Method and device for identifying void defect of steel plate concrete structure

Abstract

The invention relates to a method and a device for identifying a void defect of a steel plate concrete structure. The method comprises the following steps: acquiring a to-be-identified audio signal of the steel plate concrete structure acquired through excitation conditions; extracting feature information on each frame of the preprocessed audio signal, wherein the feature information at least comprises a Mel-frequency cepstrum coefficient, root mean square energy and two items of a spectral centroid statistical mean value and a standard deviation; the extracted frame feature information is input into a void defect recognition model, void confidence of each frame is obtained, and the void defect recognition model is obtained through training by utilizing a marked audio sample training set based on a support vector machine; in response to the fact that the void confidence coefficient corresponding to each frame of audio is higher than a confidence threshold value, judging that the corresponding frame is a void frame; and taking a judgment result of the majority frames as a final void defect identification result of the audio signal to be identified. The problems that a steel plate concrete structure is low in small-size void defect sensitivity and is greatly interfered by the thickness of a steel plate and an internal complex structure are solved.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing and defect identification technology for civil engineering structures, and in particular to a method and apparatus for identifying void defects in steel plate concrete structures. Background Technology

[0002] Steel-concrete composite (SC) structures are widely used in nuclear power engineering, large industrial buildings, and critical infrastructure due to their excellent mechanical properties and durability. However, due to differences in material properties, fluctuations in construction processes, and long-term exposure to harsh environments, localized voids can easily appear at the steel-concrete interface. These defects not only affect the overall stiffness and load-bearing capacity of the structure but may also accelerate steel corrosion and reduce structural durability. Therefore, effective identification of void defects is of great significance. Currently, void detection technologies for such composite structures mainly draw on methods used for steel-concrete composite structures, including the impact vibration method and elastic wave computed tomography (CT). These methods are based on stress wave propagation theory and determine the interface bonding state by analyzing the wave propagation characteristics in the medium, showing a certain degree of effectiveness in identifying void defects over large areas. However, steel-concrete composite structures have their own unique characteristics: on the one hand, the steel plate thickness is relatively large (usually 20mm to 40mm), significantly affecting the propagation path and attenuation characteristics of elastic waves; on the other hand, the structure often has dense shear studs, tie bars, and other connecting components, resulting in a complex internal structure that significantly interferes with non-destructive testing methods based on wave propagation theory, significantly reducing their sensitivity to small-sized void defects and increasing identification errors. In recent years, with the development of artificial intelligence technology, machine learning and deep learning methods have been gradually applied in structural health monitoring and defect identification. By learning features and patterns from large amounts of data, they can achieve efficient and automated defect identification. However, in the identification of void defects in steel-concrete composite structures, how to stably extract features related to the void state from audio signals collected in complex environments and build a highly adaptable and fast identification model still lacks a systematic solution. Therefore, developing a steel-concrete composite void defect identification method and system that can adapt to complex working conditions and has high accuracy and strong robustness has an urgent practical need and important engineering application value. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and device for identifying void defects in steel plate concrete structures, addressing the problems that current detection methods have low sensitivity to small-sized void defects and are greatly affected by the thickness of the steel plate and the complex internal structure.

[0004] The present invention provides a method for identifying void defects in steel-concrete composite structures, the method comprising:

[0005] Acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions;

[0006] Feature information is extracted from each frame of the preprocessed audio signal. The feature information includes at least two items from the following: Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid.

[0007] The extracted frame feature information is input into the void removal defect recognition model to obtain the void removal confidence of each output frame. The void removal defect recognition model is a classification model trained on a support vector machine using a labeled audio sample training set.

[0008] If the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is determined to be a de-empty frame; otherwise, it is a non-de-empty frame.

[0009] The classification results of all frames are statistically analyzed, and the judgment results of the majority of frames are used as the final defect identification result of the audio signal to be identified.

[0010] In one embodiment, the excitation condition is achieved by using a solid iron or steel hammer with a spherical end.

[0011] In one embodiment, the audio signal to be identified is acquired by a microphone or pickup with a sampling rate ≥44.1kHz.

