An internal crack detection method for hydraulic structures

Abstract

The present invention discloses a method for detecting internal cracks of hydraulic structures, belonging to the technical field of crack detection. The present invention obtains electromagnetic echoes by emitting sinusoidal electromagnetic waves; secondly, the electromagnetic echoes are segmented according to the same time length, and the local maximum and minimum values are extracted, and the growth and decay values of the maximum and minimum values are calculated; then, the degree of abnormality is calculated based on these maximum and minimum values, and an abnormal vector is constructed; then, the distribution of the maximum and minimum values is analyzed, the instability is calculated and an instability vector is constructed; finally, the extracted abnormal vector and instability vector are processed by a crack feature fusion neural network, so as to obtain an evaluation value of the internal cracks of the hydraulic structure. By comprehensively analyzing the characteristics of the electromagnetic echo signal, the present invention effectively improves the accuracy and reliability of crack detection.

Description

Technical Field

[0001] The present invention relates to the technical field of crack detection, and particularly to a method for detecting internal cracks of hydraulic structures. Background Art

[0002] Hydraulic structures are crucial infrastructure in water conservancy projects, including dams, sluices, canals, etc. Their structural safety is directly related to the stability of project operation and the ability to prevent and mitigate disasters. Currently, traditional detection methods for hydraulic structures mainly rely on manual visual inspection, which has problems such as low detection accuracy, limited detection depth, and difficulty in identifying internal micro-cracks. Especially for hydraulic structures composed of complex materials such as concrete, internal potential cracks are often difficult to detect in a timely manner, especially some deep and tiny structural cracks.

[0003] To improve the detection efficiency and accuracy, researchers have gradually turned to electromagnetic wave detection technology. As a non-invasive detection method, electromagnetic wave detection can penetrate the building surface and detect the internal structure. However, existing electromagnetic wave detection technologies still face difficulties in effectively distinguishing normal structure echo signals from crack-related abnormal signals, resulting in limitations in the reliability and accuracy of detection results. Summary of the Invention

[0004] Aiming at the above deficiencies in the prior art, a method for detecting internal cracks of hydraulic structures provided by the present invention solves the problem of low detection accuracy of internal cracks of hydraulic structures existing in the prior art.

[0005] To achieve the above invention objective, the technical solution adopted by the present invention is: A method for detecting internal cracks of hydraulic structures, comprising the following steps:

[0006] S1. Transmit a sinusoidal electromagnetic wave to the hydraulic structure to obtain an electromagnetic echo;

[0007] S2. Segment the electromagnetic echo at the same time length, extract the local maximum value and local minimum value in each segment, and calculate the maximum value growth value and the maximum value decay value;

[0008] S3. According to the maximum value growth value and the maximum value decay value of each echo segment, calculate the maximum value growth anomaly degree value and the maximum value decay anomaly degree value, and construct a maximum value growth anomaly vector and a maximum value decay anomaly vector;

[0009] S4. According to the distribution of the maximum value growth value and the maximum value decay value of each echo segment, calculate the maximum value growth instability and the maximum value decay instability, and construct a maximum value growth instability vector and a maximum value decay instability vector;

[0010] S5. Use a crack feature fusion neural network to process the maximum value growth abnormal vector, the maximum value decay abnormal vector, the maximum value growth instability vector, and the maximum value decay instability vector to obtain the internal crack evaluation value of the hydraulic structure.

[0011] Further, S2 includes the following sub-steps:

[0012] S21. Segment the electromagnetic echo by the same time length to obtain multiple echo segments;

[0013] S22. Mark the local maximum and local minimum in each echo segment, arrange the local maximum and local minimum in the order of time occurrence, and construct a local maximum and minimum sequence;

[0014] S23. Calculate the maximum value growth value according to the gap between each local minimum and the next local maximum in the local maximum and minimum sequence;

[0015] S24. Calculate the maximum value decay value according to the gap between each local maximum and the next local minimum in the local maximum and minimum sequence.

[0016] Further, the formula for the maximum value growth value in S23 is: , where γ up,j is the jth maximum value growth value, E min,j is the jth local minimum, E max,j+1 is the (j + 1)th local maximum, T1 is the time length between the jth local minimum and the (j + 1)th local maximum, and j is a positive integer;

[0017] The formula for the maximum value decay value in S24 is: , where γ down,j is the jth maximum value decay value, E max,j is the jth local maximum, E min,j+1 is the (j + 1)th local minimum, and T2 is the time length between the jth local maximum and the (j + 1)th local minimum.

