Field detection method and system for compaction quality of backfill soil foundation

Abstract

The invention discloses a backfill soil foundation compaction quality on-site detection method and system. The method comprises the steps that instantaneous acceleration signals of a vibrating wheel in the soil compaction construction process are collected in real time through a three-axis acceleration sensor; the acceleration signal comprises a three-axis acceleration component; recording the number of construction times through construction equipment, collecting position information of a construction area by using a GPS (Global Positioning System), and synchronously binding the acceleration signal with the number of construction times and the position information through timestamps to form a data set corresponding to the position of the construction area; performing spectral analysis on the acceleration signal in the data set, and extracting a dominant frequency amplitude and a first harmonic amplitude to generate a filtered acceleration signal; the filtered acceleration signals are input into a training single-hidden-layer BP neural network, and the soil dry density is calculated in combination with the number of construction times; based on the ratio of the calculated soil dry density to the standard maximum dry density, the soil compactness in the construction area is determined in combination with the position information, and a construction area compactness distribution diagram is generated through the upper computer to optimize the construction strategy.

Description

Technical Field

[0001] The present invention belongs to the field of civil engineering and intelligent detection technology, and more specifically, relates to an on-site detection method and system for backfill foundation compaction quality. Background Art

[0002] As the urbanization level in China continues to increase, the process of urban construction is moving forward in great strides, and more and more substation projects need to be newly built. Backfill foundation is the main foundation form for newly built substations. The compaction quality of the backfill foundation is crucial to the structural service performance of the newly built substation. Good compaction can effectively avoid uneven settlement of the foundation and protect the service safety of the upper substation and the underground pipeline corridor. Due to problems such as uneven fill thickness, loose fill, and high compressibility, the backfill foundation is difficult to meet the requirements of strength, deformation, and stability in the project. Appropriate treatment is required to eliminate or reduce the risks brought by poor backfill foundation and ensure the safety of the upper structure. However, the effect of foundation treatment must be evaluated through necessary detection methods. Therefore, detection work is an important task throughout foundation treatment.

[0003] The compaction quality of backfill soil is often affected by the following factors.

[0004] First, the different sources of backfill will lead to differences in its composition. For example, some backfill may be composed of construction waste or waste, while others may be natural soil. These different components will have different effects on the density of the backfill.

[0005] Secondly, the time, method and thickness of the backfill are also important factors that lead to uneven properties of the backfill. Over time, the backfill will gradually settle and rot, which will affect its density. Different backfill methods and thicknesses will result in different densities of the backfill.

[0006] Thirdly, the environment in which the project is located will also have an important impact on the properties of the backfill soil. Environmental factors include soil moisture, temperature, climate, etc. For example, in a humid environment, the density of the backfill soil may be reduced due to the influence of moisture.

[0007] Since existing technologies all adopt fixed and programmed detection methods, it is difficult to obtain accurate results when dealing with the above-mentioned influencing factors. Summary of the invention

[0008] In order to solve the problems caused by factors such as the source of backfill materials, filling time, filling method, filling thickness, and engineering environment in evaluating the density of backfill soil, this patent proposes a real-time estimation method for model parameters based on a neural network, so that the compaction degree can be accurately estimated in real time through real-time acceleration measurement data using the acceleration and compaction degree relationship model.

[0009] The present invention adopts the following technical solution.

[0010] A first aspect of the present invention provides a method for on-site detection of compaction quality of backfill foundation, comprising the following steps:

[0011] The instantaneous acceleration signal of the vibrating wheel during the soil compaction construction process is collected in real time by a three-axis acceleration sensor; the acceleration signal includes a three-axis acceleration component; the number of construction passes is recorded by the construction equipment, and the location information of the construction area is collected by using a GPS module, and the acceleration signal is synchronously bound with the number of construction passes and the location information through a timestamp to form a data set corresponding to the location of the construction area;

[0012] Perform spectrum analysis on the acceleration signal in the data set, extract the main frequency amplitude and the first harmonic amplitude to generate a filtered acceleration signal;

[0013] The filtered acceleration signal is input into the trained single hidden layer BP neural network, and the soil dry density is calculated based on the number of construction passes.

[0014] Based on the ratio of the calculated soil dry density to the standard maximum dry density, the soil compaction degree in the construction area is determined in combination with the location information, and a compaction distribution map of the construction area is generated by the host computer to optimize the construction strategy.