[0012] In one embodiment, the preprocessing includes bandpass filtering, noise reduction, framing, and trimming operations. The bandpass filtering frequency band is 0kHz to 20kHz. The framing is implemented using a Hamming window with a window length of 20ms to 40ms and a step size of 10ms to 20ms. The trimming duration is 0.05s to 3s. When the audio duration is insufficient, it is automatically padded with zeros to the specified duration.

[0013] In one embodiment, the extraction of feature information includes:

[0014] Extract at least 10 dimensions of Mel frequency cepstral coefficients from each frame of the audio signal.

[0015] In one embodiment, the labeled audio sample training set is divided into a training set, a validation set, and a test set according to a preset ratio, wherein the training set accounts for 60% to 80%, the validation set accounts for 10% to 20%, and the test set accounts for 10% to 20%.

[0016] In one embodiment, the kernel function of the support vector machine is one of the Sigmoid function, the polynomial kernel function, and the Gaussian kernel function.

[0017] In one embodiment, the confidence threshold is set in the range of 0.3 to 0.7.

[0018] In one embodiment, the method further includes:

[0019] The software system that deploys the void defect identification model on a desktop computing terminal supports users to import audio signal files or collect signals in real time through an audio interface as input, so as to automatically output void identification results.

[0020] The present invention also provides a device for identifying void defects in steel plate concrete structures, the device comprising:

[0021] The signal acquisition module is used to acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions;

[0022] The feature information extraction module is used to extract feature information from each frame of the preprocessed audio signal. The feature information includes at least two items from Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid.

[0023] The confidence acquisition module is used to input the extracted frame feature information into the void removal defect recognition model to obtain the void removal confidence of each frame output. The void removal defect recognition model is a classification model based on support vector machine and trained using a labeled audio sample training set.

[0024] The audio frame determination module is used to determine that if the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is a de-empty frame; otherwise, it is a non-de-empty frame.

[0025] The recognition result output module is used to statistically analyze the classification results of all frames, and the judgment results of the majority of frames are used as the final denuclearization defect recognition result of the audio signal to be recognized.

[0026] The aforementioned method and apparatus for identifying void defects in steel-concrete composite structures effectively solves the technical problems of low sensitivity to small-sized void defects and significant interference from steel plate thickness and complex internal structures in current detection methods. Specifically, firstly, acquiring audio signals under excitation conditions provides effective raw data input for identification. Next, by meticulously extracting Mel-frequency cepstral coefficients, root-mean-square energy, and spectral centroid statistical features from each frame of the signal, subtle acoustic feature changes caused by void defects can be captured from multiple dimensions in the time-frequency domain. This significantly enhances the sensitivity to small-sized or localized void defects and overcomes the drawbacks of feature averaging and overwhelmed characteristics when processing entire signal segments using traditional methods. Then, the extracted frame features are input into a pre-trained support vector machine classification model. Leveraging the model's powerful ability to handle high-dimensional features and nonlinear problems, the void confidence score for each frame is accurately output, achieving intelligent identification and classification of complex acoustic patterns. Finally, by comparing the confidence score of each frame with a preset threshold, frame-level fine-grained determination of the audio signal is achieved. Finally, based on the judgment results of all frames, the majority decision principle is adopted to obtain the overall recognition conclusion, which effectively suppresses misjudgments caused by local signal fluctuations or accidental interference. Thus, it achieves efficient, stable and accurate identification of void defects in steel plate concrete structures under complex working conditions, and improves the robustness and practicality of the method. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0028] Figure 1 This is a flowchart of a method for identifying void defects in steel plate concrete structures according to an embodiment of the present invention;

[0029] Figure 2 Flowchart for frame-level null classification and determination;

[0030] Figure 3 This is a schematic diagram of the steel plate concrete structure of the present invention;

[0031] Figure 4 This is the computer desktop application interface of the present invention;

[0032] Figure 5 A schematic diagram of a device for identifying void defects in steel plate concrete structures according to an embodiment of the invention;

[0033] Figure 6 This is an internal structural diagram of a computer device according to one embodiment. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] The following is combined with Figures 1-6 The present invention describes a method and apparatus for identifying void defects in steel plate concrete structures.

[0036] like Figure 1 and Figure 2 As shown, in one embodiment, a method for identifying void defects in a steel-concrete composite structure includes the following steps:

[0037] Step S110: Acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions.

