Standard penetrometer discreteness control method based on multi-modal feedback and dynamic correction

Abstract

The invention discloses a standard penetrometer discreteness control method based on multi-modal feedback and dynamic correction, and the method comprises the steps: carrying out Fourier filtering and temperature compensation on data collected by a sensor, and calculating the energy transmission efficiency eta of each hammering; constructing a dynamic response matrix based on eta and other sensor data, carrying out discreteness analysis, and triggering a corresponding correction strategy; the historical hammering data are analyzed through an LSTM network, the preset energy E1 of next hammering and probe rod adjustment parameters are predicted, and self-adaptive stratum adjustment is achieved; the height of the drop hammer is dynamically adjusted through the energy deviation delta E; when the proportion of the side friction resistance exceeds the limit, the lubricant is released through the micropores in the probe rod to reduce friction; and calculating a variable coefficient Cv of a penetration hammering number N value sequence corresponding to the latest hammering, and intelligently judging a test end point in combination with the penetration depth. Through multi-modal sensing and closed-loop dynamic correction, dispersion of penetration parameters is remarkably reduced, and real and reliable soil layer property data are provided for geotechnical engineering.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of geotechnical engineering test equipment, and particularly relates to a standard penetration instrument discreteness control method based on multi-modal feedback and dynamic correction. BACKGROUND

[0002] The standard penetration test (SPT) is a test method for simulating the bearing capacity of soil by using a specific standard hammer weight and drop height in the soil. As a core technology of geotechnical engineering investigation, the SPT is widely used in key engineering decisions such as foundation bearing capacity evaluation, sand liquefaction discrimination, soil layer division and consistency state identification due to its simple operation and rich empirical data.

[0003] However, the SPT test itself has characteristics such as high parameter variability and strong data discreteness, which makes the geotechnical parameters derived based on empirical formulas unstable and uncertain. In addition, the current standard penetration instrument in China still has deficiencies in instrument precision, calibration system and data intelligent processing, which together exacerbate the fluctuation of test results. If not effectively controlled, it will directly affect the accuracy of engineering design, may cause quality problems such as uneven settlement of foundation and cracking of building, and even threaten the overall engineering safety in serious cases The parameter discreteness of the current standard penetration test mainly comes from the following aspects: 1. Hammer energy deviation: The traditional mechanical drop hammer system has a large fluctuation in energy transmission during hammering, resulting in a large difference in the penetration effect of the soil layer by continuous hammering, which directly affects the accuracy of the penetration number.

[0004] 2. Soil disturbance accumulation: Continuous hammering penetration is a dynamic process that continuously disturbs the surrounding soil. During the continuous penetration process, the hammering action causes the compression of the soil around the probe and the change of the pore water pressure, which changes the stress state of the soil and affects the subsequent penetration resistance, resulting in systematic drift or instability of the test data.

[0005] 3. Environmental coupling interference: The complex environment of the test site, such as probe deflection and surface vibration, is not monitored and compensated in real time. These noise unrelated to the nature of the soil is coupled with the soil response during the test, further causing deviation in the test results. SUMMARY

[0006] The purpose of the present application is to overcome the shortcomings of the prior art and provide a standard penetration instrument discreteness control method based on multi-modal feedback and dynamic correction, which significantly reduces the discreteness of the penetration parameters of the standard penetration test through multi-modal sensing and dynamic correction, thereby providing reliable data that truly reflect the properties of the soil layer for geotechnical engineering investigation and design.