[0018] Further, S3 includes the following sub-steps:

[0019] S31. Screen out the maximum value growth values greater than the growth threshold in each echo segment as abnormal maximum value growth values;

[0020] S32. Screen out the maximum value decay values greater than the decay threshold in each echo segment as abnormal maximum value decay values;

[0021] S33. In each echo segment, calculate the maximum value growth abnormal degree value according to each abnormal maximum value growth value;

[0022] S34. In each echo segment, calculate the maximum and minimum attenuation anomaly degree value according to each abnormal maximum and minimum attenuation value;

[0023] S35. Use each maximum and minimum growth anomaly degree value as an element to construct a maximum and minimum growth anomaly vector;

[0024] S36. Use each maximum and minimum attenuation anomaly degree value as an element to construct a maximum and minimum attenuation anomaly vector.

[0025] Further, the formula for calculating the maximum and minimum growth anomaly degree value in S33 is: , where s up is the maximum and minimum growth anomaly degree value, γ up,un is the average value of each abnormal maximum and minimum growth value in an echo segment, and N up is the number of abnormal maximum and minimum growth values in an echo segment;

[0026] The formula for calculating the maximum and minimum attenuation anomaly degree value in S34 is: , where s down is the maximum and minimum attenuation anomaly degree value, γ down,un is the average value of each abnormal maximum and minimum attenuation value in an echo segment, and M down is the number of abnormal maximum and minimum attenuation values in an echo segment.

[0027] Further, S4 includes the following sub-steps:

[0028] S41. Calculate the average value of the maximum and minimum growth values of all echo segments to obtain the maximum and minimum growth average value;

[0029] S42. In each echo segment, calculate the standard deviation of each maximum and minimum growth value to obtain the maximum and minimum growth fluctuation value of this echo segment;

[0030] S43. Take the ratio of the maximum and minimum growth fluctuation value to the maximum and minimum growth average value as the maximum and minimum growth instability;

[0031] S44. Calculate the average value of the maximum and minimum attenuation values of all echo segments to obtain the maximum and minimum attenuation average value;

[0032] S45. In each echo segment, calculate the standard deviation of each maximum and minimum attenuation value to obtain the maximum and minimum attenuation fluctuation value of this echo segment;

[0033] S46. Take the ratio of the maximum and minimum attenuation fluctuation value to the maximum and minimum attenuation average value as the maximum and minimum attenuation instability;

[0034] S47. Use the maximum and minimum growth instability as an element to construct a maximum and minimum growth instability vector;

[0035] S48. Construct a maximum value attenuation instability vector by using the maximum value attenuation instability as an element.

[0036] Further, the crack feature fusion neural network in S5 includes: a growth data fusion layer, a decay data fusion layer, a first matrix construction layer, a second matrix construction layer, a first CNN network, a second CNN network, an adder A1, a feature self-adjustment layer, and a fully connected layer.

[0037] The first input end of the growth data fusion layer is used to input the maximum value growth abnormal vector, its second input end is used to input the maximum value growth instability vector, and its output end is connected to the input end of the first matrix construction layer; the first input end of the decay data fusion layer is used to input the maximum value decay abnormal vector, its second input end is used to input the maximum value decay instability vector, and its output end is connected to the input end of the second matrix construction layer; the input end of the first CNN network is connected to the output end of the first matrix construction layer; the input end of the second CNN network is connected to the output end of the second matrix construction layer; the first input end of the adder A1 is connected to the output end of the first CNN network, its second input end is connected to the output end of the second CNN network, and its output end is connected to the input end of the feature self-adjustment layer; the output end of the feature self-adjustment layer is connected to the input end of the fully connected layer; the output end of the fully connected layer is used as the output end of the crack feature fusion neural network.

[0038] Further, the specific processing processes of the growth data fusion layer and the decay data fusion layer are both: extracting features for each element input from the first input end through the tanh function, and splicing the features to form a first splicing vector; extracting features for each element input from the second input end through the tanh function, and splicing the features to form a second splicing vector, and multiplying the first splicing vector and the second splicing vector element by element to obtain a fusion feature vector.