[0038] The excitation conditions utilize a solid iron or steel hammer with a spherical end for vibration. Standardized excitation tools ensure stable and repeatable audio signals during the striking process, effectively simulating the acoustic differences between voided and solid areas. This avoids the problem of large signal fluctuations caused by inconsistent excitation, improves sensitivity to small voids (e.g., 10mm × 10mm), and reduces interference from steel plate thickness on signal transmission, thereby enhancing recognition accuracy and repeatability. The acquisition device uses a microphone or other pickup with a sampling rate ≥44.1kHz. The high sampling rate ensures high-fidelity audio signal capture, accurately recording subtle spectral changes (such as energy attenuation and frequency shift) caused by void defects. Addressing the challenge of background noise in complex environments such as nuclear power plants, improved signal resolution solves the problem of small void features being easily masked, enhancing applicability and recognition accuracy under harsh conditions. The number of samples is no less than 100, with at least 50 voided and 50 solid samples. Constructing a balanced dataset ensures the generalization ability of the trained model. The large sample size covers the variability of different void sizes (e.g., from 10mm×10mm to 10mm×20mm) and steel plate thickness, effectively solving the overfitting problem caused by insufficient samples in traditional methods and improving the model's sensitivity and robustness to small voids.

[0039] Step S120: Extract feature information from each frame of the preprocessed audio signal. The feature information includes at least two of the following: Mel frequency cepstral coefficients (MFCC), root mean square energy (RMS), and the statistical mean and standard deviation of the spectral centroid.

[0040] The audio signal to be recognized undergoes preprocessing, including bandpass filtering, noise reduction, framing, and trimming. The bandpass filtering bandwidth is 0kHz–20kHz. Framing is implemented using a Hamming window with a window length of 20ms–40ms and a step size of 10ms–20ms. Trimming duration is 0.05s–3s. When the audio duration is insufficient, it is automatically padded with zeros to the specified length. The bandpass filtering bandwidth of 0kHz–20kHz filters out low-frequency environmental noise and high-frequency interference, focusing on the effective frequency band related to noise reduction. This solves the problem of low signal-to-noise ratio in complex environments and reduces the interference of steel plate thickness and internal structure on audio features, thereby improving the purity of feature extraction and recognition accuracy.

[0041] The total number of frames X for a single sample after framing is calculated using the formula:

[0042]

[0043] Where T is the total duration of the audio after trimming (in ms), L is the window length (in ms), and S is the step size (in ms). The use of a Hamming window reduces spectral leakage, while the window length and step size optimize the granularity of time-frequency analysis. For non-steady-state audio signals (such as the attenuation process of a knock), frame-level processing captures instantaneous feature changes, solving the problem of low sensitivity of traditional whole-segment signal analysis to short-term clipping events and improving the resolution of small-sized clippings. The formula for calculating the total number of frames enables accurate calculation of the frame count, ensuring frame-level consistency in feature extraction. To address the non-uniformity of complex audio, the model input is optimized by quantizing the frame count, avoiding recognition errors caused by frame count fluctuations and improving the computational efficiency and robustness of clipping determination. Figure 2 The process of frame-level despace classification and determination is demonstrated, and its technical effect is further visualized through the formula, which solves the problem of dispersion of small-sized despace in the time domain.

[0044] By setting the trimming duration to 0.05s to 3s and using zero padding when it is insufficient, the feature dimension deviation caused by inconsistent signal duration is avoided through standardizing the audio length. Combined with zero padding, the integrity of frame-level processing is ensured, the problem of missing signal segments caused by environmental interference is avoided, and the stability and applicability under varying operating conditions are enhanced.

[0045] The feature information includes at least two of the following: Mel-frequency cepstral coefficients (MFCC), root-mean-square energy (RMS), and the statistical mean and standard deviation of the spectral centroid. Specifically, at least 10 dimensions of Mel-frequency cepstral coefficients are extracted from each frame of the audio signal. Through multi-dimensional feature fusion (such as MFCC capturing the spectral envelope, RMS reflecting energy distribution, and the spectral centroid indicating the frequency centroid), the acoustic characteristics of de-volumetric features are comprehensively characterized. To address the issue of weak de-volumetric features in small-sized samples, high-dimensional MFCC enhances the capture of spectral details, overcomes the deficiency of single features in terms of sensitivity to de-volumetric features, and improves the resolution and accuracy of the recognition model.