[0015] Preferably, the triaxial acceleration sensor is fixed to the vibration shaft of the vibratory roller by bolts to collect instantaneous acceleration signals of the vibration wheel under the effects of exciting force, deadweight and resistance during the construction process;

[0016] The sampling frequency of the acceleration signal is 1000 Hz. The collected three-axis signals include the vibration acceleration values ​​of the horizontal x-axis, vertical y-axis and lateral z-axis. The vertical direction signal is used to evaluate the impact of the vibration wheel on the soil, and the horizontal and lateral signals are used to monitor the slip and asymmetric force of the vibration wheel.

[0017] Preferably, the location information collected by the GPS module includes longitude, latitude and altitude;

[0018] The positioning error is corrected by real-time signal difference between the base station and the mobile station. The positioning error after correction is less than 1 meter.

[0019] Use a high-precision clock to synchronize the timestamps of the position information and the acceleration signal, where the time deviation of the timestamp synchronization does not exceed 1 millisecond;

[0020] The collected location information is divided into a plurality of continuous geographic coordinate points according to the construction path, and each coordinate point corresponds to a sampling time of the acceleration signal;

[0021] The location information is bound to the acceleration signal to generate a compaction detection dataset containing time series, geographic coordinate points and acceleration data, which is used to calculate the compaction quality distribution in the construction area.

[0022] Preferably, the performing spectrum analysis on the acceleration signal in the data set to extract the main frequency amplitude and the first harmonic amplitude to generate a filtered acceleration signal includes:

[0023] The acceleration signal is processed in segments according to a fixed number of sampling points. Each segment contains 1024 sampling points. The number of sampling points is determined according to the sampling frequency of the sensor and the target analysis frequency range.

[0024] Applying a Hanning window to each segment of the acceleration signal, performing a fast Fourier transform on the windowed signal segment, converting the time domain signal into a frequency domain signal, and generating an amplitude spectrum; the amplitude spectrum includes a correspondence between the frequency and the signal amplitude, and is used to identify the main frequency and the first harmonic component of the signal;

[0025] Extract the main frequency component and the first harmonic component of the acceleration signal from the spectrum; the main frequency component includes the frequency component with the largest amplitude in the spectrum, which corresponds to the excitation frequency of the vibration wheel; the first harmonic component includes the integer multiple frequency of the main frequency, and the amplitude of the first harmonic component is used to reflect the resistance characteristics of the soil, and the amplitude of the main frequency and the first harmonic component is used as the result of spectrum analysis;

[0026] The high-frequency components with amplitudes lower than 5% of the total energy in the spectrum are removed to complete the preliminary noise reduction processing of the spectrum signal.

[0027] Preferably, a Hanning window is applied to each acceleration signal as follows:

[0028]

[0029] Where w[n] represents the weight of the window function; N represents the number of sampling points of each signal segment, 1024.

[0030] Preferably, the single hidden layer BP neural network comprises:

[0031] The number of neurons in the input layer is 4, corresponding to the amplitude of the main frequency of the filtered acceleration signal, the amplitude of the first harmonic, the number of construction passes, and the position information;

[0032] The number of neurons in the hidden layer is 50, and the activation function uses the hyperbolic tangent function;

[0033] The output layer consists of one neuron, which is used to calculate the soil dry density corresponding to the construction point. The model weight initialization method uses the Xavier initialization method.

[0034] Preferably, the soil dry density is inversely calculated as follows:

[0035]

[0036] Where, e is the porosity; ρ is the soil density; ρ d is the soil compaction degree; α is the acceleration; ρ d is the soil dry density; A and B are model coefficients, which are determined by the output of the neural network.

[0037] Preferably, the generation of the compaction degree distribution map comprises:

[0038] The construction area is divided into grid cells with a side length of 0.5 meters, and each grid corresponds to a compaction data point;

[0039] The average value of compaction is calculated in each grid and color-coded according to the range of the average values;

[0040] The grids with a density below 90% are marked in red, indicating insufficient compaction areas; the grids with a density between 90% and 95% are marked in yellow, indicating critical compaction areas; the grids with a density above 95% are marked in green, indicating qualified compaction areas.

[0041] The color-coded compaction distribution map is stored in the host computer and can be updated in real time according to the construction path.

