a negative pulse analog voltage signal of a certain frequency;

[0020] The voltage signal is amplified and input into the corresponding ultrasonic probe, generating ultrasonic waves by the inverse piezoelectric effect;

[0021] Ultrasonic waves enter the steel structure of the canopy through the coupling medium;

[0022] Ultrasonic waves are affected by the interfaces and damage of various components in the steel structure of the canopy, and reflection, refraction, transmission and diffraction occur, resulting in continuous energy loss, increasing high-frequency components and changes in acoustic parameters.

[0023] The ultrasonic echo is received by the probe and converted into an analog voltage signal through the piezoelectric effect;

[0024] The preamplifier is used to reduce noise and amplify useful signals to improve the signal-to-noise ratio of the voltage signal;

[0025] The ultrasonic receiving module converts the improved voltage signal into digital signal through an A / D converter (ADC) to form an ultrasonic echo digital signal;

[0026] The digital oscilloscope module displays the digital signal on the display screen;

[0027] The storage module saves the digital signal for the next step of data processing and image generation.

[0028] Optionally, multi-channel ultrasonic scanning data acquisition includes: moving the ultrasonic probe to continuously acquire multi-channel ultrasonic echo signals at different detection positions.

[0029] Optionally, ultrasound data processing includes:

[0030] Perform interception, windowing, filtering and decomposition on the stored ultrasonic echo digital signal data;

[0031] The data interception method may be but is not limited to using one or more gates to intercept and retain signals that meet the requirements of the starting and ending positions and wave heights;

[0032] Data windowing methods can include, but are not limited to, rectangular, Hanning, Hamming, flat-top, Kaiser, and Blackman windows.

[0033] The data filtering method may include but is not limited to Kalman filtering, Wiener filtering and adaptive filtering;

[0034] The data decomposition method may include but is not limited to empirical mode decomposition (EMD), local mean decomposition (LMD) and empirical wavelet decomposition (EWT).

[0035] Optionally, the ultrasound data image generation includes:

[0036] A-scan waveform, B-scan grayscale, amplitude spectrum, high-order spectrum and time-frequency analysis;

[0037] The A-scan waveform is a time-domain signal diagram of a single-point detection by an ultrasonic probe, and is a one-dimensional scan diagram.

[0038] The B-scan grayscale image is a two-dimensional grayscale image formed by converting multiple A-scan waveforms detected by the ultrasonic probe into grayscale lines when it moves on the surface of the canopy steel structure.

[0039] The horizontal axis of the amplitude spectrum is frequency, and the vertical axis is amplitude. It is obtained by performing discrete Fourier transform (DFT) on the A-scan waveform.

[0040] The high-order spectrum is a frequency contour diagram obtained by applying high-order spectrum analysis to the A-scan waveform, thereby extracting the average phase information of the signal;

[0041] The time-frequency analysis diagram describes the relationship between the change of signal frequency and time. The horizontal axis is time, the vertical axis is frequency, and the color depth represents the amplitude.

[0042] Optionally, steel structure defects include: wall thickness reduction, inner wall corrosion, internal water accumulation, weld cracks and concrete defects.

[0043] Optional extraction of steel structure defect feature information, including:

[0044] For the one-dimensional A-scan waveform, traditional signal processing methods are used to effectively extract disease features. The extraction methods can be but are not limited to principal component analysis (PCA), genetic algorithm (GA) and split spectrum analysis (SSP).

[0045] For the two-dimensional B-scan grayscale image, a new two-dimensional statistical method is used to extract disease features. The extraction method can be but is not limited to local binary pattern (LBP), histogram of oriented gradients (HOG), high-order local autocorrelation (HLAC) and gradient local autocorrelation (GLAC).

[0046] Optionally, the extracted disease characteristic information is classified, including:

[0047] Use statistical machine learning methods to classify the extracted disease feature information;

[0048] Statistical machine learning methods can include but are not limited to singular value decomposition (SVD), support vector machine (SVM), sparse coding (SC), artificial neural network (ANN), random forest (RF), probabilistic neural network (PNN), Fisher discriminant analysis (FDA) and recurrent neural network (RNN);

[0049] The sample data sets used in statistical machine learning methods can be derived from data calibrated based on comparison test blocks specified in relevant testing standards, or from sufficiently representative data generated by computer simulation of limited models and disease characteristics, or from new data obtained by performing operations such as mirroring, rotation, scaling, translation, and cropping on existing data sets.

[0050] The classification of disease characteristic information is into five categories: exceeding the scrap line, between the scrap line and the quantitative line, between the quantitative line and the assessment line, between the assessment line and the background noise line, and within the background noise line.

[0051] Optionally, determine a score for each disease, including:

[0052] Based on the results of the statistical machine learning analysis, the severity of each disease is scored;

[0053] The score range is between 0 and 100 points. 100 points are given for defects exceeding the scrap line, 80 to 90 points are given for defects between the scrap line and the quantitative line, 70 to 80 points are given for defects between the quantitative line and the assessment line, 60 to 70 points are given for defects between the assessment line and the background noise line, and 0 to 60 points are given for defects within the background noise line.

[0054] The higher the score, the more serious the damage to the canopy steel structure.