[0046] Step S130: Input the extracted frame feature information into the void removal defect recognition model to obtain the void removal confidence of each output frame. The void removal defect recognition model is a classification model based on support vector machine and trained using an labeled audio sample training set.

[0047] The labeled audio samples were divided into training, validation, and test sets according to a preset ratio. Support Vector Machines (SVMs) were then used to model and train the extracted audio features to construct a model for identifying gaps in audio input. Specifically, the labeled audio sample training set was divided into training, validation, and test sets according to a preset ratio, with the training set accounting for 60%–80%, the validation set for 10%–20%, and the test set for 10%–20%. This proportionalized dataset optimized the model training and validation process. The validation set was used for parameter tuning, and the test set was used for evaluation, addressing the issues of overfitting and weak generalization ability. This ensured the model's stable performance in complex environments, thereby improving the accuracy and reliability of gap identification.

[0048] Support Vector Machines (SVMs) use one of the following kernel functions: Sigmoid, multinomial, or Gaussian. Adaptive kernel function selection addresses nonlinear classification problems. The complex distribution of missing features in the feature space enhances the model's learning ability, solves the problem of blurred small-sized missing boundaries in traditional linear methods, and improves recognition accuracy and robustness.

[0049] Step S140: In response to the fact that the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is determined to be a de-empty frame; otherwise, it is a non-de-empty frame.

[0050] The confidence threshold ranges from 0.3 to 0.7, and an adjustable threshold is used to dynamically determine the confidence level of frame-level gap removal. This addresses the risk of misjudgment caused by environmental interference. By statistically analyzing most frames to output the final result, the method's adaptability and robustness in noisy environments are improved, ensuring the stability of gap removal recognition for small sizes.

[0051] Step S150: Statistically analyze the classification results of all frames, and use the judgment results of the majority of frames as the final defect identification result of the audio signal to be identified.

[0052] By setting a judgment threshold and making statistical decisions on the judgment results of all frames, the conclusion of the majority of frames is used as the final output, which enhances the robustness to local signal fluctuations and interference and avoids misjudgment. Thus, it achieves efficient, accurate and stable identification of void defects in steel plate concrete under complex working conditions, and overcomes the limitations of traditional wave propagation theory-based methods in thick steel plates and complex internal structure scenarios.

[0053] Furthermore, the software system deploying the void defect identification model on a desktop computing terminal allows users to import audio signal files or acquire signals in real time via an audio interface as input, automatically outputting void identification results. Through systematic deployment, real-time on-site identification is achieved. This solves the problems of traditional methods relying on human experience and slow response times. Modular design (audio acquisition, preprocessing, feature extraction, etc.) improves ease of operation and engineering applicability. Figure 3 The system interface was showcased, and its technical advantage lies in the visual output of the de-voiding results, which addresses the need for rapid troubleshooting in scenarios such as nuclear power engineering.

[0054] The method for identifying void defects in steel-concrete composite structures in this embodiment effectively solves the technical problems of low sensitivity to small-sized void defects and significant interference from steel plate thickness and complex internal structures in current detection methods. Specifically, firstly, by acquiring audio signals collected under excitation conditions, effective raw data input is provided for identification. Next, by finely extracting Mel-frequency cepstral coefficients (MFCC), root-mean-square energy (RMS), and spectral centroid statistical features from each frame of the signal, subtle acoustic feature changes caused by void defects can be captured from multiple dimensions in the time-frequency domain. This significantly enhances the sensitivity to small-sized or localized void defects and overcomes the drawbacks of feature averaging and overwhelmed behavior when processing entire signal segments using traditional methods. Then, the extracted frame features are input into a pre-trained support vector machine classification model. Utilizing the model's powerful ability to handle high-dimensional features and nonlinear problems, the void confidence score for each frame is accurately output, achieving intelligent identification and classification of complex acoustic patterns. Furthermore, by comparing the confidence score of each frame with a preset threshold, frame-level fine-grained determination of the audio signal is achieved. Finally, based on the judgment results of all frames, the majority decision principle is adopted to obtain the overall recognition conclusion, which effectively suppresses misjudgments caused by local signal fluctuations or accidental interference. Thus, it achieves efficient, stable and accurate identification of void defects in steel plate concrete structures under complex working conditions, and improves the robustness and practicality of the method.