[0007] In order to achieve the above object, the technical scheme of the present application is as follows: a standard penetration instrument discreteness control method based on multi-modal feedback and dynamic correction, comprising the following steps: S1, processing the data collected by the sensors on the electromagnetic-hydraulic hybrid drive drop hammer device: filtering out high-frequency noise greater than the preset frequency f in the data based on Fourier transform, correcting the data drift in combination with the temperature compensation factor CT, and calculating the energy transfer efficiency η of each hammering according to the corrected data; S2, constructing a dynamic response matrix M by the calculated energy transfer efficiency η and in combination with the data collected by the sensors, performing single / multi-parameter discreteness analysis using the dynamic response matrix M, and triggering the corresponding correction strategy according to the discreteness analysis result; S3, using the LSTM network to analyze the data of the previous n hammerings, outputting the preset energy E1 of the next hammering and the probe rod adjustment parameters, and realizing adaptive formation adjustment in combination with the optimization strategy; S4, arranging n1 reference probes around the test point, and calculating the actual energy E2 after each hammering, and if the energy deviation ΔE is greater than the preset percentage threshold, adjusting the drop hammer height of the next hammering by a certain proportion; the energy deviation ΔE is the absolute difference between the actual energy E2 and the preset energy E1; S5, when the ratio of the real-time side friction resistance to the total penetration resistance is greater than the preset ratio, releasing the silicone oil lubricant through the micro-holes in the probe rod to reduce the friction coefficient; S6, when the rod body bending strain is greater than the preset value ε, adjusting the levelness of the equipment base; calculating the mean μ and the standard deviation σ of the current penetration hammering number N value sequence, if any N value Ni satisfies |Ni-μ|>3σ, marking the Ni as suspicious data, and repeatedly penetrating the upper and lower preset sections corresponding to the depth of the suspicious data for review; S7, based on the penetration hammering number N value sequence corresponding to the last m hammerings, calculating the coefficient of variation Cv of the sequence, and when the preset condition is met, the test is completed and terminated.

[0008] As a further solution, the specific steps of processing the data collected by the sensors and calculating the energy transfer efficiency η in step S1 are as follows: S1.1, calculating the temperature compensation factor CT, CT=1+α(T T0), wherein α is the thermal expansion coefficient of the sensor material, T0 is the calibration reference temperature, and T is the real-time temperature; S1.2, temperature compensation is performed on the original time domain signal collected by the sensor, and the corrected impact force and speed are obtained: Fh1(t)=Fh(t) / CT,Fp1(t)=Fp(t) / CT; Wherein, Fh1(t) is the corrected hammer head t time impact force original waveform, Fp1(t) is the corrected probe t time bottom counterforce original waveform, Fh(t) is the hammer head t time impact force original waveform, Fp(t) is the probe t time bottom counterforce original waveform; vh1(t)=vh(t) / CT, vp1(t)=vp(t) / CT; Wherein, vh1(t) is the corrected hammer head t time speed original signal, vp1(t) is the corrected probe bottom end t time speed original signal, vh(t) is the hammer head t time speed original signal; S1.3, Fourier transform is carried out to the corrected time domain signal Fh1(t), vh1(t), Fp1(t), vp1(t), and the effective frequency band is reserved through low pass filtering to obtain frequency domain signals Fh(f), Vh(f), Fp(f), Vp(f); Wherein, Fh(f) is the signal obtained by Fourier transform on the corrected Fh1(t), Vh(f) is the signal obtained by Fourier transform on the corrected vh1(t), Fp(f) is the signal obtained by Fourier transform on the corrected Fp1(t), and Vp(f) is the signal obtained by Fourier transform on the corrected vp1(t); S1.4, based on the frequency domain signals obtained by Fourier transform, the hammer input energy Eh and the soil layer absorbed energy Ep are calculated:

[0009] Wherein, fmax is the highest frequency, Vh'(f) is the conjugate complex number of Vh(f), It is the data after Fourier transform of Fh1(t), and Re represents taking real part;

[0010] Wherein, fmax is the highest frequency, Vp'(f) is the conjugate complex number of Vp(f), It is the data after Fourier transform of Fp1(t), and Re represents taking real part; S1.5, the energy transfer efficiency η is calculated, η=Ep / Eh*100%.

[0011] As a further solution, the single / multi-parameter discreteness analysis is carried out by using the dynamic response matrix M in step S2, and the corresponding correction strategy is triggered according to the threshold setting: S2.1, based on the calculated energy transfer efficiency η, the standard deviation σ of the number of penetration hammer N value and the penetration rebound amount Δh obtained by combining the sensor acquisition data to construct the dynamic response matrix M, M=[η,σ,Δh]; S2.2, single / multi-parameter discreteness analysis is performed on the dynamic response matrix M; S2.3, a corresponding correction strategy is triggered according to the single / multi-parameter discreteness analysis result.