[0055] Optionally, based on the score and weight of each defect, a comprehensive diagnosis of the degree of damage to the canopy steel structure is performed, including:

[0056] The weight of each disease can be given according to relevant technical requirements and assessment standards, or management personnel can propose scientific and reasonable weight value allocation based on the specific location and degree of disease occurrence;

[0057] The total disease score of the canopy steel structure is obtained by comprehensive calculation based on the score value and weight of each disease, and the total score range is between 0 and 100 points;

[0058] Diagnose the degree of damage to the canopy steel structure based on the total damage score;

[0059] Among them, the disease degrees include: Level I disease (above 90 points), Level II disease (between 80 and 90 points), Level III disease (between 70 and 80 points), Level IV disease (between 60 and 70 points) and Level V disease (less than 60 points).

[0060] In a second aspect, an embodiment of the present invention provides an ultrasonic intelligent detection device for diagnosing steel structure defects of a railway station canopy, comprising a device host, an ultrasonic probe, a scanning device, a coupling agent, and an intelligent detection software program. When the device host executes the computer program, the steps of the ultrasonic intelligent detection method for diagnosing steel structure defects of a canopy provided in the first aspect or any possible implementation method of the first aspect are implemented.

[0061] An embodiment of the present invention provides an ultrasonic intelligent detection device for diagnosing steel structure defects in railway station canopies, comprising: multi-channel ultrasonic scanning data acquisition; ultrasonic data processing and image generation; steel structure defect feature information extraction; classification of the extracted defect feature information to determine a score for each defect; and comprehensive diagnosis of the degree of damage to the canopy steel structure based on the score and weight of each defect. In this embodiment of the present invention, based on ultrasonic detection data acquisition and processing and defect feature extraction and identification, scores for each canopy steel structure defect are obtained, and a total defect score is calculated taking into account the weight of each defect, thereby achieving a diagnosis of the degree of damage to the canopy steel structure. The diagnostic process is scientific and intelligent, and the diagnostic results are comprehensive and accurate, providing beneficial support for the overall service safety of the canopy steel structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 This is a flowchart of an implementation of an ultrasonic intelligent detection method for diagnosing steel structure defects of a railway station canopy provided by an embodiment of the present invention;

[0064] Figure 2 This is a schematic diagram showing the principle of an ultrasonic intelligent detection method for diagnosing steel structure defects of a railway station canopy provided by an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the structure of a main unit of an ultrasonic intelligent detection device for diagnosing steel structure defects of a railway station canopy provided by an embodiment of the present invention;

[0066] Figure 4 (a) is a schematic diagram of a normal waveform after being intercepted by a gate in the interception method of the present invention;

[0067] Figure 4 (b) is a schematic diagram of the defect waveform after being intercepted by the gate in the interception method of the present invention;

[0068] Figure 4 (c) is a schematic diagram of a severe defect waveform after interception by a gate in the interception method of the present invention;

[0069] Figure 5 is the amplitude response diagram of Kaiser window;

[0070] Figure 6 This is the waveform diagram of initial weld crack detection;

[0071] Figure 7 This is the waveform diagram of weld crack detection after Kaiser window treatment;

[0072] Figure 8 This is a schematic diagram of the detection signal after the gate intercepts;

[0073] Figure 9 A schematic diagram of filtering results of filtering the detection signal;

[0074] Figure 10 This is a schematic diagram of the actual application effect using empirical mode decomposition (EMD) as an example;

[0075] Figure 11 The following is a schematic diagram of a one-dimensional scan during actual inspection. (a) shows the time domain signal when there is no water accumulation inside the steel structure column of the canopy; (b) shows the time domain signal when there is water accumulation inside the steel structure column of the canopy.

[0076] Figure 12 B-scan grayscale image when there is no weld crack;

[0077] Figure 13 Schematic diagram of surface opening crack;

[0078] Figure 14 This is a schematic diagram of the power amplitude spectrum in actual detection;

[0079] Figure 15 Schematic diagram of the high-order spectrum in actual detection, where (a) is a normal high-order spectrum; (b) is a high-order spectrum with three simulated defect holes;

[0080] Figure 16The following is a schematic diagram of the time-frequency analysis in actual detection; (a) is the time-frequency analysis diagram without water accumulation; (b) is the time-frequency analysis diagram with water accumulation;

[0081] Figure 17 A flow chart showing the combination of genetic algorithm and BP neural network to improve the accuracy of steel structure defect feature information extraction;

[0082] Figure 18 Schematic diagram of each classification line. DETAILED DESCRIPTION

[0083] In the following description, specific details such as particular system structures and techniques are provided for purposes of illustration, not limitation, to facilitate a thorough understanding of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.

[0084] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below with reference to the accompanying drawings.

[0085] See also Figure 1 , which shows a flowchart of an implementation of an ultrasonic intelligent detection method for diagnosing steel structure defects of a railway station canopy provided by an embodiment of the present invention; details are as follows:

[0086] S101: Multi-channel ultrasound scanning data acquisition;

[0087] S102: Ultrasound data processing and image generation;

[0088] S103: Extraction of steel structure defect feature information;

[0089] S104: Classify the extracted disease characteristic information and determine the score value of each disease;

[0090] S105: Based on the score value and weight of each disease, comprehensively diagnose the degree of disease of the canopy steel structure.