[0055] The present invention will be further described below with reference to specific embodiments.

[0056] like Figure 4The steel-concrete composite member is known to have dimensions of 700mm × 400mm × 400mm and a steel plate thickness of 20mm. Four void defects are identified, three of which have an area of ​​10mm × 10mm and one has an area of ​​10mm × 20mm. The identification of these void defects includes the following steps:

[0057] (1) A solid steel hammer with a spherical end was used as an excitation tool to strike the surface of the steel plate in the void area. Audio signals were collected using a microphone with a sampling rate of 44.1kHz. 20 audio signal samples were collected on the surface of the steel plate with a void size of 10mm×10mm and 40 audio signal samples were collected on the surface of the steel plate with a void size of 10mm×20mm. Each strike point was ensured to be unique, and a total of 100 void defect samples were collected. The surface of the steel plate in the dense area was struck 100 times, ensuring that each strike point was unique, and a total of 100 samples were collected in the dense area. The total number of audio signal samples was 200.

[0058] (2) Labels were assigned to the 200 collected audio signal samples. Empty samples were labeled with 0 and dense samples were labeled with 1, thus constructing a labeled sample dataset.

[0059] (3) The sample dataset was preprocessed. A bandpass filter was used to limit the signal frequency to the range of 0Hz to 6000Hz. Environmental noise was eliminated by wavelet denoising. The duration of each sample audio was cut to 0.1s. Each audio segment was divided into frames with a window length of 25ms and a step length of 10ms. The silent parts at the beginning and end were removed and the segments were cut into effective signal segments. The total number of frames for a single sample was 8.

[0060] (4) Extract multiple time-frequency domain feature information from each frame of audio, including extracting the Mel frequency cepstral coefficients (MFCC) of each frame and taking the first 13 dimensions as features; extracting the root mean square energy (RMS) value of each frame of audio; extracting the spectral centroid of each frame of audio; and statistically analyzing all frame features of each audio sample, calculating the mean and standard deviation of its MFCC, RMS, and spectral centroid, and using them as the global feature vector of the sample.

[0061] (5) The labeled feature dataset is divided proportionally into a training set (70%), a validation set (15%), and a test set (15%), with 140, 30, and 30 features respectively. A Support Vector Machine (SVM) classifier is used to train the feature vectors of each frame in the training set. The Sigmoid function is chosen as the SVM kernel function to construct a recognition model that can output frame-level "empty" confidence. After parameter tuning on the validation set, the model outputs the probability value of each frame belonging to the "empty" category, which serves as the "empty" confidence.

[0062] (6) After the model training is completed, a fixed confidence threshold of 0.6 is set. 30 test set samples are input for testing. The test results are shown in Table 1, with an accuracy of 0.93.

[0063]

[0064]

[0065] Table 1

[0066] Taking test sample 5 as an example, the process of determining it as a "empty frame" is explained: Sample 5 has a total of 8 frames. The model calculates the empty frame confidence score for each frame, which are 0.78, 0.89, 0.74, 0.88, 0.86, 0.83, 0.55, and 0.79, respectively. Frames with a score higher than 0.6 are determined to be "empty frames", and those without are "solid frames". A total of 7 frames are determined to be "empty frames" and 1 frame is determined to be "solid frames". Therefore, the empty frame classification label of the majority of frames is used as the final recognition result of this audio segment.

[0067] (7) Deploy the trained recognition model on the desktop application. The system supports real-time access to audio data collected at the recognition site through the audio interface, automatically completes signal processing, feature extraction, model prediction and result display, outputs the void determination result of the current recognition point of the steel plate concrete structure, and also supports uploading existing audio files for analysis and automatically outputting the determination result. It can be used in actual engineering recognition.