[0039] Further, the expressions of the matrix construction layer are both: , where H is the fusion feature matrix, h is the fusion feature vector, and T is the transpose operation.

[0040] Further, the expression of the feature self-adjustment layer is: , where is the nth feature output by the feature self-adjustment layer, x n is the nth feature output by the adder A1, w n is the nth weight of the feature self-adjustment layer, b n is the nth bias of the feature self-adjustment layer, e is the natural constant, and n is a positive integer.

[0041] The beneficial effects of the present invention are:

[0042] 1. In order to better extract the features related to cracks, the present invention emits sinusoidal electromagnetic waves to hydraulic structures. The sinusoidal electromagnetic waves have waveform continuity and can provide a stable and uniform energy distribution. Therefore, the maximum value growth value and the maximum value decay value extracted can reflect the propagation characteristics of the sinusoidal electromagnetic waves inside the hydraulic structures.

[0043] 2. The present invention processes the electromagnetic echo in segments, extracts the local maximum value and the local minimum value in each segment, calculates the maximum value growth value and the maximum value decay value to reflect the mutation situation of the electromagnetic echo, and then calculates the maximum value growth anomaly degree value and the maximum value decay anomaly degree value to reflect the anomaly situation of the maximum value growth and the maximum value decay in a segment.

[0044] 3. The present invention calculates the maximum value growth instability degree and the maximum value decay instability degree according to the distribution of each maximum value growth value and maximum value decay value in each segment to reflect the fluctuation situation of the maximum value growth value and the maximum value decay value.

[0045] 4. When there are cracks, the cracks will cause the distortion of the propagation of electromagnetic waves, resulting in the non-uniformity of the energy distribution. The cracks will cause the medium to be discontinuous, resulting in abnormal wave reflection, scattering and attenuation characteristics. Therefore, the present invention extracts the maximum value growth anomaly degree value, the maximum value decay anomaly degree value, and the maximum value growth instability degree and the maximum value decay instability degree to reflect the abnormal form of the electromagnetic echo, effectively distinguish the normal structure echo signal from the abnormal signal of the cracks, and then uses the crack feature fusion neural network to process these data to improve the internal crack detection accuracy of the hydraulic structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flowchart of a method for detecting internal cracks in a hydraulic structure;

[0047] Figure 2 is a sinusoidal electromagnetic wave of 50 MHz;

[0048] Figure 3 is a schematic structural diagram of a crack feature fusion neural network. DETAILED DESCRIPTION OF THE INVENTION

[0049] The following describes the specific embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

[0050] As Figure 1 shown, a method for detecting internal cracks in a hydraulic structure includes the following steps:

[0051] S1. Transmit a sinusoidal electromagnetic wave to the hydraulic structure to obtain an electromagnetic echo;

[0052] S2. Segment the electromagnetic echo by the same time length, extract the local maximum and local minimum in each segment, and calculate the maximum - value growth value and the maximum - value decay value;

[0053] S3. According to the maximum - value growth value and the maximum - value decay value of each echo segment, calculate the maximum - value growth anomaly degree value and the maximum - value decay anomaly degree value, and construct a maximum - value growth anomaly vector and a maximum - value decay anomaly vector;

[0054] S4. According to the distribution of the maximum - value growth value and the maximum - value decay value of each echo segment, calculate the maximum - value growth instability degree and the maximum - value decay instability degree, and construct a maximum - value growth instability vector and a maximum - value decay instability vector;

[0055] S5. Use a crack - feature fusion neural network to process the maximum - value growth anomaly vector, the maximum - value decay anomaly vector, the maximum - value growth instability vector, and the maximum - value decay instability vector to obtain the internal crack evaluation value of the hydraulic structure.

[0056] In this embodiment, as Figure 2 shown, the expression of the sinusoidal electromagnetic wave in S1 is: , where \(E(t)\) is the sinusoidal electromagnetic wave, \(E\) is the amplitude of the transmitted sinusoidal electromagnetic wave, \(\sin\) is the sine function, \(f\) is the frequency value, and \(\varphi\) is the phase.

