Resiliometer based on rebound-acoustics and strength detection method thereof

Abstract

The invention relates to a rebound-acoustics-based rebound apparatus and a strength detection method thereof.Firstly, a force sensor and an audio acquisition device are integrated in the rebound apparatus, when a rebound test is carried out, the force sensor measures the stress change before and after a spring is impacted, and then the rebound value reading is calculated by calculating the relationship between the spring elongation and the stress; after the test is finished, the magnetic attraction device automatically retracts the spring to ensure that the device can be used for the next test, then the audio acquisition device records a sound signal in each test process, the sound signal is preprocessed through a WaveNet algorithm to remove noise, then features are extracted from the collected sound signal, and the features are extracted from the collected sound signal. And combining the extracted sound features into a feature matrix, analyzing and screening out features highly related to the strength of the building material by using a correlation heat map, and finally inputting a rebound value and the screened sound features into an XGBoost regression model to predict the strength of the building material. The method supports off-line operation and is suitable for rapid detection and evaluation of various hard building materials.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of strength detection of building materials such as concrete, rock and blocks, and particularly relates to a rebound hammer based on rebound-acoustics and a strength detection method thereof. BACKGROUND

[0002] The strength of building materials is crucial for ensuring the safety and durability of building structures. Traditional strength detection mainly relies on destructive testing methods such as core drilling sampling. However, these methods are not only time-consuming and costly, but also destructive to the building, making them unsuitable for widespread application in actual buildings.

[0003] Rebound method and knocking method are widely used due to their simplicity. Traditional rebound hammers calculate the hardness value of materials by measuring the rebound distance after spring impact, but this method only provides a single index of rebound value, which is easily affected by environmental factors and human operation errors. Knocking method is affected by subjective factors of operators, and data processing is required after testing, limiting its application in field testing. SUMMARY

[0004] Therefore, the present application aims to provide a rebound hammer based on rebound-acoustics and a strength detection method thereof, which improves the accuracy and reliability of building material strength detection by combining rebound value and sound signals generated during the rebound process.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solution: a rebound hammer based on rebound-acoustics, comprising: a rebound hammer (1), the rebound hammer (1) is fixedly connected with a multifunctional accessory (2), the multifunctional accessory (2) contains an audio acquisition device (21), a force sensor (22), an electromagnetic relay (23), a spring (24), a baffle (25) and a microcontroller (26); the audio acquisition device (21), the force sensor (22), the electromagnetic relay (23), the spring (24), the baffle (25) and the microcontroller (26) are sequentially arranged; the audio acquisition device (21) is connected with the microcontroller (26), the force sensor (22) is connected with the electromagnetic relay (23), the electromagnetic relay (23) and the baffle (25) are connected through the spring (24), and the electromagnetic relay (23) is connected with the microcontroller (26).

[0006] The present application also provides a strength detection method of a rebound hammer based on rebound-acoustics, which uses the rebound hammer based on rebound-acoustics to carry out strength detection, including the following steps:

[0007] a) integrating a force sensor (22) and an audio acquisition device (21) inside the rebound hammer;

[0008] b) When the rebound test is performed, the force sensor (22) measures the force change before and after the spring (24) hits the stopper (25), and calculates the rebound value reading by calculating the relationship between the spring elongation and the force;

[0009] c) After the test is completed, the electromagnetic relay (23) automatically retracts the spring (24), ensuring that the device is ready for the next test;

[0010] d) The audio acquisition device (21) records the sound signal during each test;

[0011] e) The sound signal is preprocessed using the WaveNet algorithm to remove noise;

[0012] f) Key features are extracted from the noise-reduced audio signal;

[0013] g) Correlation heat map analysis is used to screen out features highly correlated with the strength of building materials, and the absolute value of the correlation coefficient is set to be greater than or equal to 0.5 as the screening threshold, and the features that meet the conditions are retained;

[0014] h) The rebound value and the screened audio features are input into the XGBoost regression model to predict the strength of the building materials;

[0015] i) The method microcontroller (26) supports offline operation mode, so that users can complete rapid detection and evaluation without network connection.