[0068] Because audio signals in de-void regions often exhibit regular differences in spectrum and energy distribution, traditional audio signal processing methods extract feature information from the entire audio segment for analysis. However, actual steel-concrete structures may be in complex environments, such as nuclear power plants and immersed tunnels, where significant signal interference may occur during audio signal collection. Extracting feature information from the entire audio segment may prevent the model from learning the features effectively, thus hindering accurate judgment. This invention addresses this by dividing the signal into a certain number of frames and extracting key acoustic features from each frame. These features are then used as input for model training, allowing the model to learn the feature information of each frame and calculate the probability of a frame being de-void (its de-void confidence level). Thresholds can then be set and adjusted based on actual conditions to determine whether a frame is classified as de-void. Finally, the classification results for all frames at that measurement point are statistically analyzed, and the majority judgment is used as the final identification result. This improves adaptability to unstable signals, effectively filters out interference, and ensures more stable and accurate identification results.

[0069] Meanwhile, the "void-free confidence" introduced in this invention refers to the reliability of the model's judgment that a sample has a void-free defect. It is a value between 0 and 1. The closer the void-free confidence is to 1, the more confident the model is that the current input sample has a void-free defect; the closer it is to 0, the more the model tends to believe the current input sample is solid. Therefore, it can be adjusted by setting a threshold. A threshold can be set according to the actual project situation. When the void-free confidence is greater than this threshold, it is judged as void-free; when it is less than the threshold, it is judged as solid. By setting a threshold according to the actual situation, the model output results can be made more reliable, avoiding a simple classification of either 0 or 1.

[0070] This invention effectively identifies void and non-void areas in steel-concrete composite structures by extracting multiple feature information from audio signals at the frame level and constructing a classification model using support vector machines, significantly improving the resolution and accuracy of defect identification. A fixed threshold strategy is employed to classify frame-level confidence levels, and statistical analysis is used to determine the majority classification as the final output, enhancing the model's generalization ability and robustness under different environments. The void defect identification model can be deployed as a software module on a computer terminal, automatically outputting identification results by real-time acquisition of on-site audio, without relying on subjective human experience. It boasts significant advantages such as ease of operation, rapid response, and suitability for quick on-site troubleshooting in engineering projects.

[0071] The following describes the steel plate concrete structure void defect identification device provided by the present invention. The steel plate concrete structure void defect identification device described below can be referred to in correspondence with the steel plate concrete structure void defect identification method described above.

[0072] like Figure 5 As shown, in one embodiment, a device for identifying void defects in steel plate concrete structures includes a signal acquisition module 510, a feature information extraction module 520, a confidence acquisition module 530, an audio frame determination module 540, and an identification result output module 550.

[0073] The signal acquisition module 510 is used to acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions;

[0074] The feature information extraction module 520 is used to extract feature information from each frame of the preprocessed audio signal. The feature information includes at least two items from Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid.

[0075] The confidence acquisition module 530 is used to input the extracted frame feature information into the void removal defect recognition model to obtain the void removal confidence of each frame output. The void removal defect recognition model is a classification model based on support vector machine and trained using an labeled audio sample training set.

[0076] The audio frame determination module 540 is used to determine that the corresponding frame is a frame with empty space if the confidence level of the empty space corresponding to each frame of audio is higher than the confidence threshold, otherwise it is a frame without empty space.

[0077] The recognition result output module 550 is used to statistically analyze the classification results of all frames, and the judgment results of the majority of frames are used as the final de-voiding defect recognition result of the audio signal to be recognized.

[0078] Figure 6 This example illustrates a schematic diagram of the physical structure of an electronic device, which can be a smart terminal. Its internal structure diagram can be as follows: Figure 6 As shown. The electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for identifying void defects in steel-concrete composite structures, the method including:

[0079] Acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions;

[0080] Feature information is extracted from each frame of the preprocessed audio signal. The feature information includes at least two items from the following: Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid.

[0081] The extracted frame feature information is input into the void removal defect recognition model to obtain the void removal confidence of each output frame. The void removal defect recognition model is a classification model trained on a support vector machine using a labeled audio sample training set.

[0082] If the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is determined to be a de-empty frame; otherwise, it is a non-de-empty frame.

[0083] The classification results of all frames are statistically analyzed, and the judgment results of the majority of frames are used as the final defect identification result of the audio signal to be identified.

[0084] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the electronic device to which the present invention is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0085] On the other hand, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements a method for identifying void defects in steel-concrete composite structures, the method comprising:

[0086] Acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions;

[0087] Feature information is extracted from each frame of the preprocessed audio signal. The feature information includes at least two items from the following: Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid.

