Ultrasonic imaging method for concrete beam damage detection

Abstract

The invention relates to the field of concrete beam damage detection, in particular to an ultrasonic imaging method for concrete beam damage detection, which comprises the following steps: arranging a linear array comprising a plurality of ultrasonic probes on a single-side contactable surface of a to-be-detected concrete beam; preparing a multi-channel ultrasonic transmitting and receiving instrument, and connecting the multi-channel ultrasonic transmitting and receiving instrument with the sensor array and the control computer; a multi-channel ultrasonic device is used to work in a full-matrix capture mode, in the full-matrix capture mode, each probe is sequentially excited to emit ultrasonic pulses, and time domain signals received by all the probes of the array are synchronously recorded to obtain an original full-matrix data set. According to the method, strong scattering noise generated by the concrete aggregate can be inhibited through the deep convolutional neural network so as to improve the signal-to-noise ratio of the image, the sound velocity field conforming to the non-uniform characteristic of the concrete can be constructed through local sound velocity inversion and iteration based on the strong reflection points, the problems of imaging artifacts and form distortion are avoided, and the accuracy of the imaging process is improved. And the concrete damage detection precision is ensured.

Description

Technical Field

[0001] This invention relates to the field of concrete beam damage detection technology, and in particular to an ultrasonic imaging method for concrete beam damage detection. Background Technology

[0002] Concrete beams are crucial load-bearing components in modern buildings, bridges, and various infrastructures. Under the combined effects of long-term loads, environmental erosion, and material aging, they are prone to developing hidden damage such as cracks, voids, and spalling, which seriously threaten structural safety and durability.

[0003] Non-destructive testing of internal damage in concrete beams primarily relies on ultrasonic technology. However, when existing ultrasonic imaging methods are applied to concrete, the aggregates within the concrete generate strong random scattering signals, severely interfering with imaging, resulting in blurred images and the inability to identify minute damage. Furthermore, the actual spatial distribution of sound velocity in concrete is uneven, and traditional SAFT (Self-Assessing and Analyzing Ultrasonic Forces) uses a single average sound velocity for calculation, causing imaging artifacts, inaccurate positioning, and morphological distortion. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an ultrasonic imaging method for detecting damage in concrete beams.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An ultrasonic imaging method for detecting damage in concrete beams includes the following steps:

[0007] S1: Deployment detection;

[0008] A: A linear array containing multiple ultrasonic probes is arranged on one accessible surface of the concrete beam to be tested.

[0009] B: Prepare a multi-channel ultrasonic transmitter and receiver, and connect it to the sensor array and control computer;

[0010] C: Use a multi-channel ultrasound device to operate in full matrix capture mode. In full matrix capture mode, each probe is excited to emit an ultrasonic pulse in sequence, and the time domain signals received by all probes in the array are recorded synchronously to obtain the original full matrix dataset.

[0011] S2: Data preprocessing and noise suppression;

[0012] A: Perform routine preprocessing on the raw data;

[0013] B: Input the preprocessed full matrix data into the pre-trained deep convolutional neural network model;

[0014] C: Obtain the descattering enhanced signal matrix of the network output, in which the background noise of material inhomogeneity is suppressed, while the damage-related signal features are preserved and enhanced;

[0015] S3: Model Imaging and Computation;

[0016] A: Set an initial uniform sound velocity field C0, and use the enhanced signal matrix obtained above to perform the first round of synthetic aperture focusing technology processing to generate the initial imaging result L1;

[0017] B: In L1, automatically identify several strong reflection points where the energy is focused most clearly;

[0018] C: Based on the signal travel time of strong reflection points on the transmit-receive path, a local sound velocity inversion problem is constructed, and the sound velocity distribution in the area surrounding these points is inverted by the ray tracing tomography method.