[0091] In the embodiment of the present invention, multi-channel ultrasound is used to scan and collect data on the canopy steel structure, and then the data is processed to generate an image. The defect feature information is extracted through intelligent recognition of the image, and the statistical machine learning method is used to classify, predict and output the score value of each defect. The total defect score is further calculated in combination with the weight relationship of each defect. Then, the degree of defect of the canopy steel structure is diagnosed according to the total defect score. The diagnostic result is accurate and the diagnostic method is advanced, which can realize accurate and effective evaluation of the service safety of the canopy steel structure, provide protection for the safe use of the railway station canopy steel structure, and improve the safety of the platform.

[0092] An ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies comprises the following steps:

[0093] S101 multi-channel ultrasound scanning data acquisition:

[0094] S1011: Using three ultrasonic transmitter modules to generate three frequency-adjustable negative pulse analog voltage signals;

[0095] S1012: The three voltage signals are amplified by the power amplifier module and then respectively input into the three ultrasonic probes. The ultrasonic probes generate three ultrasonic signals due to the inverse piezoelectric effect.

[0096] The three ultrasonic probes are ultrasonic longitudinal wave straight probe, shear wave oblique probe and guided wave probe.

[0097] The ultrasonic longitudinal wave straight probe is used to excite ultrasonic longitudinal waves in the steel structure of the canopy, the ultrasonic shear wave oblique probe is used to excite ultrasonic shear waves in the steel structure of the canopy, and the ultrasonic guided wave probe is used to excite ultrasonic guided waves in the steel structure of the canopy.

[0098] Ultrasonic longitudinal and shear waves are high-frequency ultrasonic signals with frequencies between 2MHz and 10MHz, while ultrasonic guided waves are low-frequency ultrasonic signals with frequencies between 20kHz and 500kHz.

[0099] These three ultrasonic waves can simultaneously excite different parts of the canopy steel structure. For example, an ultrasonic longitudinal wave straight probe is placed vertically somewhere at the bottom of the platform steel pipe column, an ultrasonic shear wave oblique probe is placed at the welding point between the top steel pipe truss and the top steel pipe crossbeam, and an ultrasonic guided wave probe is placed in the middle of the platform steel pipe column.

[0100] S1013: Three ultrasonic waves enter the canopy's steel structure through a coupling medium. Because the ultrasonic waves are affected by the interfaces and damage of the various components within the canopy's steel structure, they cause reflection, refraction, transmission, and diffraction. This results in continuous energy loss, an increase in high-frequency components, and changes in acoustic parameters.

[0101] S1014: The three ultrasonic echoes are received by the corresponding probes and converted into analog voltage signals through the piezoelectric effect. The analog voltage signals are input into the preamplifier for noise reduction and amplification of the useful signal, thereby improving the signal-to-noise ratio of the voltage signal.

[0102] S1015: The three-channel ultrasonic receiving module performs analog-to-digital conversion on the improved voltage signal through an A / D converter (ADC) to generate three-channel ultrasonic echo digital signals. The three-channel digital oscilloscope module displays the digital signals on the display module.

[0103] S1016: The storage module saves the three-channel ultrasonic echo digital signals for the next step of data processing and image generation;

[0104] S1017: The three-channel ultrasonic probe moves on the steel structure of the canopy through the scanning device to collect the multi-channel ultrasonic echo signal of the next detection position, and repeats steps S1011 to S1016.

[0105] S102: Ultrasound data processing and image generation:

[0106] S1021: Process the stored ultrasonic echo digital signal, such as interception, windowing, filtering, and decomposition.

[0107] The interception method may be to use one or more gates to intercept and retain the signals that meet the requirements of the starting and ending positions and wave heights;

[0108] For example, when detecting concrete defects, it is advisable to use a gate to intercept the first or second frequency range. This is because when ultrasonic guided waves propagate in damaged concrete, the damage destroys the continuity of the concrete, complicating the propagation process. The superposition of direct waves, diffracted waves, and projected waves passing through the concrete distorts the received waveform, making the waveform identification generally limited to the first and second frequency ranges.

[0109] The normal waveform, defect waveform, and severe defect waveform captured by the gate are as follows: Figure 4 As shown in (a), (b), and (c).

[0110] Windowing methods include rectangular, Hanning, Hamming, flat-top, Kaiser, or Blackman windows.

[0111] Taking the Kaiser window as an example, the expression of its window function is:

[0112] (1)

[0113] Where, is a deformed zero-order Bessel function of the first kind; The main lobe width and side lobe level can be adjusted simultaneously. For example, to obtain a side lobe attenuation of The Kaiser window of dB can be calculated using the following :

[0114] (2)

[0115] The larger the The narrower the window, the smaller the side lobes of the spectrum, but the width of the main lobe also increases accordingly.

[0116] The amplitude response of Kaiser window is shown in the figure below: Figure 5 shown.

[0117] For example, when detecting cracks in the weld of the top steel tube truss, the initial ultrasonic shear wave flaw detection waveform Figure 6 shown.