[0016] In a preferred embodiment: the force sensor (22) is used to measure the force change before and after the spring (24) hits the stopper (25), and the rebound value reading is calculated by analyzing the relationship between the spring elongation and the force, which is one of the features participating in the subsequent strength prediction; the maximum impact force F is represented as:

[0017]

[0018] where k s is the sensor sensitivity, and V is the voltage output; then based on Hooke's law:

[0019] F=kx+F0

[0020] where k is the spring stiffness and F0 is the spring initial parameter; the maximum compression amount x of the spring is calculated to obtain the rebound value reading.

[0021] In a preferred embodiment: the audio acquisition device (21) can record the sound signal during each test; the total frame number M:

[0022]

[0023] Where N is the total number of audio sampling points, L is the frame length, and S is the sampling speed; after applying the Hamming window w[n] to each frame of signal, the FFT is performed to generate the spectrum matrix, and the short-time Fourier transform STFT result X m [k'] is:

[0024]

[0025] Where x m [n] is the mth signal sequence, m is the frame index, k' is the frequency index; compressed to MxK mel spectrum through mel filter, j is the imaginary unit;

[0026] The conversion formula between mel frequency m' and hertz frequency f is:

[0027]

[0028] Where f is the hertz frequency, and m' is the mel frequency;

[0029] Let Hk[k] be the frequency response of the kth mel filter, and Sm[k"] be the spectrum of the mth frame. The mel spectrum coefficient calculation formula is:

[0030]

[0031] Where the kth mel spectrum coefficient Y m [k"] of the mth frame; the initial signal-to-noise ratio SNR calculation formula is as follows:

[0032]

[0033] Where P singal is the average power of the signal, and P noise is the average power of the noise.

[0034] In a preferred embodiment: the offline noise reduction algorithm is WaveNet; uses dilated causal convolution to capture the time dependence of the audio signal at different scales; for a one-dimensional signal x[n], the output y[n] of the dilated causal convolution is calculated as follows:

[0035]

[0036] Where K is the size of the convolution kernel; w[k] is the weight of the convolution kernel, k=0,1,…; d is the dilation rate, and n is the time index of the output signal; uses a gated activation unit to enhance the nonlinear representation capability of the model; the output h[n] for each time step n is calculated as follows:

[0037]

[0038] Where: z f[n] is the feature vector calculated by the hyperbolic tangent function tanh, z g [n] is the gating vector calculated by the Sigmoid function σ, used to control z f [n] which elements can pass through; ⊙ represents element-wise multiplication; w f [k] and w g [k] are the convolution kernel weights of the feature branch and the gating branch respectively; the mean square error of the benchmark signal SNR is used as the loss function:

[0039]

[0040] where N is the length of the audio signal, is the original audio signal, y clean [n] is the denoised audio signal; by minimizing this loss function, the clean audio signal is recovered from the noisy audio signal;

[0041] The key features are extracted from the denoised audio signal, the root mean square amplitude reflects the effective value of signal energy, which is positively correlated with the damping characteristics of the material, and the calculation formula is as follows:

[0042]

[0043] where A rms is the root mean square amplitude, x n is the signal value of the nth sampling point; the zero crossing rate is the number of sign changes of adjacent sampling points in the audio signal, and the calculation formula is as follows:

[0044]

[0045] where ZCR is the zero crossing rate of the audio signal, and u(x) is the sign judgment function; the short-time energy is the energy of the framed signal, and the calculation formula is as follows:

[0046]

[0047] The kurtosis is the degree of sharpness of the signal amplitude distribution, and the calculation formula is as follows:

[0048]

[0049] where K is the kurtosis value, μ is the signal mean, and σ is the signal standard deviation; the skewness describes the symmetry of the signal amplitude distribution, and the calculation formula is as follows:

[0050]

[0051] where S is the skewness; the formula for calculating the dominant frequency is as follows:

[0052]

[0053] where f is the fundamental frequency, k is the frequency index corresponding to the spectral peak, f s is the sampling frequency; a-b Hz energy ratio is the proportion of energy in the frequency interval [a, b] to the total energy, calculated as follows:

[0054]

[0055] where a, b are the endpoints of the frequency interval, X k is the frequency domain amplitude of the signal; MFCC is calculated based on the Mel spectrum, taking the logarithm of the Mel spectrum first, and then performing discrete cosine transform DCT to obtain the Mel frequency cepstrum coefficient:

[0056]

[0057] where MFCC i is the i-th Mel frequency cepstrum coefficient, Y[k] is the k-th frequency band energy of the Mel spectrum.