[0042] Preferably, after the compaction distribution map is generated, it is used to detect abnormal points in the construction area and generate construction parameter optimization suggestions, which include:

[0043] Parameter recommendations for increasing the number of construction passes or increasing the exciting force of the vibrating wheel in areas with low compaction;

[0044] Adjust the construction path to avoid repeated compaction of high compaction areas;

[0045] The optimized construction path and parameters are sent to the vibratory roller control system for automatic adjustment.

[0046] A second aspect of the present invention provides a backfill foundation compaction quality on-site detection system, comprising: a signal acquisition module, a signal processing module, a dry density calculation module and a compaction degree analysis module;

[0047] The signal acquisition module is used to collect the instantaneous acceleration signal of the vibration wheel and the location information of the construction area in real time during the soil compaction process. The acceleration signal is obtained by the three-axis acceleration sensor installed on the vibration shaft, and the location information is collected by the GPS module and synchronously bound with the acceleration signal through the timestamp to form an acceleration data set;

[0048] The signal processing module is used to perform segmentation processing, windowing, fast Fourier transform and spectrum analysis on the acceleration data set provided by the signal acquisition module, extract the main frequency amplitude and the first harmonic amplitude, and generate a filtered acceleration signal through bandpass filtering and noise reduction operations;

[0049] The dry density calculation module is used to receive the filtered acceleration signal generated by the signal processing module, and input it into the trained single hidden layer BP neural network model in combination with the number of construction passes and position information, and output the predicted value of soil dry density corresponding to the construction point through calculation;

[0050] The compaction analysis module is used to calculate the soil compaction at each point in the construction area based on the ratio of the soil dry density and the standard maximum dry density provided by the dry density calculation module, and generate a compaction distribution map according to the grid division, and display different compaction ranges through color coding.

[0051] Compared with the prior art, the beneficial effects of the present invention include at least: the present invention proposes a parameter real-time estimation method based on a neural network acceleration and compaction relationship model, which solves the influence of factors such as backfill source, backfill time, backfill method, backfill thickness, and engineering environment on the evaluation accuracy of backfill density. The prior art adopts a fixed and programmed approach, so it is difficult to obtain accurate results when dealing with the above-mentioned influencing factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is an error histogram of the accuracy test results of the neural network model after training provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0054] Embodiment 1 of the present invention provides a method for on-site detection of compaction quality of backfill foundation, comprising the following steps:

[0055] The instantaneous acceleration signal of the vibrating wheel during the soil compaction construction process is collected in real time by a three-axis acceleration sensor; the acceleration signal includes a three-axis acceleration component; the number of construction passes is recorded by the construction equipment, and the location information of the construction area is collected by using a GPS module, and the acceleration signal is synchronously bound with the number of construction passes and the location information through a timestamp to form a data set corresponding to the location of the construction area;

[0056] Specifically, the acceleration sensor is installed on the vibratory roller by bolting, and then the instantaneous acceleration of the vibrating wheel under the action of the exciting force, the gravity of the vibrating wheel itself, and the resistance force is measured. Considering that the vertically installed acceleration sensor can most effectively reflect the changes in the vibration acceleration of the roller, this is because the shock wave generated by the vibrating wheel is fan-shaped and spreads to the deep layer of the backfill soil with the contact surface between the vibrating wheel and the ground as the center, so the shock wave in the vertical direction first reaches the bottom layer of the backfill soil and is fed back to the vibrating wheel. However, current technologies mostly use single-axis accelerometers to be installed on the vibrating shaft with the sensitive axis vertically downward. Due to the influence of vibration and installation factors, the fact is that the sensitive axis of the single-axis accelerometer is often not completely vertically downward. In order to solve this problem, this patent uses a three-axis accelerometer, that is, the three-axis accelerometer is installed on the vibrating shaft by bolting. After using a three-axis accelerometer, the instantaneous acceleration a 瞬时 The calculation formula is:

[0057]

[0058] Among them, a x 、a y and a z The three axes are the measured acceleration of the three-axis accelerometer, and g is the acceleration due to gravity. The data is uploaded to the host computer for processing via Bluetooth or WiFi. The GPS signal is also uploaded together with the acceleration signal, so that the host computer can dynamically display the compaction degree of each point in the construction area.

[0059] Preferably, the triaxial acceleration sensor is fixed to the vibration shaft of the vibratory roller by bolts to collect instantaneous acceleration signals of the vibration wheel under the effects of exciting force, deadweight and resistance during the construction process;

[0060] The sampling frequency of the acceleration signal is 1000 Hz. The collected three-axis signals include the vibration acceleration values ​​of the horizontal x-axis, vertical y-axis and lateral z-axis. The vertical direction signal is used to evaluate the impact of the vibration wheel on the soil, and the horizontal and lateral signals are used to monitor the slip and asymmetric force of the vibration wheel.