[0012] As a further solution, in step S2.2, the single-parameter discreteness analysis includes: When the energy transfer efficiency η is greater than a first limit value, it is judged that the energy transfer is abnormal; When the standard deviation σ of the penetration hammer number N value exceeds a second limit value, it is judged that the soil layer is uneven; When the instantaneous change rate of the penetration rebound amount Δh is greater than a third limit value, it is judged that the stratum interface is suddenly changed; The multi-parameter discreteness analysis includes: Correlation analysis is performed on the data in the dynamic response matrix M: When η is not associated with σ and Δh, it is judged that the energy transfer is abnormal; When σ and Δh are positively correlated and both are negatively correlated with η, it is judged that the soil layer is uneven; When η, σ and Δh deviate from the limit value, it is judged that the stratum interface is suddenly changed.

[0013] As a further solution, in step S2.3, the way of triggering the corresponding correction strategy according to the single / multi-parameter discreteness analysis result is as follows: S2.3.1, single-parameter correction: for energy transfer abnormality, adjust the drop hammer height or mercury liquid weight cavity hammer weight through the electromagnetic-hydraulic drive system, so that the energy transfer efficiency η returns to the preset efficiency range; S2.3.2, multi-parameter correction: use LSTM network and Q-learning reinforcement learning algorithm to optimize, input the energy transfer efficiency η, the standard deviation σ of the penetration hammer number N value, and the penetration rebound amount Δh of the previous n times hammer, to predict the optimal energy En and the probe rod parameter combination, so that the matrix norm ∥M∥ of the corrected dynamic response matrix M converges to the preset range.

[0014] As a further solution, in step S3, LSTM network is used to analyze the previous n times hammer data, and the preset energy E1 and the probe rod adjustment parameter of the next hammer are output, and the adaptive stratum adjustment is realized by combining the optimization strategy. The specific process is as follows: S3.1, timing data processing: input the timing data sequence X of the previous n hammering, X=[Fh(t), Fp(t), vp(t), η(t), Rs] into the trained LSTM network; wherein, Fh(t) is the original waveform of the hammer head t impact force, Fp(t) is the original waveform of the probe rod t bottom reaction force, vp(t) is the original signal of the probe rod bottom t speed, η(t) is the t energy transmission efficiency; Rs is the ratio of side friction resistance to total penetration resistance; S3.2, parameter prediction: the LSTM network outputs the preset energy E1 of the next hammering and the probe rod adjustment parameter based on the timing data sequence, the probe rod adjustment parameter includes sleeve shrinkage rate and damping coefficient; S3.3, strategy optimization: the preset energy E1 and probe rod adjustment parameter are optimized and corrected based on the Q-learning reinforcement learning algorithm, the Q-learning reinforcement learning algorithm selects the combination mode of sleeve shrinkage rate and release of lubricant according to the current state, so that the corrected side friction resistance ratio Rs is reduced; wherein, the reward function R is designed as: when ΔE<z1 and Cv<z2, a positive reward coefficient R1 is given; when Rs>z3 or Cv>z2, a negative penalty R2 is given; wherein, Cv is the N value coefficient of variation, z1, z2, z3, z4 are preset variation coefficient thresholds; S3.4, cooperative execution: based on the optimized parameters, the hammer height is adjusted through the electromagnetic-hydraulic driving system, and the probe rod damping characteristics are adjusted through the damping adjustment device.

[0015] As a further solution, step S7 further comprises: If the coefficient of variation Cv of the N value sequence corresponding to the last m hammering is less than or equal to z2, and the penetration depth reaches the preset depth, it is determined that the test is completed, and the final N value sequence and related penetration parameters are output; If the coefficient of variation Cv of the N value sequence corresponding to the last m hammering is greater than z2, or the penetration depth does not reach the preset depth, the penetration is reviewed, and steps S4 to S7 are repeated.

[0016] Compared with the prior art, the beneficial effects of the present scheme are: The scheme effectively avoids the errors introduced by human factors in traditional operations through automatic data acquisition and processing, and significantly improves the accuracy and reliability of the detection results through fusion analysis of multi-modal and multi-sample time series data. Based on the dynamic feedback mechanism, the energy output of each hammering can be ensured to be highly stable, and adaptive adjustment can be made according to the mechanical properties of different strata, effectively reducing the dispersion of the number of penetration hammers N, and ensuring effective penetration testing under various stratum conditions. In addition, by constructing a dynamic response matrix of multi-parameter fusion, the state change of the soil layer can be identified, and the drop hammer strategy and probe rod parameters can be adjusted accordingly, effectively suppressing data fluctuations caused by soil disturbance, probe rod friction and other factors, and ensuring the continuity and stability of the test results at different depths and stratum conditions.