[0019] D: Using the aforementioned local sound velocity values ​​as constraints, update the two-dimensional sound velocity field model of the imaging region to optimize C0 into C1(x,z);

[0020] E: Recalculate the precise wave propagation time of the pixel using the updated sound velocity field C1(x,z), and perform synthetic aperture focusing superposition on the enhanced signal matrix again to generate the focused imaging result I2. Repeat this step to obtain the final high-resolution image L_final.

[0021] S4: Damage assessment;

[0022] A: Smooth the final image L_final to show the location, shape, and relative scale of the internal damage;

[0023] B: Based on the design drawings and structure of the concrete beam, analyze the shape, size, depth, and orientation of the highlighted areas in the image to determine the type and severity of the damage.

[0024] Preferably, the number of ultrasonic probes in step S1-A is greater than 16.

[0025] Preferably, the conventional preprocessing in step S2-A includes DC offset removal, bandpass filtering, and time-varying gain compensation.

[0026] Preferably, the deep convolutional neural network model in step S2-B is trained based on concrete sample data to identify and separate incoherent noise caused by random scattering of aggregates from coherent reflection signals generated by macroscopic damage interfaces.

[0027] Preferably, the input to the deep convolutional neural network model is a preprocessed fragment of full matrix data collected from healthy concrete samples or an initial low-quality SAFT image patch generated therefrom. The network's supervision label is the clean signal or image of the corresponding region after removing aggregate scattering noise, and it retains only the reflection signal features caused by macroscopic damage.

[0028] Preferably, the two-dimensional sound velocity field model in step S3-D is updated through spatial interpolation and extrapolation.

[0029] Preferably, the process in step S3-E can be iterated 1 to 2 times until the sound velocity field model is stable and the final high-resolution image is obtained.

[0030] Preferably, the imaging smoothing process in step S4-A is threshold segmentation and feature extraction.

[0031] Preferably, the damage types in step S4-B include horizontal cracks, diagonal cracks, and cavities.

[0032] The beneficial effects of this invention are as follows: the strong scattering noise generated by concrete aggregate can be suppressed by deep convolutional neural networks to improve the image signal-to-noise ratio; by inverting and iterating the local sound velocity based on strong reflection points, a sound velocity field that conforms to the non-uniform characteristics of concrete can be constructed, avoiding the problems of imaging artifacts and morphological distortion, and ensuring the accuracy of concrete damage detection. Detailed Implementation

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0034] An ultrasonic imaging method for detecting damage in concrete beams includes the following steps:

[0035] S1: Deployment detection;

[0036] A: A linear array containing multiple ultrasonic probes is arranged on one accessible surface of the concrete beam to be tested; and by adopting a one-sided detection mode, it is applicable to situations where only one surface can be contacted on site (such as the bottom of the beam), thus improving the practicality of this method.

[0037] B: Prepare a multi-channel ultrasonic transmitter and receiver, and connect it to the sensor array and control computer;

[0038] C: Use a multi-channel ultrasound device to operate in full matrix capture mode. In full matrix capture mode, each probe is excited to emit an ultrasonic pulse in sequence, and the time domain signals received by all probes in the array are recorded synchronously to obtain the original full matrix dataset.

[0039] S2: Data preprocessing and noise suppression;

[0040] A: Perform routine preprocessing on the raw data;

[0041] B: Input the preprocessed full matrix data into the pre-trained deep convolutional neural network model;

[0042] C: Obtain the descattering enhanced signal matrix of the network output, in which the background noise of material inhomogeneity is suppressed, while the damage-related signal features are preserved and enhanced;

[0043] S3: Model Imaging and Computation;

[0044] A: Set an initial uniform sound velocity field C0, and use the enhanced signal matrix obtained above to perform the first round of synthetic aperture focusing technology to generate an initial imaging result L1; this can provide an imaging starting point, and the first imaging using the enhanced signal can obtain a clearer initial result than using the original signal, which is beneficial for the subsequent accurate identification of strong reflection points used for inversion.