[0061] Preferably, the location information collected by the GPS module includes longitude, latitude and altitude;

[0062] The positioning error is corrected by real-time signal difference between the base station and the mobile station. The positioning error after correction is less than 1 meter.

[0063] Use a high-precision clock to synchronize the timestamps of the position information and the acceleration signal, where the time deviation of the timestamp synchronization does not exceed 1 millisecond;

[0064] The collected location information is divided into a plurality of continuous geographic coordinate points according to the construction path, and each coordinate point corresponds to a sampling time of the acceleration signal;

[0065] The location information is bound to the acceleration signal to generate a compaction detection dataset containing time series, geographic coordinate points and acceleration data, which is used to calculate the compaction quality distribution in the construction area.

[0066] Perform spectrum analysis on the acceleration signal in the data set, extract the main frequency amplitude and the first harmonic amplitude to generate a filtered acceleration signal;

[0067] Specifically, the fast Fourier transform (FFT) technology is used to perform spectrum analysis on the acceleration signal to obtain the amplitude of each frequency component under different stress states. Finally, the signal within one cycle is intercepted and the amplitude of other order signals is removed to find the amplitude of the main frequency signal and the amplitude of the resistance signal. The main frequency signal can be considered as the amplitude of the exciting force signal, and the first harmonic component is the main influencing factor and can be considered as the amplitude of the resistance signal. The main frequency signal and the first harmonic signal are retained, and the harmonic signals above the first order are removed, and then the inverse Fourier transform is performed to obtain the filtered acceleration signal.

[0068] The main frequency of the exciting force applied by the vibratory roller is fixed and known. The plasticity of uncompacted soil is very large. During the compaction process, the soil will not affect the vibration frequency of the vibrating wheel (unless the soil gradation is not good, such as gravel mixed with gravel, or the sudden change of the road surface causes a sudden change in the acceleration signal). Most of the frequency components other than the main frequency are noise, such as the noise caused by the vibration of the transmission system, so multi-band analysis is meaningless. Therefore, as you said, the purpose of Fourier transform is to achieve bandpass filtering.

[0069] Preferably, the performing spectrum analysis on the acceleration signal in the data set to extract the main frequency amplitude and the first harmonic amplitude to generate a filtered acceleration signal includes:

[0070] The acceleration signal is processed in segments according to a fixed number of sampling points. Each segment contains 1024 sampling points. The number of sampling points is determined according to the sampling frequency of the sensor and the target analysis frequency range.

[0071] Applying a Hanning window to each segment of the acceleration signal, performing a fast Fourier transform on the windowed signal segment, converting the time domain signal into a frequency domain signal, and generating an amplitude spectrum; the amplitude spectrum includes a correspondence between the frequency and the signal amplitude, and is used to identify the main frequency and the first harmonic component of the signal;

[0072] Extract the main frequency component and the first harmonic component of the acceleration signal from the spectrum; the main frequency component includes the frequency component with the largest amplitude in the spectrum, which corresponds to the excitation frequency of the vibration wheel; the first harmonic component includes the integer multiple frequency of the main frequency, and the amplitude of the first harmonic component is used to reflect the resistance characteristics of the soil, and the amplitude of the main frequency and the first harmonic component is used as the result of spectrum analysis;

[0073] The high-frequency components with amplitudes lower than 5% of the total energy in the spectrum are removed to complete the preliminary noise reduction processing of the spectrum signal.

[0074] Preferably, a Hanning window is applied to each acceleration signal as follows:

[0075]

[0076] Where w[n] represents the weight of the window function; N represents the number of sampling points of each signal segment, 1024.

[0077] The filtered acceleration signal is input into the trained single hidden layer BP neural network, and the soil dry density is calculated based on the number of construction passes.

[0078] Preferably, the single hidden layer BP neural network comprises:

[0079] The number of neurons in the input layer is 4, corresponding to the amplitude of the main frequency of the filtered acceleration signal, the amplitude of the first harmonic, the number of construction passes, and the position information;

[0080] The number of neurons in the hidden layer is 50, and the activation function uses the hyperbolic tangent function;

[0081] The output layer consists of one neuron, which is used to calculate the soil dry density corresponding to the construction point. The model weight initialization method uses the Xavier initialization method.