[0017] In summary, the scheme not only controls the main error source from the source, but also improves the data quality, engineering applicability and geological interpretation ability of the standard penetration test through the whole adaptive regulation mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A step flowchart of the method for controlling the dispersion of the standard penetration instrument based on multi-modal feedback and dynamic correction in an embodiment of the present application is shown. Figure 2 A flowchart of the method for controlling the dispersion of the standard penetration instrument based on multi-modal feedback and dynamic correction in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments.

[0020] As shown in Figure 1 and Figure 2 , the present application provides a method for controlling the dispersion of a standard penetration instrument based on multi-modal feedback and dynamic correction, comprising the following steps: S1, processing the data collected by the sensors on the electromagnetic-hydraulic hybrid drive drop hammer device: filtering out high-frequency noise greater than a preset frequency f in the data based on Fourier transform, correcting data drift in combination with a temperature compensation factor CT; and calculating the energy transfer efficiency η of each hammering according to the corrected data; S2, constructing a dynamic response matrix M by the calculated energy transfer efficiency η and in combination with the data collected by the sensors, using the dynamic response matrix M for single / multi-parameter dispersion analysis, and triggering the corresponding correction strategy according to the dispersion analysis result; S3, adopt the LSTM network to analyze the previous n times hammering data, output the preset energy E1 of the next hammering and the probe rod adjusting parameter, and realize adaptive formation adjustment in combination with the optimization strategy; S4, n1 reference probe rods are arranged around the test point, and the actual energy E2 is calculated after each hammering; if the energy deviation AE between the actual energy E2 and the preset energy E1 is greater than the preset percentage threshold, the drop hammer height of the next hammering is adjusted in proportion; the energy deviation AE is the absolute difference between the actual energy E2 and the preset energy E1; S5, when the ratio of the real-time side friction to the total penetration resistance is greater than the preset ratio, the silicon oil lubricant is released through the micro-holes in the probe rod to reduce the friction coefficient; S6, when the rod body bending strain is greater than the preset value ε, the equipment base levelness is adjusted; the mean μ and the standard deviation σ of the current penetration hammering number N value sequence are calculated, if any N value Ni satisfies |Ni-μ|>3σ, the Ni is marked as suspicious data, and the upper and lower preset sections of the depth corresponding to the suspicious data are repeatedly penetrated for review; S7, based on the penetration hammering number N value sequence corresponding to the last m times hammering, the coefficient of variation Cv of the sequence is calculated, and when the preset condition is met, the test is completed and terminated.

[0021] This embodiment takes the test of a clay-sand interbedded site as an example to illustrate the specific implementation process of the present application in detail: In this embodiment, the core driving device of the standard penetration instrument is an electromagnetic-hydraulic hybrid driving drop hammer system, the mercury liquid dynamic counterweight cavity of the drop hammer mass is adjusted to 63.5 kg, and the initial sleeve diameter is set to 50 mm. Various sensors such as accelerometers, strain gauges, temperature sensors and inclinometers are integrated on the probe rod system and the drop hammer device for real-time collection of multi-modal data in the penetration process.

[0022] The control flow of the method includes the following steps: Data preprocessing and energy efficiency calculation: The original data collected by the sensors (such as acceleration, force, temperature) are preprocessed. Specifically, the preset frequency f is set to 1 kHz, and Fourier transform is used to filter out high-frequency electrical noise above this frequency. At the same time, the temperature compensation factor CT is used to correct the drift of the sensor readings with temperature changes. Based on the corrected force-velocity data, the actual energy transfer efficiency η of each hammering is calculated.

[0023] Dynamic response analysis and correction triggering: The energy transfer efficiency η calculated in step S1 is fused with the data collected by the sensors in this step to construct a dynamic response matrix M, and the matrix is used for discreteness analysis, and the corresponding correction strategy is triggered according to the discreteness analysis result.