[0118] After adding Kaiser window to the waveform signal, the waveform is as follows Figure 7 shown.

[0119] As can be seen from the figure above, after filtering with the Kaiser window function, the signal becomes relatively smooth, with a significant denoising effect. The reflection signal and defect signal at the steel pipe end are easy to distinguish, making it easier to perform subsequent image recognition and defect scoring.

[0120] Filtering methods such as Kalman filtering, Wiener filtering or adaptive filtering;

[0121] The surface of the steel structure of the canopy is usually coated with an anti-corrosion layer. In order to save money and time during the inspection process, some anti-corrosion layers are not allowed to be removed. However, due to the presence of the anti-corrosion layer, the reflected echo of the anti-corrosion layer and the first reflected echo signal of the steel structure overlap, resulting in echo aliasing, which affects the inspection results. The example of using recursive least squares adaptive filtering of the signal to improve the detection accuracy of the thinning of the wall thickness of the steel structure of the canopy with anti-corrosion layer is used to illustrate the theoretical part as follows:

[0122] Adaptive filtering is a method that uses the filter parameters obtained at the previous moment to automatically adjust the filter parameters at the current moment to adapt to the unknown or time-varying statistical characteristics of the signal and noise, thereby achieving optimal filtering. Compared with the Wiener filter and the Kalman filter, adaptive filtering does not require prior knowledge of the statistical characteristics of the signal and noise. The input signal u (n) passes through a parameter-adjustable digital filter to generate an output signal y (n), which is compared with the desired signal d (n) to form an error signal e (n). The calculation formula for e (n) is:

[0123] (3)

[0124] The filter parameters are adjusted through an adaptive algorithm to minimize the mean square value of e(n).

[0125] The adaptive filtering recursive rule is to make the weighted cumulative error cost function obtain the minimum value as shown in formula (4):

[0126] (4)

[0127] in Expected signal With output signal The difference, is the forgetting factor, which is a positive number ranging from 0 to 1.

[0128] (5)

[0129] in yes The input vector of the tap at time instant, yes The weight vector of the tap at each moment is equal to the order of the filter In order to minimize the cost function, the derivative of the cost function with respect to the weight vector is calculated and made zero, and the following canonical equation can be obtained:

[0130] (6)

[0131] in is the autocorrelation matrix. If we want to find The best estimate of , then we need to solve The inverse of , the recursive process of the recursive least squares algorithm is:

[0132] Calculate the nth gain vector:

[0133] (7)

[0134] Compute the a priori estimate error:

[0135] (8)

[0136] Calculate the weight vector:

[0137] (9)

[0138] Compute the inverse of the autocorrelation matrix:

[0139] (10)

[0140] As one of the adaptive filter algorithms, the recursive least squares algorithm has a complex recursive process but a fast convergence speed. The value of With error The change is achieved so that the output signal at the next moment is close to the expected signal.

[0141] The actual application steps are as follows:

[0142] Step 1: The ultrasonic detection signal diagram after interception through the gate, such as Figure 8 shown.

[0143] Step 2: Selection of filter order and forgetting factor

[0144] In the least squares recursive adaptive filtering algorithm, the order of the filter is and forgetting factor The error, signal-to-noise ratio, and convergence speed of the output signal are affected, so determining the appropriate parameters is very important for signal processing. The optimal filter order and forgetting factor are selected based on the signal-to-noise ratio of the output signal to obtain the optimal separation signal.

[0145] Step 3: Adaptive filtering of the detection signal

[0146] The filtering results are as follows Figure 9 shown.

[0147] At this time, the filtering results show that the wall thickness reduction value of the canopy steel structure is 0.524mm. This result has high accuracy and is not affected by the thickness of the anti-corrosion layer.

[0148] Data decomposition methods include empirical mode decomposition (EMD), local mean decomposition (LMD) or empirical wavelet decomposition (EWT).

[0149] Take Empirical Mode Decomposition (EMD) as an example:

[0150] The theoretical part is as follows:

[0151] The decomposition process is: find all the maximum points of the original data sequence X(t) and fit them with the cubic spline interpolation function to form the upper envelope of the original data; similarly, find all the minimum points and fit them with the cubic spline interpolation function to form the lower envelope of the data. The mean of the upper and lower envelopes is denoted as ml. The original data sequence X(t) is subtracted from the average envelope ml to obtain a new data sequence hl:

[0152] hl=X(t)-ml (11)

[0153] If the new data after subtracting the envelope average from the original data still has negative local maxima and positive local minima, it means that this is not an eigenmode function and needs to be further "screened". Figure 10 shown.

[0154] The decomposed IMF components contain local characteristic signals of the original ultrasonic signal at different time scales. The empirical mode decomposition method can stabilize non-stationary data.

[0155] S1022: Generate an image of the processed ultrasonic echo digital signal, generating an A-scan waveform image, a B-scan grayscale image, an amplitude spectrum image, a high-order spectrum image, and a time-frequency analysis image.