[0045] B: In L1, automatically identify several strong reflection points where the energy is focused most clearly;

[0046] C: Based on the signal travel time of strong reflection points on the transmit-receive path, a local sound velocity inversion problem is constructed, and the sound velocity distribution in the area surrounding these points is inverted by the ray tracing tomography method.

[0047] D: Using the aforementioned local sound velocity values ​​as constraints, update the two-dimensional sound velocity field model of the imaging region to optimize C0 into C1(x,z);

[0048] E: Recalculate the precise wave propagation time of the pixel using the updated sound velocity field C1(x,z), and perform synthetic aperture focusing superposition on the enhanced signal matrix again to generate the focused imaging result I2. Repeat this step to obtain the final high-resolution image L_final.

[0049] S4: Damage assessment;

[0050] A: Smooth the final image L_final to show the location, shape, and relative scale of the internal damage;

[0051] B: Based on the design drawings and structure of the concrete beam, analyze the shape, size, depth, and orientation of the highlighted areas in the image to determine the type and severity of the damage.

[0052] In step S1-A, the number of ultrasound probes is greater than 16. This ensures sufficient spatial sampling density to obtain high spatial resolution imaging results.

[0053] The routine preprocessing in step S2-A includes DC offset removal, bandpass filtering, and time-varying gain compensation.

[0054] In step S2-B, the deep convolutional neural network model is trained based on concrete sample data and is used to identify and separate incoherent noise caused by random scattering of aggregates and coherent reflection signals generated by macroscopic damage interfaces.

[0055] The input to the deep convolutional neural network model is a preprocessed fragment of full matrix data collected from healthy concrete samples or an initial low-quality SAFT image patch generated from it. The network's supervision label is the clean signal or image of the corresponding region after removing aggregate scattering noise, and it only retains the reflection signal features caused by macroscopic damage.

[0056] In step S3-D, the two-dimensional sound velocity field model is updated through spatial interpolation and extrapolation.

[0057] The process in steps S3-E can be iterated 1 to 2 times until the sound velocity field model stabilizes and the final high-resolution image is obtained. This is because a small number of iterations are usually sufficient to stabilize the sound velocity field.

[0058] In step S4-A, the imaging smoothing process involves threshold segmentation and feature extraction.

[0059] The damage types in step S4-B include horizontal cracks, diagonal cracks, and cavities.

[0060] In this embodiment, a linear array sensor containing at least 16 ultrasonic probes is first arranged on one accessible surface of the concrete beam to be tested. This array is then connected to a multi-channel ultrasonic transmitter / receiver for control by a computer. The system employs a single-sided arrangement, solving the problem of not being able to simultaneously place equipment on both sides of the beam on-site.

[0061] The system then operates in full-matrix capture mode, sequentially exciting each probe in the array to emit ultrasonic pulses at a specific center frequency, while simultaneously recording the time-domain reflection signals received by all probes in the array. This process is repeated until each probe has completed its emission, thus obtaining a complete raw dataset containing all possible sound wave propagation paths.

[0062] After data acquisition, preprocessing is performed to preserve valid frequency bands. The preprocessed data is then input into a deep convolutional neural network model to obtain the enhanced signal matrix.

[0063] An initial uniform sound velocity value is set, and the enhanced signal matrix is ​​processed using the first round of synthetic aperture focusing technology to generate an initial image. Several strong reflection points with good energy focusing are automatically identified in this image. Using the precise travel time information of these points along all transmit and receive paths, the sound velocity distribution in their local areas is inverted using ray-tracing tomography.

[0064] Then, using this inversion result as a constraint, the two-dimensional non-uniform sound velocity field model of the imaging region is updated through spatial interpolation. The wave propagation time is then recalculated using the updated sound velocity field, and the enhanced signal matrix is ​​re-focused using synthetic aperture to generate a higher-resolution image. The sound velocity field inversion and focusing imaging process can be iterated 1 to 2 times until the result stabilizes, thus obtaining a high-resolution final image.