[0024] Intelligent prediction and adaptive adjustment: This step adopts a pre-trained LSTM neural network, whose structure includes an input layer, a hidden layer and an output layer. Through the network analysis of time series data, the preset energy E1 of the next hammering and the probe rod adjustment parameter (such as the expected amount of lubricant release) are output. The prediction result is fine-tuned through an optimization strategy combined with the Q-learning reinforcement learning algorithm, so that the system can better adapt to the change of the stratum from soft clay to dense sand.

[0025] Reference probe rod arrangement: In order to obtain reliable reference data of the surrounding soil, n1=3 reference probe rods are arranged in a regular triangle around the main test probe rod. In this application, the probe rod arrangement can be arranged in a polygon according to actual needs, and the regular triangle is the preferred scheme in this embodiment. These reference probe rods are used to measure the in-situ stress and pore water pressure at different positions, providing an environmental reference for judging the rationality of the main probe rod data.

[0026] Energy closed-loop correction: After each hammering, the actual penetration energy E2 is calculated according to the sensor data, the energy deviation ΔE is calculated, ΔE=|E2-E1|, and when ΔE>n*3%, the drop hammer height of the next hammering is adjusted in proportion. For example, if ΔE is 15%, the system will adjust the drop hammer height by about 7.5% according to the characteristic curve of the electromagnetic-hydraulic drive system, to realize accurate closed-loop control of energy.

[0027] Real-time friction compensation: The ratio of the side friction resistance to the total penetration resistance is calculated in real time, and if the real-time side friction resistance ratio is greater than n*4%, it is determined that the interference of the rod side friction on the test is too large, and the lubrication mechanism is started immediately. The silicon oil lubricant is released through the micro-holes pre-installed in the probe rod, lasting for 1-2 seconds, so as to effectively reduce the friction coefficient between the probe rod and the soil.

[0028] Attitude correction and data review: When the strain value monitored by the rod bending strain sensor is greater than the preset value ε, the hydraulic leveling mechanism is triggered to adjust the levelness of the equipment base to ensure that the probe rod is vertically penetrated. At the same time, the mean μ and standard deviation σ of the sequence of the recorded penetration hammering number N values are calculated, and the Ralda criterion (3σ criterion) is applied for outlier identification. If any N value Ni satisfies |Ni-μ|>3σ, the Ni is marked as suspicious data, and the upper and lower depth sections of the depth corresponding to the suspicious data are repeatedly penetrated for review to confirm whether the abnormal value is caused by stratum mutation or test interference.

[0029] Quality evaluation: The coefficient of variation Cv is calculated by taking the N value sequence of the last m=5 hammerings.

[0030] Termination condition one: if Cv ≤ z2, and the cumulative penetration depth has reached the preset depth required by the standard, it is determined that the test data quality is excellent, the system terminates the test, and outputs the final N value sequence and related penetration parameters (such as average energy, friction curve, etc.).

[0031] Termination condition two: if the penetration depth does not reach the preset depth, or Cv > z2. This case indicates that the data dispersion is high, or the test point has been disturbed by previous hammering, and the current results are not sufficient to represent the true properties of the undisturbed soil layer. To prevent further testing at the disturbed location, the system will shift to a completely new, undisturbed test point, and return to step S4 (n1 reference probes are arranged in a regular triangle centered on the test point), and re-execute the subsequent correction and test cycle until the penetration depth reaches the predetermined depth while meeting the coefficient of variation requirement.

[0032] Through the above closed-loop control process, the traditional standard penetration test is transformed from an open-loop, passive recording process to a closed-loop, actively optimized intelligent testing scheme. The final N value sequence can more truly and clearly reflect the changes in the mechanical properties of the soil layer, significantly improving the accuracy of geological stratification analysis and the reliability of geotechnical engineering design.