[0156] The A-scan waveform is a time-domain signal diagram of a single-point detection by an ultrasound probe. It is a one-dimensional scan diagram. Figure 11 This is an example of a one-dimensional scan during an actual inspection. Comparing the two figures, we can see that due to the accumulation of water inside, the ultrasonic echo amplitude decays rapidly as the time series increases.

[0157] In addition, the experiment also found that inner wall corrosion will also cause the ultrasonic echo amplitude to decay rapidly, but the attenuation rate is different. At this time, the wall thickness thinning detection value can also be introduced. If the wall thickness detection value remains unchanged compared to the original wall thickness, it can be determined that it is only caused by internal water accumulation.

[0158] The B-scan grayscale image is a two-dimensional grayscale image formed by converting multiple A-scan waveforms detected by the ultrasonic probe into grayscale lines when it moves on the surface of the canopy steel structure.

[0159] The following figure is an example of a 2D B-scan in actual testing. Figure 12 and Figure 13 shown.

[0160] By comparing the two-dimensional B-scan images with and without cracks, the location, length, size and other information of the cracks can be clearly obtained.

[0161] The horizontal axis of the amplitude spectrum is frequency, and the vertical axis is amplitude. It is obtained by performing discrete Fourier transform (DFT) on the A-scan waveform. Figure 14 This is an example of the power amplitude spectrum in actual detection.

[0162] By comparing the power amplitude spectrum with and without damage, it can be found that damage causes the loss of ultrasonic echo energy, resulting in greater attenuation and an overall lower amplitude than when there is no damage.

[0163] The high-order spectrum is a frequency contour diagram obtained by applying high-order spectrum analysis to the A-scan waveform, thereby extracting the average phase information of the signal; see Figure 15 This is an example of the high-order spectrum in actual detection.

[0164] The results of the high-order spectrum analysis above show that the displayed frequencies of the high-order spectrograms are consistent with the frequencies of the corresponding analyzed signals. Furthermore, the high-order spectrograms of defective steel trusses differ significantly from those of normal steel trusses, particularly in terms of phase information, which is difficult to replicate with other methods.

[0165] The time-frequency analysis diagram describes the relationship between the signal frequency and time. The horizontal axis is time, the vertical axis is frequency, and the color depth represents the amplitude. Figure 16 This is a legend of time-frequency analysis in actual detection.

[0166] A 2.5MHz ultrasonic probe is used in the actual detection, and a time-frequency analysis diagram can be obtained through wavelet time-frequency transform. Compared with the above analysis diagram, it can be seen that when there is no water accumulation, the energy is concentrated on the center frequency of 2.5MHz, and the high-order harmonics are clear. When there is water accumulation, the energy is dispersed in the range of 0~10MHz, and the high-order harmonics are fuzzy. This feature can be used to identify water accumulation inside the steel structure of the canopy.

[0167] From the above analysis, it can be seen that through ultrasonic data processing and image generation, the disease can be presented intuitively.

[0168] However, currently, most methods rely on manual analysis to perform qualitative and quantitative analysis of defect images. However, these methods are susceptible to subjective interference, resulting in missed or misjudgment of defects. Intelligent defect detection systems based on machine learning can achieve highly accurate defect detection while avoiding the shortcomings of manual inspection and significantly improving defect detection speed.

[0169] Therefore, the present invention uses A-scan one-dimensional images to represent ultrasonic detection signals, and also uses two-dimensional images to represent ultrasonic scanning signals. Defect identification is a classification problem, and statistical machine learning methods excel at classifying one-dimensional data and two-dimensional images. Therefore, compared with manual defect identification methods, machine learning has an unparalleled advantage in extracting deep features from defect signals.

[0170] S103: Extraction of steel structure defect feature information:

[0171] The extracted damage information includes five items: wall thickness reduction, inner wall corrosion, internal water accumulation, weld cracks and concrete defects;

[0172] For one-dimensional images such as A-scan waveforms and amplitude spectra, traditional signal processing methods are used to effectively extract disease features, such as principal component analysis (PCA), genetic algorithm (GA), and split spectrum analysis (SSP).

[0173] For two-dimensional images such as B-scan grayscale images, high-order spectrograms and time-frequency analysis images, new two-dimensional statistical methods are used to extract disease features, such as local binary pattern (LBP), histogram of oriented gradients (HOG), high-order local autocorrelation (HLAC) and gradient local autocorrelation (GLAC).

[0174] Take the genetic algorithm (GA) as an example:

[0175] The essence of genetic algorithm is to find the optimal solution through a certain strategy. Genetic algorithm can be used to extract the characteristic information of steel structure defect images.

[0176] The main steps of a genetic algorithm are: selecting an initial population, encoding, constructing a fitness selection function, selection, crossover, and mutation. The first generation of populations produces the next generation through inheritance, i.e., selection, crossover, and mutation. The new population can repeat the above selection, crossover, and mutation inheritance process. The genetic algorithm is applied to image segmentation to find the optimal segmentation threshold. The algorithm design process is as follows:

[0177] Step 1: Generate the initial population. There are generally two methods for generating the initial population. One is completely random generation, which is suitable for situations where there is no prior knowledge of the solution to the problem. Some prior knowledge can be transformed into a set of requirements that must be met. Then, samples are randomly selected from the solutions that meet these requirements, allowing the genetic algorithm to reach the optimal solution. The initialization concept of the present invention is: first, a 1×8 matrix is ​​randomly generated, and then grayscale transformation is performed on it. Grayscale greater than 0.5 is set to 1, and grayscale less than 0 is set to 0.