[0082] Preferably, the soil dry density is inversely calculated as follows:

[0083]

[0084] Where, e is the porosity; ρ is the soil density; ρ d is the soil compaction degree; α is the acceleration; ρ d is the soil dry density; A and B are model coefficients, which are determined by the output of the neural network.

[0085] like Figure 1As shown, the error histogram of the accuracy test results of the neural network model after training provided in accordance with the embodiment of the present invention is used to intuitively evaluate the prediction accuracy of the neural network model in soil dry density calculation. Each bar in the histogram represents the distribution of the prediction error, the horizontal axis is the error range, and the vertical axis is the number of samples within the corresponding error range. By observing the concentration and distribution range of the error values ​​in the figure, the prediction accuracy and stability of the model can be judged, providing a reference for optimizing the neural network structure and input parameters.

[0086] Specifically, soil gradation, backfill thickness, number of passes, driving speed, vibration amplitude, and vibration frequency are direct factors affecting compaction, so these factors will inevitably change the model coefficients shown in formula (2). The engineering environment (temperature and humidity) will affect the moisture content of the soil, thereby affecting the compaction effect, and will inevitably affect the model coefficients shown in formula (2). In view of this, the input of the neural network is determined to be the quantized soil gradation, backfill thickness, engineering environment (temperature and humidity), number of passes, driving speed, vibration amplitude, and vibration frequency, and the output is the model coefficients A, B, C, and D. Among them, soil gradation and backfill thickness are artificial pre-input quantities, and the remaining input quantities are measured by temperature and humidity sensors (engineering environment), odometers (number of passes), encoders (speed), and vibration sensors (vibration amplitude, vibration frequency). The Universal Approximation Theorem shows that a neural network with one hidden layer and containing enough neurons can approximate any continuous function in a closed interval. Therefore, the neural network used in this patent is a single hidden layer BP network with 23 hidden layer neurons, and the "Fletcher-Reeves conjugate gradient" algorithm is selected as the model training algorithm. Before inputting into the neural network, the quantitative data such as soil gradation are normalized. For soil gradation, the gradation coefficient of a certain particle size can be used as normalized data. For input quantities such as backfill thickness, its value can be divided by the maximum possible value and used as normalized data. In order to obtain the actual soil compaction, after compaction once, multiple sampling points are uniformly and randomly selected in the construction area, and then the compaction of these points is obtained by methods such as sand filling method, and finally the average compaction of these sampling points is used as the compaction of the construction area.

[0087] Based on the ratio of the calculated soil dry density to the standard maximum dry density, the soil compaction degree in the construction area is determined in combination with the location information, and a compaction distribution map of the construction area is generated by the host computer to optimize the construction strategy.

[0088] Preferably, the generation of the compaction degree distribution map comprises:

[0089] The construction area is divided into grid cells with a side length of 0.5 meters, and each grid corresponds to a compaction data point;

[0090] The average value of compaction is calculated in each grid and color-coded according to the range of the average values;

[0091] The grids with a density below 90% are marked in red, indicating insufficient compaction areas; the grids with a density between 90% and 95% are marked in yellow, indicating critical compaction areas; the grids with a density above 95% are marked in green, indicating qualified compaction areas.

[0092] The color-coded compaction distribution map is stored in the host computer and can be updated in real time according to the construction path.

[0093] Preferably, after the compaction distribution map is generated, it is used to detect abnormal points in the construction area and generate construction parameter optimization suggestions, which include:

[0094] Parameter recommendations for increasing the number of construction passes or increasing the exciting force of the vibrating wheel in areas with low compaction;

[0095] Adjust the construction path to avoid repeated compaction of high compaction areas;

[0096] The optimized construction path and parameters are sent to the vibratory roller control system for automatic adjustment.

[0097] Specifically, the compaction degree of each point evenly distributed in the construction area is displayed in a centralized manner in combination with GPS data, so as to observe the soil compaction condition in the construction area in real time. If an abnormal compaction point is found, the cause of the abnormality is analyzed on site, such as whether the soil gradation or moisture content at the abnormal point is abnormal. If necessary, the soil can be compacted again directly, or the soil near the abnormal point can be replaced before compaction.