[0065] Finally, the imaging results are interpreted and evaluated. Combining the design drawings and structural knowledge of the beam, the shape, size, depth and orientation of the damaged area are analyzed to determine the type and severity of damage such as cracks and voids.

[0066] In this invention, a deep convolutional neural network can suppress strong scattering noise generated by concrete aggregates to improve the image signal-to-noise ratio. By inverting and iterating the local sound velocity based on strong reflection points, a sound velocity field that conforms to the non-uniform characteristics of concrete can be constructed, avoiding the problems of imaging artifacts and morphological distortion, and ensuring the accuracy of concrete damage detection.

[0067] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An ultrasonic imaging method for detecting damage in concrete beams, characterized in that, Includes the following steps: S1: Deployment detection; A: A linear array containing multiple ultrasonic probes is arranged on one accessible surface of the concrete beam to be tested. B: Prepare a multi-channel ultrasonic transmitter and receiver, and connect it to the sensor array and control computer; C: Use a multi-channel ultrasound device to operate in full matrix capture mode. In full matrix capture mode, each probe is excited to emit an ultrasonic pulse in sequence, and the time domain signals received by all probes in the array are recorded synchronously to obtain the original full matrix dataset. S2: Data preprocessing and noise suppression; A: Perform routine preprocessing on the raw data; B: Input the preprocessed full matrix data into the pre-trained deep convolutional neural network model; C: Obtain the descattering enhanced signal matrix of the network output, in which the background noise of material inhomogeneity is suppressed, while the damage-related signal features are preserved and enhanced; S3: Model Imaging and Computation; A: Set an initial uniform sound velocity field C0, and use the enhanced signal matrix obtained above to perform the first round of synthetic aperture focusing technology processing to generate the initial imaging result L1; B: In L1, automatically identify several strong reflection points where the energy is focused most clearly; C: Based on the signal travel time of strong reflection points on the transmit-receive path, a local sound velocity inversion problem is constructed, and the sound velocity distribution in the area surrounding these points is inverted by the ray tracing tomography method. D: Using the aforementioned local sound velocity values ​​as constraints, update the two-dimensional sound velocity field model of the imaging region to optimize C0 into C1(x,z); E: Recalculate the precise wave propagation time of the pixel using the updated sound velocity field C1(x,z), and perform synthetic aperture focusing superposition on the enhanced signal matrix again to generate the focused imaging result I2. Repeat this step to obtain the final high-resolution image L_final. S4: Damage assessment; A: Smooth the final image L_final to show the location, shape, and relative scale of the internal damage; B: Based on the design drawings and structure of the concrete beam, analyze the shape, size, depth, and orientation of the highlighted areas in the image to determine the type and severity of the damage.

2. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The number of ultrasonic probes in step S1-A is greater than 16.

3. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The conventional preprocessing in step S2-A includes DC offset removal, bandpass filtering, and time-varying gain compensation.

4. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The deep convolutional neural network model in step S2-B is trained based on concrete sample data and is used to identify and separate incoherent noise caused by random scattering of aggregates and coherent reflection signals generated by macroscopic damage interfaces.

5. The ultrasonic imaging method for detecting damage in concrete beams according to claim 4, characterized in that, The input to the deep convolutional neural network model is a preprocessed fragment of full matrix data collected from healthy concrete samples or an initial low-quality SAFT image patch generated from it. The network's supervision label is the clean signal or image of the corresponding region after removing aggregate scattering noise, and it only retains the reflection signal features caused by macroscopic damage.

6. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The two-dimensional sound velocity field model in step S3-D is updated through spatial interpolation and extrapolation.

7. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The process in step S3-E can be iterated 1 to 2 times until the sound velocity field model is stable and the final high-resolution image is obtained.

8. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The imaging smoothing process in step S4-A involves threshold segmentation and feature extraction.

9. The ultrasonic imaging method for detecting damage in concrete beams according to claim 1, characterized in that, The damage types in step S4-B include horizontal cracks, diagonal cracks, and cavities.