[0178] Step 2: Encoding. Since the grayscale value of the image is between 0 and 255, the 8-bit binary code in the one-dimensional threshold segmentation needs to be changed to 16 bits, with the first 8 bits representing a threshold and the last 8 bits representing a threshold.

[0179] Step 3: Decoding: The upper 8 bits of the 16-bit binary code are decoded into the previous threshold, and the lower 8 bits are decoded into another threshold.

[0180] Step 4: Fitness selection function. The fitness selection function should reflect the evolutionary excellence of the individual, that is, the degree to which the individual is likely to reach or approach the optimal solution to the problem. In the genetic algorithm, it is the basis for genetic operations on the individual and therefore has a direct impact on the efficiency and performance of the algorithm.

[0181] Step 5: Selection Operator. This invention uses an elite selection operation. In genetic algorithms, elite selection is a very successful strategy for generating new individuals. It directly brings the best individuals into the next generation as elites without any changes.

[0182] Step 6: Crossover operator: The present invention first determines the crossover rate, and then crosses the random positions of the current generation with the current generation.

[0183] Step 7: Mutation Operator. This invention uses binary encoding. Mutation is a bitwise inversion, controlling the rate at which new genes are introduced into the population. If the mutation probability is too low, some useful genes may not be introduced; if the mutation probability is too high, the offspring may lose the beneficial characteristics of both parents. Therefore, this invention first determines the total number of genes, then determines the number of genes to be mutated, and then inverts the bits of the genes to be mutated.

[0184] Step 8: Termination criteria: The present invention uses the maximum evolution algebra control algorithm to terminate, thereby extracting the characteristic information of steel structure defects.

[0185] In addition, genetic algorithms can also be combined with BP neural networks to improve the accuracy of steel structure disease feature information extraction. The process is as follows: Figure 17 shown.

[0186] S104: Classify the extracted disease characteristic information and determine the score value of each disease:

[0187] S1041: Classify the extracted disease feature information using statistical machine learning methods, such as singular value decomposition (SVD), support vector machine (SVM), sparse coding (SC), artificial neural network (ANN), random forest (RF), probabilistic neural network (PNN), Fisher discriminant analysis (FDA) and recurrent neural network (RNN).

[0188] Taking artificial neural network (ANN) as an example, a deep sequential convolutional neural network is used to identify diseases in ultrasound images, such as 7 convolutional layers, 7 batch normalization layers, 7 ReLU (a commonly used activation function) layers, 3 short-circuit modules, 3 short-circuit combined ReLU layers and 1 fully connected layer.

[0189] S1042: The sample data set used in statistical machine learning methods can be derived from data calibrated using comparison test blocks based on relevant testing standards, or from sufficiently representative data generated by computer simulation of limited models and disease characteristics, or from new data generated by operations such as mirroring, rotation, scaling, translation, and cropping on existing data sets.

[0190] S1043: Based on the results of the statistical machine learning analysis, the severity of each disease is scored.

[0191] The classification of disease characteristic information is as follows: exceeding the scrap line, between the scrap line and the quantitative line, between the quantitative line and the assessment line, between the assessment line and the background noise line, and within the background noise line.

[0192] The scoring range is between 0 and 100 points. 100 points are given for defects exceeding the scrap line, 80 to 90 points are given between the scrap line and the quantitative line, 70 to 80 points are given between the quantitative line and the assessment line, 60 to 70 points are given between the assessment line and the background noise line, and 0 to 60 points are given within the background noise line. The higher the score, the more serious the defect of the canopy steel structure.

[0193] Specific classification lines are as follows Figure 18 As shown:

[0194] For example, according to relevant technical requirements, the wall thickness should not be less than 0.85 times the original wall thickness. Therefore, when the wall thickness is reduced to 0.85 times or more of the original wall thickness, the wall thickness reduction defect value is 100 points.

[0195] For example, according to relevant technical requirements, the weld crack is larger than φ 2-6dB side through hole equivalent is considered serious injury, so when the weld crack is greater than or equal to φ 2-6dB SDH equivalent, weld crack damage value is 100 points;

[0196] S105: Based on the score and weight of each disease, comprehensively diagnose the degree of disease of the canopy steel structure:

[0197] S1051: Give the weight value of each disease.

[0198] The weight of each disease can be given according to relevant technical requirements and assessment standards, or the management personnel can propose a scientific and reasonable weight value allocation based on the specific location and degree of the disease;

[0199] S1052: The total defect score of the canopy steel structure is calculated comprehensively through the score value and weight of each defect. The total score ranges from 0 to 100 points.

[0200] The calculation method is: For example, the scores of the five defects, such as wall thickness reduction, inner wall corrosion, internal water accumulation, weld cracks and concrete defects, are: ;

[0201] The weights of these five diseases are: ;

[0202] The total disease score for .

[0203] S1053: Based on the total damage score, diagnose the damage degree of the canopy steel structure.