[0058] In a preferred embodiment: the correlation heat map analysis is used to screen features highly correlated with the strength of building materials, and the absolute value of the correlation coefficient is greater than or equal to a preset threshold as the screening condition, the preset threshold is configured to effectively reduce the feature dimension while ensuring the prediction accuracy, thereby improving the training efficiency and prediction speed of the XGBoost regression model; the calculation formula of the correlation coefficient r is as follows:

[0059]

[0060] where X i and Y i are the i-th observation value of the feature, and are the mean values of the two variables, and n is the number of features.

[0061] In a preferred embodiment, the feature extraction includes mean, variance, peak, zero crossing rate, main frequency and spectral energy, etc. time domain and frequency domain features.

[0062] In a preferred embodiment: the XGBoost regression model combines the screened sound features and the rebound value to predict the actual strength of the building materials.

[0063] In a preferred embodiment: the method uses a microcontroller (26) to support data transmission functions, allowing measured audio data, rebound values and final strength prediction results to be transmitted to other devices or systems for further analysis and report generation.

[0064] Compared with the prior art, the present application has the following beneficial effects:

[0065] (1) The force sensor and the audio acquisition device are integrated in the traditional rebound hammer, so that the rebound value and the audio signal of the test can be automatically collected;

[0066] (2) The acoustic parameters are introduced for comprehensive analysis, so that the problem that the single rebound value cannot accurately reflect the strength of the building material is effectively overcome;

[0067] (3) The rebound hammer is set to have constant work each time, so that the influence of the subjectivity of the operator in the knocking process on the detection result is reduced, and the detection precision is improved;

[0068] (4) The rebound value and the sound signal are processed by combining the noise reduction algorithm and the machine learning model, so that the accuracy of the detection result is improved;

[0069] (5) All necessary calculation and data processing steps are completed on the local device, without the need of retraining the model, so that the application requirements of various field environments are met. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 is the total flow chart of the method implemented in the embodiment of the present application.

[0071] Figure 2 is a schematic diagram of the rebound hammer in the embodiment.

[0072] Figure 3 is a schematic diagram of the force sensor and the sound acquisition device in the embodiment.

[0073] Figure 4 is a schematic diagram of the audio signal before and after noise reduction in the embodiment.

[0074] Figure 5 is a schematic diagram of the audio feature extraction in the embodiment.

[0075] Figure 6 is a schematic diagram of the rebound value and the audio feature extraction in the embodiment.

[0076] Figures 2-3 is a schematic diagram of the rebound value and the audio feature extraction in the embodiment. DETAILED DESCRIPTION

[0077] The present application will be further described below in conjunction with the drawings and embodiments.

[0078] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0079] It is to be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application; as used in this description, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0080] Reference Figures 1-6 A rebound-acoustic-based rebound hammer strength detection method uses a rebound-acoustic-based rebound hammer, which includes a rebound hammer 1 fixedly connected with a multifunctional accessory 2, the multifunctional accessory 2 containing an audio acquisition device 21, a force sensor 22, an electromagnetic relay 23, a spring 24, a baffle 25, and a microcontroller 26; the audio acquisition device 21, the force sensor 22, the electromagnetic relay 23, the spring 24, the baffle 25, and the microcontroller 26 are sequentially arranged; the audio acquisition device 21 is connected with the microcontroller 26, the force sensor 22 is connected with the electromagnetic relay 23, the electromagnetic relay 23 and the baffle 25 are connected through the spring 24, and the electromagnetic relay 23 is connected with the microcontroller 26.

[0081] The strength detection method includes the following steps:

[0082] S1, the rebound-acoustic-based rebound hammer and the strength detection method provided by the embodiment include a rebound hammer 1 fixedly connected with a multifunctional accessory 2, the multifunctional accessory 2 containing an audio acquisition device 21, a force sensor 22, an electromagnetic relay 23, a spring 24, a baffle 25, and a microcontroller 26. The audio acquisition device 21 can be a microphone.