[0057] In this embodiment, S2 includes the following sub - steps:

[0058] S21. Segment the electromagnetic echo by the same time length to obtain multiple echo segments;

[0059] S22. Mark the local maximum and local minimum in each echo segment, arrange the local maximum and local minimum in the order of occurrence time, and construct a local maximum - minimum sequence;

[0060] For example: an echo segment is \([0.1,0.3,0.5,0.6,0.7,0.4,0.3,0.2,0.1,0.5,0.6,0.8,0.3,0.2,0.4]\), the local maximum values are \(0.7\) and \(0.8\), the local minimum values are \(0.1\) and \(0.2\), then the local maximum - minimum sequence is \([0.7,0.1,0.8,0.2]\);

[0061] S23. Calculate the maximum - value growth value according to the gap between each local minimum and the next local maximum in the local maximum - minimum sequence;

[0062] S24. Calculate the maximum - value decay value according to the gap between each local maximum and the next local minimum in the local maximum - minimum sequence.

[0063] In the present invention, the local minimum is defined as: the amplitude of the selected point is lower than the amplitudes of the neighboring points on both sides; the local maximum is defined as: the amplitude of the selected point is higher than the amplitudes of the neighboring points on both sides.

[0064] In this embodiment, the formula for the maximum / minimum growth value in S23 is: , where γ up,j is the j-th maximum / minimum growth value, E min,j is the j-th local minimum, E max,j+1 is the (j + 1)-th local maximum, T1 is the time length between the j-th local minimum and the (j + 1)-th local maximum, j is a positive integer, and E min,j and E max,j+1 are adjacent;

[0065] The formula for the maximum / minimum decay value in S24 is: , where γ down,j is the j-th maximum / minimum decay value, E max,j is the j-th local maximum, E min,j+1 is the (j + 1)-th local minimum, T2 is the time length between the j-th local maximum and the (j + 1)-th local minimum, and E max,j and E min,j+1 are adjacent.

[0066] From the perspective of structural integrity assessment, a healthy building structure enables electromagnetic waves to propagate in a relatively uniform and predictable manner. When internal cracks exist, the propagation path of electromagnetic waves changes, causing non-uniformity in energy distribution. By extracting the maximum / minimum growth value and the maximum / minimum decay value, such subtle changes in propagation characteristics can be sensitively captured.

[0067] In the analysis of wave propagation mechanism, when a sinusoidal electromagnetic wave propagates in a homogeneous medium, its energy attenuation and growth show regularity. Once encountering a crack, it will cause medium discontinuity, resulting in abnormal wave reflection, scattering, and attenuation characteristics. The maximum / minimum growth value reflects the wave energy rising speed, and the maximum / minimum decay value reflects the wave energy decay speed.

[0068] In this embodiment, S3 includes the following sub-steps:

[0069] S31. Screen out the maximum / minimum growth values greater than the growth threshold in each echo segment as abnormal maximum / minimum growth values;

[0070] S32. Screen out the maximum / minimum decay values greater than the decay threshold in each echo segment as abnormal maximum / minimum decay values;

[0071] S33. In each echo segment, calculate the maximum / minimum growth abnormality degree value according to each abnormal maximum / minimum growth value;

[0072] S34. In each echo segment, calculate the maximum and minimum attenuation anomaly degree value according to each abnormal maximum and minimum attenuation value;

[0073] S35. Use each maximum and minimum growth anomaly degree value as an element to construct a maximum and minimum growth anomaly vector;

[0074] S36. Use each maximum and minimum attenuation anomaly degree value as an element to construct a maximum and minimum attenuation anomaly vector.

[0075] In the present invention, the elements in the vector are arranged in the order of the time of occurrence.

[0076] In this embodiment, the growth threshold is the threshold set for the maximum and minimum growth value, and the attenuation threshold is the threshold set for the maximum and minimum attenuation value. The growth threshold and the attenuation threshold can be specifically set according to experience and experiments. For example, the present invention emits a 50 MHz sinusoidal electromagnetic wave. For a 50 MHz sinusoidal electromagnetic wave, the length of one cycle is 20 nanoseconds, and the distance between its adjacent local maximum and minimum values is half of the cycle length. For a 50 MHz sinusoidal electromagnetic wave: the maximum and minimum growth values and the maximum and minimum attenuation values are both 2E / (10 nanoseconds). Considering the energy attenuation, therefore, the maximum and minimum growth values and the maximum and minimum attenuation values of the electromagnetic wave echo are both 2Eμ / (10 nanoseconds), where E is the amplitude of the sinusoidal electromagnetic wave and μ is the attenuation ratio. Therefore, 2Eμ / (10 nanoseconds) can be set as the corresponding threshold.