[0098] Example 2 of the present invention provides a backfill soil foundation compaction quality on-site detection system, including: a signal acquisition module, a signal processing module, a dry density calculation module and a compaction degree analysis module;

[0099] The signal acquisition module is used to collect the instantaneous acceleration signal of the vibration wheel and the location information of the construction area in real time during the soil compaction process. The acceleration signal is obtained by the three-axis acceleration sensor installed on the vibration shaft, and the location information is collected by the GPS module and synchronously bound with the acceleration signal through the timestamp to form an acceleration data set;

[0100] The signal processing module is used to perform segmentation processing, windowing, fast Fourier transform and spectrum analysis on the acceleration data set provided by the signal acquisition module, extract the main frequency amplitude and the first harmonic amplitude, and generate a filtered acceleration signal through bandpass filtering and noise reduction operations;

[0101] The dry density calculation module is used to receive the filtered acceleration signal generated by the signal processing module, and input it into the trained single hidden layer BP neural network model in combination with the number of construction passes and position information, and output the predicted value of soil dry density corresponding to the construction point through calculation;

[0102] The compaction analysis module is used to calculate the soil compaction at each point in the construction area based on the ratio of the soil dry density and the standard maximum dry density provided by the dry density calculation module, and generate a compaction distribution map according to the grid division, and display different compaction ranges through color coding.

[0103] The results after training are shown in Table 1, Table 2, Table 3 and Table 4:

[0104] Table 1 Weight matrix from input layer to hidden layer

[0105]

[0106] Table 2 Weight matrix from hidden layer to output layer

[0107]

[0108] Table 3 Bias vector of hidden layer

[0109]

[0110]

[0111] Table 4 Bias vector of output layer

[0112]

[0113] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for on-site detection of backfill soil foundation compaction quality, characterized in that: The following steps are involved: The instantaneous acceleration signal of the vibrating wheel during the soil compaction construction process is collected in real time by a three-axis acceleration sensor; the acceleration signal includes a three-axis acceleration component; the number of construction passes is recorded by the construction equipment, and the location information of the construction area is collected by using a GPS module, and the acceleration signal is synchronously bound with the number of construction passes and the location information through a timestamp to form a data set corresponding to the location of the construction area; Perform spectrum analysis on the acceleration signal in the data set, extract the main frequency amplitude and the first harmonic amplitude to generate a filtered acceleration signal; The filtered acceleration signal is input into the trained single hidden layer BP neural network, and the soil dry density is calculated based on the number of construction passes. Based on the ratio of the calculated soil dry density to the standard maximum dry density, the soil compaction degree in the construction area is determined in combination with the location information, and a compaction distribution map of the construction area is generated by the host computer to optimize the construction strategy.

2. A method for on-site detection of backfill foundation compaction quality according to claim 1, characterized in that: The triaxial acceleration sensor is fixed to the vibration shaft of the vibratory roller by bolts to collect instantaneous acceleration signals of the vibration wheel under the effects of excitation force, deadweight and resistance during construction. The sampling frequency of the acceleration signal is 1000 Hz. The collected three-axis signals include the vibration acceleration values ​​of the horizontal x-axis, vertical y-axis and lateral z-axis. The vertical direction signal is used to evaluate the impact of the vibration wheel on the soil, and the horizontal and lateral signals are used to monitor the slip and asymmetric force of the vibration wheel.

3. The method for on-site detection of backfill foundation compaction quality according to claim 1, characterized in that: The location information collected by the GPS module includes longitude, latitude and altitude; The positioning error is corrected by real-time signal difference between the base station and the mobile station. The positioning error after correction is less than 1 meter. Use a high-precision clock to synchronize the timestamps of the position information and the acceleration signal, where the time deviation of the timestamp synchronization does not exceed 1 millisecond; The collected location information is divided into a plurality of continuous geographic coordinate points according to the construction path, and each coordinate point corresponds to a sampling time of the acceleration signal; The location information is bound to the acceleration signal to generate a compaction detection dataset containing time series, geographic coordinate points and acceleration data, which is used to calculate the compaction quality distribution in the construction area.