[0083] Before the test starts, the rebound hammer needs to be systematically checked for hardware and function verification to ensure that the equipment state meets the detection standard. The rebound hammer is as shown in Figure 2 The multifunctional accessory is as shown in Figure 3 Firstly, check whether the internal microphone 21, the force sensor 22, the electromagnetic relay 23, the spring 24, the baffle 25, and the microcontroller 26 are dislocated, damaged, or have other abnormal conditions. Test the elastic deformation of the spring 24 by manually triggering the extension and retraction of the spring 24 to ensure that the elastic deformation is uniform, and observe whether the reciprocating motion is smooth, stuck, or has abnormal sound. Secondly, use a multimeter to test the continuity of the microcontroller 26 circuit to ensure that the hardware components are fault-free. Use the rebound hammer to vertically adhere to the standard anvil, press down the rebounding rod at a constant force and speed, and record each rebound value. If the average value deviates from the standard value of the anvil, the internal components need to be adjusted until the calibration is qualified.

[0084] In this embodiment, the rebound hammer is vertically attached to the surface of the test block to strike at a 90° vertical angle; then enter S2.

[0085] S2, the force sensor 22 can capture the force-time dynamic curve of the spring hitting the material at the moment. The force sensor 22 adopts an integrated design, which is fixed on the multifunctional accessory to reduce mechanical hysteresis error. The collection frequency is set to 10 kHz, which can meet the sampling requirements of high-frequency signals according to the Nyquist sampling theorem, and ensure that the signal has no aliasing distortion. The original force signal is first filtered through a 5th order Butterworth low-pass filter to filter out environmental noise and high-frequency interference; then the sliding window technology is used for mean filtering to smooth the signal burr. The system automatically identifies the peak point of the force curve, calculates the maximum impact force (Fmax) and rebound displacement and other key indicators. The spring displacement length is calculated by Hooke's law to obtain the rebound value reading. Then enter S3.

[0086] S3, the audio acquisition device 21 adopts a stereo microphone, which is fixed on the tail cover part of the rebound hammer through a magnetic support to reduce air propagation delay error. The original audio signal is transmitted to the embedded processor, which uses WaveNet algorithm for noise reduction processing. First, a neural network model containing multiple layers of dilated causal convolution is constructed, the input is the mel spectrum of the original audio, and the noise-reduced reference signal is generated through adversarial training. The specific processing process includes: 1) frame processing; 2) short-time Fourier transform to generate a frequency spectrum matrix; 3) compressed to multiple mel frequency bands through a mel filter bank; 4) predict the noise component through the WaveNet model and subtract it, and evaluate it by calculating the root mean square error between the reference signal and the original signal. If the error is large, the audio needs to be reacquired. In addition, in order to prevent structural noise interference, testing in strong airflow environment is prohibited; audio acquisition and force signal acquisition need to be time-aligned through a synchronous trigger module to ensure the accuracy of subsequent feature correlation analysis, and the measured and noise-reduced sound signal is as shown in Figure 4 ; then enter S4.

[0087] S4, time domain features are extracted from the reference audio signal: peak amplitude, root mean square amplitude, zero crossing rate, short-time energy, kurtosis, skewness; frequency domain features: main frequency, frequency band energy distribution, mel frequency cepstral coefficient, spectral entropy. After feature extraction, correlation heat map analysis is used, and the absolute value of the correlation coefficient greater than or equal to 0.5 is selected as the feature that meets the condition, and the screening process is as shown in Figure 5 ; then enter S5.

[0088] S5, the filtered audio features and rebound values are combined into an input vector, which is input into the XGBoost regression model for training. The root mean square error is used as the loss function. Before training, the data set needs to be divided into training set and test set in proportion, and the hyperparameters are optimized through cross-validation. When predicting, the model output value needs to be converted to the actual strength value through the material strength conversion formula, such asFigure 6 The rebound hammer is shown.

[0089] Specifically, before the test starts, the internal components of the rebound hammer are checked, the rebound hammer is calibrated using a standard steel anvil, and then the hammer is struck perpendicularly to the object under test.

[0090] Specifically, the force signal is obtained using the force sensor, the data collected by the force sensor 22 is analyzed, the specific elongation of the spring is calculated, and the rebound value reading is obtained. The maximum impact force F is represented as:

[0091]

[0092] where k s is the sensor sensitivity, and V is the voltage output. Then, based on Hooke's law:

[0093] F = kx + F0

[0094] where k is the spring stiffness, and F0 is the initial parameter of the spring. The maximum compression amount x of the spring is calculated, and the rebound value reading is obtained.