[0077] The present invention selects the maximum and minimum growth values and the maximum and minimum attenuation values greater than the threshold to characterize the abnormal signal features related to cracks, that is, the abnormal maximum and minimum growth values and the abnormal maximum and minimum attenuation values in S31 and S32 are the abnormal signal features related to cracks, and the maximum and minimum growth values less than or equal to the growth threshold and the maximum and minimum attenuation values less than or equal to the attenuation threshold are regarded as normal echo signal features.

[0078] In this embodiment, the formula for calculating the maximum and minimum growth anomaly degree value in S33 is: , where s up is the maximum and minimum growth anomaly degree value, γ up,un is the average value of each abnormal maximum and minimum growth value in an echo segment, and N up is the number of abnormal maximum and minimum growth values in an echo segment;

[0079] The formula for calculating the maximum and minimum attenuation anomaly degree value in S34 is: , where s down is the maximum and minimum attenuation anomaly degree value, γ down,un is the average value of each abnormal maximum and minimum attenuation value in an echo segment, and M down is the number of abnormal maximum and minimum attenuation values in an echo segment.

[0080] The present invention calculates the average values of the abnormal maximum growth values and the abnormal maximum decay values of each echo segment, which reflects the abnormality degree of an echo segment. Then, the number of the abnormal maximum growth values and the abnormal maximum decay values is further used to enhance the abnormality degree. When the average value is larger and the number is more, the abnormality degree value is larger.

[0081] In this embodiment, S4 includes the following sub-steps:

[0082] S41. Calculate the average value of the maximum growth values of all echo segments to obtain the maximum growth average value;

[0083] S42. In each echo segment, calculate the standard deviation of each maximum growth value to obtain the maximum growth fluctuation value of this echo segment, where the maximum growth fluctuation value is equal to the standard deviation of each maximum growth value;

[0084] S43. Take the ratio of the maximum growth fluctuation value to the maximum growth average value as the maximum growth instability;

[0085] S44. Calculate the average value of the maximum decay values of all echo segments to obtain the maximum decay average value;

[0086] S45. In each echo segment, calculate the standard deviation of each maximum decay value to obtain the maximum decay fluctuation value of this echo segment, where the maximum decay fluctuation value is equal to the standard deviation of each maximum decay value;

[0087] S46. Take the ratio of the maximum decay fluctuation value to the maximum decay average value as the maximum decay instability;

[0088] S47. Use the maximum growth instability as an element to construct a maximum growth instability vector;

[0089] S48. Use the maximum decay instability as an element to construct a maximum decay instability vector.

[0090] The maximum growth instability reflects the fluctuation characteristics in the process of electromagnetic wave energy accumulation. In a uniform and complete building structure, the energy accumulation of electromagnetic waves should show a relatively stable and predictable trend. When there are internal cracks, this energy accumulation process will show significant irregularities. By quantifying this irregularity, tiny structural abnormalities can be sensitively captured, providing a more accurate quantitative index for crack detection.

[0091] The maximum value attenuation instability reveals the complex dynamic process of the energy loss of electromagnetic waves. In an ideal homogeneous medium, the attenuation of electromagnetic waves should follow relatively stable physical laws. However, internal cracks can cause discontinuities in the propagation path of electromagnetic waves, leading to significant changes in the energy attenuation pattern. Calculating the maximum value attenuation instability is actually quantifying the fluctuation characteristics in this non-ideal attenuation process, providing a new analysis dimension for identifying potential cracks.

[0092] As Figure 3 shown, the crack feature fusion neural network in S5 includes: a growth data fusion layer, an attenuation data fusion layer, a first matrix construction layer, a second matrix construction layer, a first CNN network, a second CNN network, an adder A1, a feature self-adjustment layer, and a fully connected layer;

[0093] The first input end of the growth data fusion layer is used to input the maximum value growth abnormal vector, its second input end is used to input the maximum value growth instability vector, and its output end is connected to the input end of the first matrix construction layer; the first input end of the attenuation data fusion layer is used to input the maximum value attenuation abnormal vector, its second input end is used to input the maximum value attenuation instability vector, and its output end is connected to the input end of the second matrix construction layer; the input end of the first CNN network is connected to the output end of the first matrix construction layer; the input end of the second CNN network is connected to the output end of the second matrix construction layer; the first input end of the adder A1 is connected to the output end of the first CNN network, its second input end is connected to the output end of the second CNN network, and its output end is connected to the input end of the feature self-adjustment layer; the output end of the feature self-adjustment layer is connected to the input end of the fully connected layer; the output end of the fully connected layer is used as the output end of the crack feature fusion neural network.