[0033] In a preferred embodiment of the present application, the specific implementation of processing the data collected by the sensor in step S1 and calculating the energy transfer efficiency η is further refined as follows: Based on Fourier transform, high-frequency noise higher than the preset frequency f in the signal is filtered out; According to the temperature compensation factor CT, the data is drift-corrected, and based on the corrected data, the energy transfer efficiency η of each hammering is calculated, and the calculation process is as follows: S1.1, calculate the temperature compensation factor CT, CT = 1 + α (T T0), where α is the thermal expansion coefficient of the sensor material, T0 is the calibration reference temperature, and T is the real-time temperature; S1.2, temperature compensation is performed on the original time domain signal collected by the sensor to obtain the corrected impact force and velocity: Fh1(t) = Fh(t) / CT, Fp1(t) = Fp(t) / CT; Where Fh1(t) is the corrected impact force original waveform of the hammer head at time t, Fp1(t) is the corrected bottom reaction force original waveform of the probe at time t, Fh(t) is the impact force original waveform of the hammer head at time t, and Fp(t) is the bottom reaction force original waveform of the probe at time t; vh1(t) = vh(t) / CT, vp1(t) = vp(t) / CT; Wherein, vh1(t) is the modified hammer head t time speed original signal, vp1(t) is the modified probe bottom t time speed original signal, and vh(t) is the hammer head t time speed original signal; S1.3, Fourier transform is carried out to the modified time domain signal Fh1(t), vh1(t), Fp1(t), vp1(t), and frequency domain signal Fh(f), Vh(f), Fp(f), Vp(f) is obtained, a low pass filter is used for filtering, so as to filter out high frequency noise and retain effective low frequency components representing the mechanical response of the soil body; Wherein, Fh(f) is the signal obtained by Fourier transform on the modified Fh1(t), Vh(f) is the signal obtained by Fourier transform on the modified vh1(t), Fp(f) is the signal obtained by Fourier transform on the modified Fp1(t), and Vp(f) is the signal obtained by Fourier transform on the modified vp1(t); S1.4, based on the frequency domain signal obtained by Fourier transform, the hammer input energy E h And the soil layer absorption energy E p :

[0034] Wherein, fmax is the highest frequency, Vh'(f) is the conjugate complex number of Vh(f), Fh1(t) Fourier transform data, Re represents taking real part;

[0035] Wherein, fmax is the highest frequency, Vp'(f) is the conjugate complex number of Vp(f), Fp1(t) Fourier transform data, Re represents taking real part; S1.5, calculate the energy transfer efficiency η, η=Ep / Eh*100%; According to the embodiment of the application, by processing the original data collected by the equipment, filtering out high frequency noise higher than the preset frequency f, and calculating the temperature compensation factor CT, the original impact force and speed signal are corrected for temperature drift, and then the hammer input energy and the soil layer absorption energy are accurately calculated based on the corrected signal, and finally the reliable energy transfer efficiency η is obtained. Specifically, through the temperature compensation mechanism, the data deviation caused by the change of environmental temperature can be effectively reduced, and the collected waveform and energy parameters are closer to the real physical process, so as to improve the test accuracy and data consistency of the standard penetration instrument in complex stratum as a whole.

[0036] According to another embodiment of the present application, a dynamic response matrix M is constructed by the calculated energy transmission efficiency η and the data collected by the sensors, the single / multi-parameter discreteness analysis is performed by using the dynamic response matrix M, and the corresponding correction strategy is triggered according to the discreteness analysis result: S2.1, based on the calculated energy transmission efficiency η, the standard deviation σ of the number of penetration hammer strikes N and the penetration rebound amount Δh obtained by combining the data collected by the sensors to construct a dynamic response matrix M, M=[η, σ, Δh]; S2.2, single / multi-parameter discreteness analysis is performed on the data in the dynamic response vector M: the data in the matrix M is analyzed to diagnose the abnormal root cause; The single-parameter discreteness analysis includes: When the energy transmission efficiency η is greater than the first limit value, it is judged that the energy transmission is abnormal; When the standard deviation σ of the number of penetration hammer strikes N exceeds the second limit value, it is judged that the soil layer is uneven; When the instantaneous change rate of the penetration rebound amount Δh is greater than the third limit value, it is judged that the stratum interface is suddenly changed; The multi-parameter discreteness analysis includes: The energy transmission efficiency η, the standard deviation σ of the number of penetration hammer strikes N, and the penetration rebound amount Δh are analyzed for correlation: When the energy transmission efficiency η is not related to the standard deviation σ and the rebound amount Δh, it is judged that the energy transmission is abnormal; When the standard deviation σ and the rebound amount Δh are positively correlated, and both are negatively correlated with the efficiency η, it is judged that the soil layer is uneven; When the three parameters η, σ, and Δh simultaneously deviate from their respective normal fluctuation ranges, it is judged that the stratum interface is suddenly changed.