[0095] In this example, the audio acquisition device 21 synchronously records the audio signal of the impact on the surface of the material, the original audio feature is intercepted for 50 ms before and after the strike, the sampling speed is 44100 points, the signal is divided into short frames with a frame length of 512 points and an overlap of 256 points, and the total number of frames M is:

[0096]

[0097] where N is the total number of audio sampling points, L is the frame length, and S is the sampling speed. After applying the Hamming window w[n] to each frame of signal, FFT is performed to generate the frequency spectrum matrix, and the short-time Fourier transform (STFT) result X m [k] is:

[0098]

[0099] where m is the frame index, and k is the frequency index. The Mel filter is compressed to MxK Mel spectrum. The conversion formula between the Mel frequency m and the Hertz frequency f is:

[0100]

[0101] where f is the Hertz frequency, and m is the Mel frequency. Let Hk[k] be the frequency response of the kth Mel filter, and Sm[k] be the spectrum of the mth frame. The Mel spectrum coefficient calculation formula is:

[0102]

[0103] where the kth Mel spectrum coefficient Y m [k] of the mth frame. The signal-to-noise ratio SNR calculation formula is as follows:

[0104]

[0105] where P singal is the average power of the signal, P noise is the average power of the noise.

[0106] In this example, the WaveNet algorithm is used for preprocessing to remove noise and obtain the reference audio signal. The dilated causal convolution is used to capture the time-dependent relationship of the audio signal at different scales. For a one-dimensional signal x[n], the output y[n] of the dilated causal convolution is calculated as follows:

[0107]

[0108] where K is the size of the convolution kernel. w[k] is the weight of the convolution kernel, k = 0, 1, …, d is the dilation rate, and n is the time index of the output signal. A gated activation unit is used to enhance the nonlinear representation capability of the model. The output h[n] at each time step n is calculated as follows:

[0109]

[0110] where: z f [n] is the feature vector calculated by the hyperbolic tangent function tanh, z g [n] is the gating vector calculated by the Sigmoid function σ, which controls which elements in z f [n] can pass through. ⊙ represents element-wise multiplication. w f [k] and w g [k] are the convolution kernel weights of the feature branch and the gating branch, respectively. The mean square error of the reference signal SNR is used as the loss function:

[0111]

[0112] where N is the length of the audio signal, is the original audio signal, y clean [n] is the denoised audio signal. By minimizing this loss function, the clean audio signal can be recovered from the noisy audio signal.

[0113] In this example, key features such as the root mean square amplitude and the zero-crossing rate are extracted from the denoised audio signal. The root mean square amplitude reflects the effective value of the signal energy and is positively correlated with the damping characteristics of the material, and is calculated as follows.

[0114]

[0115] where A rms is the root mean square amplitude, xn is the signal value of the nth sample point. The zero-crossing rate is the number of times of sign change between adjacent sample points in the audio signal, and the calculation formula is as follows.

[0116]

[0117] where ZCR is the zero-crossing rate of the audio signal, and u(x) is the sign judgment function. The short-time energy is the energy of the framed signal, and the calculation formula is as follows.

[0118]

[0119] The kurtosis is the degree of sharpness of the amplitude distribution of the signal, and the calculation formula is as follows.

[0120]

[0121] where K is the kurtosis value, μ is the mean value of the signal, and σ is the standard deviation of the signal. The skewness describes the symmetry of the amplitude distribution of the signal, and the calculation formula is as follows.

[0122]

[0123] where S is the skewness. The dominant frequency calculation formula is as follows.

[0124]

[0125] where f is the dominant frequency, k is the frequency index corresponding to the spectral peak, and f s is the sampling frequency. The a-b Hz energy ratio is the proportion of the energy in the frequency interval [a, b] to the total energy, and the calculation formula is as follows.

[0126]

[0127] where a and b are the end points of the frequency interval, and X k is the frequency domain amplitude of the signal. The MFCC is calculated based on the Mel spectrum, and the Mel spectrum is first taken logarithm, and then the discrete cosine transform (DCT) is performed to obtain the Mel frequency cepstral coefficient:

[0128]

[0129] where MFCC i is the ith Mel frequency cepstral coefficient, i takes 1, 2, 3, etc., and Y[k] is the energy of the kth frequency band of the Mel spectrum.