[0094] In this embodiment, the specific processing processes of the growth data fusion layer and the attenuation data fusion layer are both: extracting features from each element input to the first input end through the tanh function, and splicing the features to form a first splicing vector; extracting features from each element input to the second input end through the tanh function, and splicing the features to form a second splicing vector, and multiplying the first splicing vector and the second splicing vector element by element to obtain a fusion feature vector.

[0095] The expression for extracting features by the tanh function is: , where is the m-th feature extracted by the tanh function, r m is the m-th element, w m is the weight of r m , b m is the bias of r m , m is a positive integer, and f is the tanh function.

[0096] For the growth data fusion layer, its input data includes: the maximum growth anomaly vector and the maximum growth instability vector. Therefore, after processing each element through the tanh function, the feature concatenation vector corresponding to the maximum growth anomaly vector and the feature concatenation vector corresponding to the maximum growth instability vector are obtained. Multiplying these two concatenation vectors element by element realizes the fusion of the maximum growth anomaly features belonging to the same segment of electromagnetic echo.

[0097] For the decay data fusion layer, its input data includes: the maximum decay anomaly vector and the maximum decay instability vector. Therefore, after processing each element through the tanh function, the feature concatenation vector corresponding to the maximum decay anomaly vector and the feature concatenation vector corresponding to the maximum decay instability vector are obtained. Multiplying these two concatenation vectors element by element realizes the fusion of the maximum decay anomaly features belonging to the same segment of electromagnetic echo.

[0098] In this embodiment, the expressions of the matrix construction layer are all: , where H is the fusion feature matrix, h is the fusion feature vector, and T is the transpose operation.

[0099] The present invention uses the matrix construction layer to transform the vectors output by the growth data fusion layer and the decay data fusion layer into matrices, which is convenient for the CNN network to extract features. Then, through the adder A1, bit-by-bit addition is realized, and the features in terms of maximum growth and maximum decay are fused to achieve a comprehensive evaluation of the crack situation.

[0100] In this embodiment, the expression of the feature self-adjustment layer is: , where is the nth feature output by the feature self-adjustment layer, x n is the nth feature output by the adder A1, w n is the nth weight of the feature self-adjustment layer, b n is the nth bias of the feature self-adjustment layer, e is the natural constant, and n is a positive integer.

[0101] The present invention uses the feature self-adjustment layer to allocate the adjustment ratio , so that the crack feature fusion neural network has the ability to automatically adjust the feature attention.

[0102] In this embodiment, the weights and biases in the crack feature fusion neural network can be obtained through the existing gradient descent method. For example, collect multiple maximum growth abnormal vectors, maximum decay abnormal vectors, maximum growth unstable vectors, and maximum decay unstable vectors as samples, and use the manually annotated internal crack evaluation value of the hydraulic structure as the label. Input each sample into the crack feature fusion neural network. According to the difference between the actual evaluation value output by the crack feature fusion neural network and the label, based on the existing mean square error loss function, obtain the gap between the actual and the label, and update each weight and bias according to the gap. The formula for updating the weights and biases in the implementation of this application is:

[0103] , , where w k is the weight at the k-th training, w k-1 is the weight at the (k - 1)-th training, η is a hyperparameter, η is used to control the step size for each parameter update, L is the mean square error loss function, b k is the bias at the k-th training, b k-1 is the bias at the (k - 1)-th training, and k is the numbering of the training times.

[0104] In this embodiment, the range of the label in the training process is set to 0 - 1. Therefore, the output range of the crack feature fusion neural network is between 0 - 1. When the internal crack evaluation value output by the fully connected layer is greater than the evaluation threshold of 0.5, such as the evaluation value is 0.7, it indicates that the possibility of cracks existing is relatively large, and it can be preliminarily judged that there are cracks inside the hydraulic structure. When the internal crack evaluation value output by the fully connected layer is less than or equal to 0.5, such as the evaluation value is 0.3, it indicates that the possibility of cracks existing is relatively small. The specific threshold setting can be verified through multiple experiments. The label setting range and the evaluation threshold are set according to specific needs, and are not limited to the values given in this invention. The label setting range is artificially set during the training process, and the specific setting range does not affect the implementation of this invention. The evaluation threshold can take the middle value of the label range, which is convenient to divide the output into two categories. When it is greater than the evaluation threshold, the possibility of cracks existing is greater, and when it is less than the evaluation threshold, the possibility of cracks existing is smaller.