4. The method for on-site detection of compaction quality of backfill soil foundation according to claim 1, characterized in that: The performing spectrum analysis on the acceleration signal in the data set to extract the main frequency amplitude and the first harmonic amplitude to generate a filtered acceleration signal includes: The acceleration signal is processed in segments according to a fixed number of sampling points. Each segment contains 1024 sampling points. The number of sampling points is determined according to the sampling frequency of the sensor and the target analysis frequency range. Applying a Hanning window to each segment of the acceleration signal, performing a fast Fourier transform on the windowed signal segment, converting the time domain signal into a frequency domain signal, and generating an amplitude spectrum; the amplitude spectrum includes a correspondence between the frequency and the signal amplitude, and is used to identify the main frequency and the first harmonic component of the signal; Extract the main frequency component and the first harmonic component of the acceleration signal from the spectrum; the main frequency component includes the frequency component with the largest amplitude in the spectrum, which corresponds to the excitation frequency of the vibration wheel; the first harmonic component includes the integer multiple frequency of the main frequency, and the amplitude of the first harmonic component is used to reflect the resistance characteristics of the soil, and the amplitude of the main frequency and the first harmonic component is used as the result of spectrum analysis; The high-frequency components with amplitudes lower than 5% of the total energy in the spectrum are removed to complete the preliminary noise reduction processing of the spectrum signal.

5. The method for on-site detection of compaction quality of backfill soil foundation according to claim 1, characterized in that: Apply the Hanning window to each acceleration signal as follows: Where w[n] represents the weight of the window function; N represents the number of sampling points of each signal segment, 1024.

6. The method for on-site detection of backfill foundation compaction quality according to claim 1, characterized in that: The single hidden layer BP neural network comprises: The number of neurons in the input layer is 4, corresponding to the amplitude of the main frequency of the filtered acceleration signal, the amplitude of the first harmonic, the number of construction passes, and the position information; The number of neurons in the hidden layer is 50, and the activation function uses the hyperbolic tangent function; The output layer consists of one neuron, which is used to calculate the soil dry density corresponding to the construction point. The model weight initialization method uses the Xavier initialization method.

7. The method for on-site detection of backfill foundation compaction quality according to claim 1, characterized in that: The inverse calculation of soil dry density is as follows: Where, e is the porosity; ρ is the soil density; ρ d is the soil compaction degree; α is the acceleration; ρ d is the soil dry density; A and B are model coefficients, which are determined by the output of the neural network.

8. The method for on-site detection of backfill foundation compaction quality according to claim 1, characterized in that: The generation of the compaction distribution map includes: The construction area is divided into grid cells with a side length of 0.5 meters, and each grid corresponds to a compaction data point; The average value of compaction is calculated in each grid and color-coded according to the range of the average values; The grids with a density below 90% are marked in red, indicating insufficient compaction areas; the grids with a density between 90% and 95% are marked in yellow, indicating critical compaction areas; the grids with a density above 95% are marked in green, indicating qualified compaction areas. The color-coded compaction distribution map is stored in the host computer and can be updated in real time according to the construction path.

9. A method for on-site detection of backfill foundation compaction quality according to claim 8, characterized in that: The compaction distribution map is generated to detect abnormal points in the construction area and generate construction parameter optimization suggestions, including: Parameter recommendations for increasing the number of construction passes or increasing the exciting force of the vibrating wheel in areas with low compaction; Adjust the construction path to avoid repeated compaction of high compaction areas; The optimized construction path and parameters are sent to the vibratory roller control system for automatic adjustment.

10. A backfill soil foundation compaction quality on-site detection system, comprising: Signal acquisition module, signal processing module, dry density calculation module and compaction analysis module; characterized in that: The signal acquisition module is used to collect the instantaneous acceleration signal of the vibration wheel and the location information of the construction area in real time during the soil compaction process. The acceleration signal is obtained by the three-axis acceleration sensor installed on the vibration shaft, and the location information is collected by the GPS module and synchronously bound with the acceleration signal through the timestamp to form an acceleration data set; The signal processing module is used to perform segmentation processing, windowing, fast Fourier transform and spectrum analysis on the acceleration data set provided by the signal acquisition module, extract the main frequency amplitude and the first harmonic amplitude, and generate a filtered acceleration signal through bandpass filtering and noise reduction operations; The dry density calculation module is used to receive the filtered acceleration signal generated by the signal processing module, and input it into the trained single hidden layer BP neural network model in combination with the number of construction passes and position information, and output the predicted value of soil dry density corresponding to the construction point through calculation; The compaction analysis module is used to calculate the soil compaction at each point in the construction area based on the ratio of the soil dry density and the standard maximum dry density provided by the dry density calculation module, and generate a compaction distribution map according to the grid division, and display different compaction ranges through color coding.

Images