[0130] In this example, the features highly correlated with the strength of building materials are screened out using correlation heat map analysis, the key audio features are retained, and the calculation complexity is reduced, and the model training efficiency is improved. The calculation formula of the correlation coefficient r is as follows:

[0131]

[0132] where X i and Y i are the i-th observation of the feature, and are the mean of the two variables, n is the number of features.

[0133] In this example, the absolute value of the correlation coefficient is set to be greater than or equal to 0.5 as the screening threshold, and the features that meet the condition are retained. The weight of the retained features is calculated, and the final feature values are [0.35, -0.18, 0.27, 0.41, -0.23].

[0134] In this example, the screened audio features and the rebound value are used as input, and an XGBoost regression model is applied for comprehensive analysis. The prediction accuracy is improved by stacking multiple decision trees, and the built-in regularization term prevents overfitting. The learning rate is 0.01, the CART tree depth is 3, and overfitting is avoided by avoiding too deep. The objective function is minimized The calculation formula is:

[0135]

[0136] where is the loss function, Ω(f k ) is the regularization term of the k-th tree, T is the number of leaf nodes of the tree, w j is the prediction value of the j-th leaf node, and γ and λ are regularization hyperparameters.

[0137] In this example, the training set is 100 groups of stone test data (rebound value 28-42, corresponding to compressive strength 90-100 MPa), and the test set is 20 groups of data.

[0138] In this example, the XGBoost model input vector [0.35, -0.18, 0.27, 0.41, -0.23, 38] outputs the actual strength prediction value of the building material (stone) as 95.2 MPa.

[0139] The embodiments of the present application are only examples, and the present application is not limited to the above-described embodiments. The method is suitable for strength detection of hard materials such as concrete, rock and block. Those skilled in the art can make various improvements and modifications to the present application according to actual needs without departing from the purpose of the present application.

Claims

1. A rebound hammer based on rebound-acoustics, characterized in that, include: A rebounder (1) is fixedly connected to a multi-functional accessory (2). The multi-functional accessory (2) contains an audio acquisition device (21), a force sensor (22), an electromagnetic relay (23), a spring (24), a baffle (25), and a microcontroller (26). The audio acquisition device (21), force sensor (22), electromagnetic relay (23), spring (24), baffle (25), and microcontroller (26) are arranged in sequence. The audio acquisition device (21) is connected to the microcontroller (26), the force sensor (22) is connected to the electromagnetic relay (23), the electromagnetic relay (23) is connected to the baffle (25) through the spring (24), and the electromagnetic relay (23) is connected to the microcontroller (26).

2. A method for strength testing using a rebound hammer based on rebound-acoustics, characterized in that... Using the rebound hammer based on rebound-acoustics as described in claim 1, strength testing is performed, including the following steps: a) Integrate a force sensor (22) and an audio acquisition device (21) inside the rebound hammer; b) When performing the rebound test, the force sensor (22) measures the change in force before and after the spring (24) hits the baffle (25), and calculates the rebound value reading by calculating the relationship between the spring elongation and the force. c) After the test is completed, the electromagnetic relay (23) automatically retracts the spring (24) to ensure that the equipment can be used for the next test; d) The audio acquisition device (21) records the sound signal during each test; e) The sound signal is preprocessed using the WaveNet algorithm to remove noise; f) Extract key features from the noise-reduced audio signal; g) Use correlation heatmap analysis to screen out features that are highly correlated with the strength of building materials, set the absolute value of the correlation coefficient to be greater than or equal to 0.5 as the screening threshold, and retain features that meet the conditions. h) Input the rebound value and the filtered audio features into the XGBoost regression model to predict the strength of building materials; i) The microcontroller (26) of the method supports offline operation mode, which allows users to complete rapid detection and evaluation without network connection.

3. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that: The force sensor (22) is used to measure the change in force before and after the spring (24) impacts the baffle (25), and to calculate the rebound value reading by analyzing the relationship between the spring elongation and the force. This reading is used as one of the features in subsequent strength prediction; the maximum impact force F is expressed as: Where k s It refers to the sensor sensitivity, where V is the voltage output; then, based on Hooke's Law: F = kx + F0 Where k is the spring stiffness and F0 is the initial spring parameter; calculate the maximum spring compression x and obtain the rebound value reading.

4. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that: The audio acquisition device (21) is capable of recording sound signals during each test; the total number of frames M: Where N is the total number of audio sampling points, L is the frame length, and S is the sampling rate; After applying a Hamming window w[n] to each frame of the signal, an FFT is performed to generate a spectrum matrix. The Short Time Fourier Transform (STFT) result X is then obtained. m [k'] is: Where, x m [n] is the m-th signal sequence, k' is the frequency index; it is compressed to an M×K Mel spectrum through a Mel filter, and j is the imaginary unit; The conversion formula between Mel frequency m' and Hertz frequency f is: Where f is the Hertz frequency and m' is the Mel frequency; Let Hk[k] be the frequency response of the k”th Mel filter, and Sm[k”] be the spectrum of the mth frame. The formula for calculating the Mel spectrum coefficients is: Where the k” Mel-spectral coefficient Y of the m-th frame m [k”]; The initial signal-to-noise ratio (SNR) is calculated using the following formula: Where P singal It is the average power of the signal, P noise It is the average power of the noise.

5. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that: The offline noise reduction algorithm is WaveNet; it uses dilated causal convolution to capture the temporal dependencies of audio signals at different scales; for a one-dimensional signal x[n], the output y[n] of the dilated causal convolution is calculated as follows: Where K is the size of the convolution kernel; w[k] is the weight of the convolution kernel, k = 0, 1, ...; d is the dilation rate; and n is the time index of the output signal. Gated activation units are used to enhance the nonlinear representation capability of the model. The output h[n] for each time step n is calculated as follows: Where: z f [n] is the eigenvector obtained by calculating the hyperbolic tangent function tanh, z g [n] is the gate vector calculated using the Sigmoid function σ, used to control z. f Which elements in [n] can be passed; ⊙ indicates element-wise multiplication; w f [k] and w g [k] represents the convolution kernel weights for the feature branch and the gated branch, respectively; the mean square error of the reference signal SNR is used as the loss function. Where N is the length of the audio signal. It is the original audio signal, y clean [n] is the denoised audio signal; by minimizing this loss function, a clean audio signal can be recovered from the noisy audio signal. Key features are extracted from the noise-reduced audio signal. The root mean square amplitude reflects the effective value of the signal energy and is positively correlated with the material damping characteristics. The calculation formula is as follows: Where A rms It is the root mean square amplitude, x n It is the signal value of the nth sampling point; the zero-crossing rate is the number of sign changes between adjacent sampling points in the audio signal, and the calculation formula is as follows: Where ZCR is the zero-crossing rate of the audio signal, u(x) is the sign determination function; short-time energy is the frame signal energy, calculated as follows: Kurtosis describes the degree of peaks in the amplitude distribution of a signal, and is calculated using the following formula: Where K is the kurtosis value, μ is the signal mean, and σ is the signal standard deviation; skewness describes the symmetry of the signal amplitude distribution, and is calculated using the following formula: Where S is the skewness; the formula for calculating the main frequency is as follows: Where f is the main frequency, and k is the frequency index corresponding to the spectral peak. s This refers to the sampling frequency; the a-bHz energy ratio is the proportion of energy within the frequency range [a,b] to the total energy, calculated using the following formula: Where a and b are the endpoints of the frequency interval, X k It represents the frequency domain amplitude of the signal; MFCC is calculated based on the Mel spectrum, first taking the logarithm of the Mel spectrum, then performing a Discrete Cosine Transform (DCT) to obtain the Mel frequency cepstral coefficients. Among them MFCC i Y[k] is the cepstral coefficient of the i-th Mel frequency, and Y[k] is the energy of the k-th band of the Mel spectrum.

6. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that: Correlation heatmap analysis is used to screen features highly correlated with building material strength, with the absolute value of the correlation coefficient being greater than or equal to a preset threshold as the screening criterion; the formula for calculating the correlation coefficient r is as follows: Where X i and Y i These are the i-th observations of the feature. and These are the means of the two variables, and n is the number of features.

7. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that, The feature extraction includes time-domain and frequency-domain features such as mean, variance, peak value, zero-crossing rate, dominant frequency, and spectral energy.

8. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that: The XGBoost regression model combines the filtered sound features and rebound values ​​to predict the actual strength of building materials.

9. The strength testing method of a rebound hammer based on rebound-acoustics according to claim 2, characterized in that: The method uses a microcontroller (26) to support data transmission, allowing the measured audio data, rebound value and final intensity prediction results to be transmitted to other devices or systems for further analysis and report generation.

Images