[0105] In order to better extract the features related to cracks in this invention, a sine electromagnetic wave is emitted to the hydraulic structure. The sine electromagnetic wave has waveform continuity and can provide a stable and uniform energy distribution. Therefore, the maximum growth value and the maximum decay value extracted can reflect the propagation characteristics of the sine electromagnetic wave inside the hydraulic structure.

[0106] The present invention processes electromagnetic echoes in segments, extracts local maximum values and local minimum values by segment, calculates the growth values of the maximum and minimum values and the decay values of the maximum and minimum values, reflects the mutation situation of the electromagnetic echoes, and then calculates the abnormal degree values of the growth of the maximum and minimum values and the abnormal degree values of the decay of the maximum and minimum values, reflecting the abnormal situation of the growth of the maximum and minimum values and the abnormal situation of the decay of the maximum and minimum values in a segment.

[0107] The present invention calculates the instability degree of the growth of the maximum and minimum values and the instability degree of the decay of the maximum and minimum values according to the distribution of the growth values of the maximum and minimum values and the decay values of the maximum and minimum values in each segment, reflecting the fluctuation situation of the growth values of the maximum and minimum values and the decay values of the maximum and minimum values.

[0108] When there are cracks, the cracks will cause the distortion of the propagation of electromagnetic waves, resulting in the non-uniformity of the energy distribution. The cracks will cause the medium to be discontinuous, resulting in abnormal reflection, scattering and attenuation characteristics of the waves. Therefore, the present invention extracts the abnormal degree values of the growth of the maximum and minimum values, the abnormal degree values of the decay of the maximum and minimum values, and the instability degree of the growth of the maximum and minimum values and the instability degree of the decay of the maximum and minimum values to reflect the abnormal form of the electromagnetic echoes, effectively distinguishing the normal structure echo signals from the abnormal signals of the cracks, and then using the crack feature fusion neural network to process these data to improve the internal crack detection accuracy of hydraulic structures.