[0088] The extracted frame feature information is input into the void removal defect recognition model to obtain the void removal confidence of each output frame. The void removal defect recognition model is a classification model trained on a support vector machine using a labeled audio sample training set.

[0089] If the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is determined to be a de-empty frame; otherwise, it is a non-de-empty frame.

[0090] The classification results of all frames are statistically analyzed, and the judgment results of the majority of frames are used as the final defect identification result of the audio signal to be identified.

[0091] In another aspect, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, it implements a method for identifying void defects in steel-concrete composite structures, the method comprising:

[0092] Acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions;

[0093] Feature information is extracted from each frame of the preprocessed audio signal. The feature information includes at least two items from the following: Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid.

[0094] The extracted frame feature information is input into the void removal defect recognition model to obtain the void removal confidence of each output frame. The void removal defect recognition model is a classification model trained on a support vector machine using a labeled audio sample training set.

[0095] If the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is determined to be a de-empty frame; otherwise, it is a non-de-empty frame.

[0096] The classification results of all frames are statistically analyzed, and the judgment results of the majority of frames are used as the final defect identification result of the audio signal to be identified.

[0097] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0098] By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for identifying void defects in steel-concrete composite structures, characterized in that, The method includes: Acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions; Feature information is extracted from each frame of the preprocessed audio signal. The feature information includes at least two items from the following: Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid. The extracted frame feature information is input into the void removal defect recognition model to obtain the void removal confidence of each output frame. The void removal defect recognition model is a classification model trained on a support vector machine using a labeled audio sample training set. If the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is determined to be a de-empty frame; otherwise, it is a non-de-empty frame. The classification results of all frames are statistically analyzed, and the judgment results of the majority of frames are used as the final defect identification result of the audio signal to be identified.

2. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The excitation condition is achieved by using a solid iron or steel hammer with a spherical end.

3. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The audio signal to be identified is acquired through a microphone or pickup, and the sampling rate is ≥44.1kHz.

4. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The preprocessing includes bandpass filtering, noise reduction, framing, and trimming operations. The bandpass filtering frequency band is 0kHz to 20kHz. The framing is implemented using a Hamming window with a window length of 20ms to 40ms and a step size of 10ms to 20ms. The trimming duration is 0.05s to 3s. When the audio duration is insufficient, it is automatically padded with zeros to the specified duration.

5. The method for identifying void defects in steel-concrete composite structures according to claim 4, characterized in that, The extracted feature information includes: Extract at least 10 dimensions of Mel frequency cepstral coefficients from each frame of the audio signal.

6. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The labeled audio sample training set is divided into a training set, a validation set, and a test set according to a preset ratio, with the training set accounting for 60% to 80%, the validation set accounting for 10% to 20%, and the test set accounting for 10% to 20%.

7. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The kernel function of the support vector machine is one of the following: the sigmoid function, the polynomial kernel function, and the Gaussian kernel function.

8. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The confidence threshold is set in the range of 0.3 to 0.

7.

9. The method for identifying void defects in steel-concrete composite structures according to claim 1, characterized in that, The method further includes: The software system that deploys the void defect identification model on a desktop computing terminal supports users to import audio signal files or collect signals in real time through an audio interface as input, so as to automatically output void identification results.

10. A device for identifying void defects in steel-concrete composite structures, characterized in that, The device includes: The signal acquisition module is used to acquire the audio signal to be identified from the steel plate concrete structure acquired under excitation conditions; The feature information extraction module is used to extract feature information from each frame of the preprocessed audio signal. The feature information includes at least two items from Mel frequency cepstral coefficients, root mean square energy, and the statistical mean and standard deviation of the spectral centroid. The confidence acquisition module is used to input the extracted frame feature information into the void removal defect recognition model to obtain the void removal confidence of each frame output. The void removal defect recognition model is a classification model based on support vector machine and trained using a labeled audio sample training set. The audio frame determination module is used to determine that if the confidence level of the de-empty frame corresponding to each audio frame is higher than the confidence threshold, the corresponding frame is a de-empty frame; otherwise, it is a non-de-empty frame. The recognition result output module is used to statistically analyze the classification results of all frames, and the judgment results of the majority of frames are used as the final denuclearization defect recognition result of the audio signal to be recognized.

Images