[0037] S2.3, according to the above analysis result, the corresponding correction strategy is triggered: S2.3.1, single-parameter correction: If it is diagnosed that the energy transmission is abnormal, the height of the drop hammer or the weight of the mercury liquid weight cavity hammer is adjusted by the electromagnetic-hydraulic drive system, so that the energy transmission efficiency η returns to the preset efficiency range; S2.3.2, multi-parameter correction: If the complex working condition (such as uneven soil layer or interface mutation) is diagnosed, the LSTM network is used in cooperation with the Q-learning reinforcement learning algorithm for optimization. The data sequence of the previous n hammering in the matrix M is taken as the input, the time-dependent characteristics are learned through the recurrent neural network, and the long-term benefits of different correction actions are evaluated by the Q-learning algorithm, and finally an optimal preset energy En and probe rod parameter combination is output. The goal of this combination is to minimize the norm of the corrected dynamic response matrix M, that is, to make the dynamic response of the whole system converge to a stable and low-dispersion state.

[0038] In this embodiment, the dynamic response matrix M based on the energy transmission efficiency η, the standard deviation σ of the number of penetration hammering N and the penetration rebound amount Δh is constructed, and the single-parameter threshold judgment and multi-parameter correlation analysis are combined, so that different types of complex working conditions such as energy transmission abnormality, uneven soil layer and stratum interface mutation are accurately distinguished and diagnosed. The joint diagnosis mechanism of multi-mode and multi-criterion overcomes the limitation of the traditional method which relies on a single parameter and is prone to misjudgment. For different dispersion sources diagnosed, for the single-parameter abnormality, single-parameter correction is adopted, for example, when the energy transmission efficiency η deviates from the limited interval, the hammer height is adjusted through the electromagnetic-hydraulic driving system to compensate for the energy loss; when the standard deviation σ of the number of penetration hammering N deviates from the limited interval, the silicone lubricant is released to reduce the friction coefficient; when the penetration rebound amount Δh deviates from the limited interval, the hammering frequency is reduced. For complex working conditions involving multi-parameter coupling, the LSTM network is used in cooperation with the Q-learning reinforcement learning algorithm for optimization strategy. Specifically, the optimization strategy can learn the time sequence rule of the previous n hammering, and through the optimization of the reward function, the optimal energy En and probe rod parameter combination that can make the norm of the dynamic response matrix M converge are predicted. Through the correction method combining feedforward prediction and feedback optimization, not only the reaction is rapid, but also the method has the advantages of foresight and self-learning ability, so as to actively adapt to the complex and changeable stratum conditions, ensure the accuracy and reliability of the data through the respective processing conditions of different factors, adjust the data deviation in time, and finally ensure that the obtained N value sequence has a lower coefficient of variation and higher repeatability, thereby reducing the engineering misjudgment and potential risk from the technical level.

[0039] According to another embodiment of the present application, the operation of pre-hammering is started, low-energy pre-hammering is first performed, then the LSTM network is used to analyze the previous n hammering data, and the preset energy E1 and probe rod adjustment parameter of the next hammering are output, and the adaptive stratum adjustment is realized in combination with the optimization strategy: S3.1, time series data processing: input the time series data sequence X t of the previous n hammering, including Fh(t), Fp(t), vp(t), η(t), and Rs, into the trained LSTM network; wherein, Fh(t) is the original waveform of the hammer head impact force at time t, Fp(t) is the original waveform of the bottom reaction force of the probe at time t, vp(t) is the original signal of the speed of the probe bottom at time t, η(t) is the energy transmission efficiency at time t, and Rs is the ratio of side friction resistance to total penetration resistance; S3.2, parameter prediction: the LSTM network outputs the preset energy E1 of the next hammering and the probe adjustment parameters, including sleeve shrinkage rate and damping coefficient, based on the time series data sequence; S3.3, strategy optimization: the Q-learning reinforcement learning algorithm is used to optimize the strategy to optimize and correct the preset energy E1 and the probe adjustment parameters, wherein the Q-learning reinforcement learning algorithm selects the combination of sleeve shrinkage rate and release of lubricant according to the current state, so that the corrected side friction resistance ratio Rs is reduced; wherein, the reward function R is designed as: when ΔE < z1 and Cv < z2, a positive reward coefficient R1 is given; when Rs > z3 or Cv > z2, a negative penalty R2 is given; wherein, Cv is the N-value coefficient of variation, and z1, z2, and z3 are preset coefficient of variation thresholds; S3.4, collaborative execution: based on the optimized parameters, the electromagnetic-hydraulic drive system is used to adjust the drop hammer height, and the damping adjustment device is used to adjust the damping characteristics of the probe.