[0109] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting internal cracks in a hydraulic structure, characterized in that: The following steps are involved: S1. Transmit sinusoidal electromagnetic waves to hydraulic structures to obtain electromagnetic echoes; S2, dividing the electromagnetic echo into segments of the same time length, extracting the local maximum and local minimum in each segment, and calculating the maximum growth value and the maximum attenuation value; The S2 comprises the following sub-steps: S21, dividing the electromagnetic echo into segments of the same time length to obtain multiple echo segments; S22, marking the local maximum and the local minimum in each echo segment, arranging the local maximum and the local minimum in the order of their occurrence, and constructing a local maximum sequence; S23, calculating the maximum value growth value according to the gap between each local minimum value and the next local maximum value in the local maximum value sequence; The formula for the maximum growth value in S23 is: , where γ up，j is the jth maximum growth value, E min，j is the jth local minimum, E max，j+1 is the j+1th local maximum, T1 is the time length between the jth local minimum and the j+1th local maximum, and j is a positive integer; S24, calculating the maximum attenuation value according to the gap between each local maximum value and the next local minimum value in the local maximum value sequence; The formula for the maximum attenuation value in S24 is: , where γ down,j is the jth maximum attenuation value, E max，j is the jth local maximum, E min，j+1 is the j+1th local minimum, T2 is the time length between the jth local maximum and the j+1th local minimum; S3, according to the maximum growth value and the maximum attenuation value of each echo segment, calculate the maximum growth abnormality degree value and the maximum attenuation abnormality degree value, and construct the maximum growth abnormality vector and the maximum attenuation abnormality vector; The S3 comprises the following sub-steps: S31, screening out the maximum growth value greater than the growth threshold in each echo segment as the abnormal maximum growth value; S32, screening out the maximum attenuation value greater than the attenuation threshold in each echo segment as the abnormal maximum attenuation value; S33, in each echo segment, according to each abnormal maximum growth value, calculating the maximum growth abnormality degree value; The formula for calculating the maximum growth abnormality value in S33 is: , where s up is the maximum growth abnormality value, γ up，un is the average value of the maximum growth of each anomaly in an echo segment, N up It is the number of abnormal maximum growth values ​​in an echo segment; S34, in each echo segment, according to each abnormal maximum attenuation value, calculating the maximum attenuation abnormality degree value; The formula for calculating the maximum attenuation abnormality value in S34 is: , where s down is the maximum attenuation abnormality value, γ down，un is the average value of the maximum attenuation value of each abnormality in an echo segment, M down It is the number of abnormal maximum attenuation values ​​in an echo segment; S35, taking each maximum growth anomaly degree value as an element, constructing a maximum growth anomaly vector; S36, taking each maximum attenuation anomaly degree value as an element, constructing a maximum attenuation anomaly vector; S4. Calculate the maximum growth instability and the maximum attenuation instability according to the distribution of the maximum growth value and the maximum attenuation value of each echo segment, and construct the maximum growth instability vector and the maximum attenuation instability vector; The S4 comprises the following sub-steps: S41, calculating the average value of the maximum growth values ​​of all echo segments to obtain the maximum growth average value; S42, in each echo segment, calculating the standard deviation of each maximum growth value to obtain the maximum growth fluctuation value of the echo segment; S43, taking the ratio of the maximum growth fluctuation value to the maximum growth average value as the maximum growth instability; S44, calculating the average value of the maximum attenuation values ​​of all echo segments to obtain the maximum attenuation average value; S45, in each echo segment, calculating the standard deviation of each maximum attenuation value to obtain the maximum attenuation fluctuation value of the echo segment; S46, taking the ratio of the maximum attenuation fluctuation value to the maximum attenuation average value as the maximum attenuation instability; S47, taking the maximum growth instability as an element, constructing a maximum growth instability vector; S48, taking the maximum attenuation instability as an element, constructing a maximum attenuation instability vector; S5. Use crack feature fusion neural network to process the maximum growth anomaly vector, the maximum decay anomaly vector, the maximum growth unstable vector and the maximum decay unstable vector to obtain the internal crack assessment value of the hydraulic structure.

2. The internal crack detection method of a hydraulic structure according to claim 1, characterized in that: The crack feature fusion neural network in S5 includes: a growth data fusion layer, a decay data fusion layer, a first matrix construction layer, a second matrix construction layer, a first CNN network, a second CNN network, an adder A1, a feature self-adjustment layer and a fully connected layer; The first input end of the growth data fusion layer is used to input the maximum growth abnormal vector, the second input end is used to input the maximum growth unstable vector, and the output end is connected to the input end of the first matrix construction layer; The first input end of the attenuation data fusion layer is used to input the maximum attenuation abnormal vector, the second input end is used to input the maximum attenuation unstable vector, and the output end is connected to the input end of the second matrix construction layer; The input end of the first CNN network is connected to the output end of the first matrix construction layer; The input end of the second CNN network is connected to the output end of the second matrix construction layer; The first input end of the adder A1 is connected to the output end of the first CNN network, the second input end thereof is connected to the output end of the second CNN network, and the output end thereof is connected to the input end of the feature self-adjusting layer; The output end of the feature self-adjustment layer is connected to the input end of the fully connected layer; The output end of the fully connected layer serves as the output end of the crack feature fusion neural network.

3. The internal crack detection method of a hydraulic structure according to claim 2, characterized in that: The specific processing processes of the growth data fusion layer and the attenuation data fusion layer are: extracting features from each element input at the first input end through the tanh function, and splicing the features to form a first splicing vector; extracting features from each element input at the second input end through the tanh function, and splicing the features to form a second splicing vector, and multiplying the first splicing vector and the second splicing vector element by element to obtain a fused feature vector.

4. The internal crack detection method of a hydraulic structure according to claim 3, characterized in that: The expressions of the matrix construction layers are all: , where H is the fused feature matrix, h is the fused feature vector, and T is the transpose operation.

5. The internal crack detection method of a hydraulic structure according to claim 4, characterized in that: The expression of the characteristic self-adjusting layer is: ,in, is the nth feature output by the feature self-adjustment layer, x n is the nth feature output by adder A1, w n is the nth weight of the feature self-adjusting layer, b n is the nth bias of the feature self-adjusting layer, e is a natural constant, and n is a positive integer.

Images