[0204] The disease severity includes: Level I disease (above 90 points), Level II disease (between 80 and 90 points), Level III disease (between 70 and 80 points), Level IV disease (between 60 and 70 points) and Level V disease (less than 60 points).

[0205] Figure 2 The following is a schematic diagram showing the principle of an ultrasonic intelligent detection method for diagnosing steel structure defects in railway station canopies; the details are as follows:

[0206] The ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies is divided into five steps. The first step is multi-channel ultrasonic scanning data acquisition, which includes ultrasonic transmitting module, power amplification module, ultrasonic probe, ultrasonic receiving module, preamplifier, ultrasonic echo, storage module, display module and scanning device; the second step is ultrasonic data processing and image generation, which includes interception, windowing, filtering, decomposition, A-scan waveform diagram, B-scan grayscale diagram, amplitude spectrum diagram, high-order spectrum diagram and time-frequency analysis diagram; the third step is steel structure defect feature information extraction, which includes traditional signal processing The first step is to classify the extracted defect feature information and determine the score of each defect, which includes statistical machine learning methods and the score of each defect; the second step is to comprehensively diagnose the degree of defect of the canopy steel structure based on the score of each defect and its weight, which includes the weight of each disease, the total disease score and the diagnosis of the disease degree; through the above five steps, a comprehensive and accurate intelligent diagnosis of the service safety of the canopy steel structure is finally achieved.

[0207] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0208] The following are device embodiments of the present invention. For details not fully described therein, reference may be made to the corresponding method embodiments described above.

[0209] Figure 3 The following is a schematic diagram of the structure of a main unit of an ultrasonic intelligent detection device for diagnosing steel structure defects in a railway station canopy. For ease of explanation, only the parts related to the embodiment of the present invention are shown, which are described in detail as follows:

[0210] like Figure 3As shown, the host of the ultrasonic intelligent detection device for diagnosing defects in steel structures of canopies includes: ARM CPU, FPGA digital processing, multi-channel high-speed AD, control circuit, low-frequency transmission, high-frequency transmission, probe receiving end, probe transmitting end, temperature / humidity acquisition, power supply, peripherals, SDRAM, SDROM, display, keyboard, thickness measurement channel, corrosion detection channel, water accumulation detection channel, weld detection channel and concrete detection channel. The ARM CPU is the central processing unit of the device host, which is used to receive keyboard information input and load and run the intelligent detection software program; the ARM CPU outputs the various excitation parameters set by the intelligent detection software program to the control circuit, and the control circuit automatically switches to select the thickness measurement channel, corrosion detection channel, water accumulation detection channel, weld detection channel and concrete detection channel, and automatically triggers the low-frequency transmission or high-frequency transmission module required by the corresponding channel, thereby stimulating the ultrasonic probe at the probe transmitting end to generate ultrasonic waves of the required frequency; the FPGA digital processing is used to receive the ARM The CPU issues instructions and controls the multi-channel high-speed AD to convert the ultrasonic analog signal collected by the probe receiving end into a digital signal according to the instruction requirements; the intelligent detection software program is stored in the SDRAM memory; the collected digital signal is stored in the SDROM memory, and the intelligent detection software program processes the data in the SDROM memory and generates images, and displays the results on the monitor; in addition, the data in the SDROM memory can also be exported to external storage devices such as USB flash drives and TF cards through peripherals; temperature / humidity acquisition is used to detect the surface temperature of the canopy steel structure and the temperature and humidity data of the ambient air. These data are used to extract the characteristic information of steel structure defects and make temperature / humidity compensation to ensure that the quantitative evaluation and scoring of defects are more accurate; the power supply is a detachable rechargeable battery, which can be used for 8 hours after a full charge, providing power for each operating link of the device host during on-site detection.

[0211] For example, the host of the ultrasonic intelligent detection device for diagnosing defects in canopy steel structures is mainly composed of a TFT display, a high-voltage trigger module, a high-voltage synchronous transmission module, a programmable high-gain amplifier, a band-pass filter, a high-speed ADC, a multi-channel transceiver module, 1-channel canopy steel structure temperature acquisition and 1-channel air temperature and humidity acquisition module; the interface consists of BNC, USB, network port, SD card and RS232, etc.; the host structure is made of high-end materials.

[0212] In the embodiment of the present invention, multi-channel ultrasound is used to scan and collect data on the canopy steel structure, and then the data is processed to generate an image. The defect feature information is extracted through intelligent recognition of the image, and the statistical machine learning method is used to classify and predict the output score value of each defect. The total defect score is further calculated in combination with the weight relationship of each defect, and then the degree of the defect of the canopy steel structure is diagnosed according to the total defect score. The diagnostic result is accurate and the diagnostic method is advanced. It solves the problem that the existing method can only perform single or non-intelligent identification detection and evaluation of the defects of the railway station canopy steel structure, and realizes a method and device for comprehensive and integrated intelligent detection and diagnosis of these defects, which greatly alleviates the safety hazards of the railway station canopy.

[0213] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. An ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies, characterized by: multiple Ultrasonic scanning data acquisition; ultrasonic data processing and image generation; extraction of steel structure defect feature information; classification of the extracted defect feature information to determine the score value of each defect; comprehensive diagnosis of the degree of defect of the canopy steel structure based on the score value of each defect and its weight; multi-channel ultrasound includes: ultrasonic longitudinal wave straight probe, shear wave oblique probe and guided wave probe excite three ultrasonic waves of ultrasonic longitudinal wave, shear wave and guided wave in the canopy steel structure; multi-channel ultrasound is simultaneously excited at different parts of the canopy steel structure; multi-channel ultrasonic scanning moves on the canopy steel structure through the scanning device to collect multi-channel ultrasonic echo signals at the next detection position; ultrasonic data processing is the processing of stored ultrasonic echo digital signals, including interception, windowing, filtering and decomposition; ultrasonic image generation includes: generation of A-scan waveform diagram, B-scan grayscale diagram, amplitude spectrum diagram, high-order spectrum diagram and time-frequency analysis diagram; steel structure defect information includes wall thickness thinning, inner wall rust , internal water accumulation, weld cracks and concrete defects; feature information extraction includes the use of traditional signal processing methods to effectively extract defect features for the one-dimensional graphs of A-scan waveform and amplitude spectrum; for the two-dimensional graphs of B-scan grayscale, high-order spectrum and time-frequency analysis, a new two-dimensional statistical method is used to extract defect features; the score value of each defect is determined as follows: the scoring range is between 0 and 100 points, 100 points are given for defects exceeding the scrap line, 80 to 90 points are given for defects between the scrap line and the quantitative line, 70 to 80 points are given for defects between the quantitative line and the assessment line, 60 to 70 points are given for defects between the assessment line and the background noise line, and 0 to 60 points are given for defects within the background noise line; the extracted defect feature information is classified as follows: defects exceeding the scrap line, between the scrap line and the quantitative line, between the quantitative line and the assessment line, between the assessment line and the background noise line, and within the background noise line.

2. The ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies according to claim 1 is characterized in that: The method used for classifying the extracted disease characteristic information is a statistical machine learning method. The sample data set used in the statistical machine learning method comes from data calibrated by comparison test blocks based on relevant testing standards, or from sufficiently representative data generated by centralized computer simulation of limited models and disease characteristics, or from new data obtained by mirroring, rotating, scaling, translating, and cropping existing data sets.

3. The ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies according to claim 1 is characterized in that: The weights of the various diseases are given according to relevant technical requirements and assessment standards, or scientific and reasonable weight value allocation is proposed by management personnel based on the specific location and degree of occurrence of the disease.

4. The ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies according to claim 1 is characterized in that: The comprehensive diagnosis of the degree of damage to the canopy steel structure includes: Level I damage: above 90 points, Level II damage: between 80 and 90 points, Level III damage: between 70 and 80 points, Level IV damage: between 60 and 70 points and Level V damage: less than 60 points.

5. An ultrasonic intelligent detection device for diagnosing steel structure defects of railway station canopies, characterized by: The device comprises a host device, an ultrasonic probe, a scanning device, a coupling agent and an intelligent detection software program. When the host device runs the computer program, the steps of the ultrasonic intelligent detection method for diagnosing steel structure defects of railway station canopies as claimed in claim 1 are executed.

6. The ultrasonic intelligent detection device for diagnosing steel structure defects of a railway station canopy according to claim 5, comprising: ARM CPU, FPGA digital processing, multi-channel high-speed AD, control circuit, low-frequency transmission, high-frequency transmission, probe receiving end, probe transmitting end, temperature / humidity acquisition, power supply, peripherals, SDRAM, SDROM, display, keyboard, thickness measurement channel, corrosion detection channel, water accumulation detection channel, weld detection channel and concrete detection channel; it is characterized in that the ARM CPU is the central processing unit of the device host, which is used to receive keyboard information input and load and run the intelligent detection software program; the ARM CPU outputs the various excitation parameters set by the intelligent detection software program to the control circuit, and the control circuit automatically switches to select the thickness measurement channel, corrosion detection channel, water accumulation detection channel, weld detection channel and concrete detection channel, and automatically triggers the low-frequency transmission or high-frequency transmission module required by the corresponding channel, thereby stimulating the ultrasonic probe at the probe transmitting end to generate ultrasonic waves of the required frequency; the FPGA digital processing is used to receive the ARM The CPU issues instructions and controls the multi-channel high-speed AD to convert the ultrasonic analog signal collected by the probe receiving end into a digital signal according to the instruction requirements; the intelligent detection software program is stored in the SDRAM memory; the collected digital signal is stored in the SDROM memory, and the intelligent detection software program processes the data in the SDROM memory and generates images, and displays the results on the monitor; in addition, the data in the SDROM memory is also exported to the external storage device through the peripheral device; temperature / humidity acquisition is used to detect the surface temperature of the canopy steel structure and the temperature and humidity data of the ambient air. These data are used to extract the characteristic information of steel structure defects and make temperature / humidity compensation to ensure that the quantitative evaluation and scoring of defects are more accurate; the power supply is a detachable rechargeable battery, which can be used for 8 hours after a full charge, providing power for each operating link of the device host during on-site detection.

Images