[0040] In this embodiment, the LSTM network processes the time series data sequence of the previous n hammering, including impact force, bottom reaction force, speed signal, energy transmission efficiency, and side friction resistance ratio, etc. The LSTM network learns and remembers the dynamic characteristics in the penetration process, thereby realizing accurate prediction of the parameters of the next hammering. The Q-learning reinforcement learning algorithm optimizes the long-term control benefit through the reward function mechanism, wherein the reward function considers multiple key indicators such as energy deviation ΔE, N-value coefficient of variation Cv, and side friction resistance ratio Rs. When the energy deviation is small and the data stability is high, a positive reward is given, and when the side friction resistance is too large or the data dispersion increases, a negative penalty is applied, thereby effectively guiding the system to evolve in the direction of overall optimization. Finally, the electromagnetic-hydraulic drive system and the damping adjustment device execute the optimized control instructions, realize dynamic adjustment of the drop hammer energy and the probe parameters according to the real-time working conditions, reduce the dependence on the experience of the operators, and reduce the errors caused by human factors, thereby obtaining more accurate stratigraphic identification ability and more reliable geotechnical parameter evaluation results.

[0041] The above are only embodiments of the present application, and common knowledge of specific structures and / or characteristics in the scheme is not described in detail here. It should be noted that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can also be made, which should also be considered as the protection scope of the present application, which will not affect the effect and practicality of the patent. The protection scope claimed in this application should be subject to the content of its claims, and the specific implementation mode and the like recorded in the specification can be used to explain the content of the claims.

[0042] The above are only embodiments of the present application, and common knowledge of specific structures and / or characteristics in the scheme is not described in detail here. It should be noted that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can also be made, which should also be considered as the protection scope of the present application, which will not affect the effect and practicality of the patent. The protection scope claimed in this application should be subject to the content of its claims, and the specific implementation mode and the like recorded in the specification can be used to explain the content of the claims.

Claims

1. A discrete control method for a standard penetrator based on multimodal feedback and dynamic correction, wherein the standard penetrator includes an electromagnetic-hydraulic hybrid drive drop hammer device and a probe, characterized in that, Includes the following steps: S1. Process the data collected by the sensors on the electromagnetic-hydraulic hybrid drive drop hammer device: filter out high-frequency noise greater than the preset frequency f in the data based on Fourier transform, and correct the data drift by combining the temperature compensation factor CT; and calculate the energy transfer efficiency η of each hammer blow based on the corrected data. S2. Construct a dynamic response matrix M by calculating the energy transfer efficiency η and combining it with the data collected by the sensor. Use the dynamic response matrix M to perform single / multi-parameter discreteness analysis and trigger the corresponding correction strategy based on the discreteness analysis results. S3. Use an LSTM network to analyze the data from the previous n hammer blows, output the preset energy E1 for the next hammer blow and the probe adjustment parameters, and combine the optimization strategy to achieve adaptive formation adjustment. S4. Set up n1 reference probes centered on the test point, and calculate the actual energy E2 after each hammer blow; if the energy deviation ΔE is greater than the preset percentage threshold, adjust the hammer drop height for the next hammer blow proportionally; the energy deviation ΔE is the absolute difference between the actual energy E2 and the preset energy E1. S5. When the ratio of real-time side friction resistance to total penetration resistance is greater than the preset ratio, silicone oil lubricant is released through the micro-holes inside the probe to reduce the coefficient of friction. S6. When the bending strain of the rod body is greater than the preset value ε, adjust the levelness of the equipment base; calculate the mean μ and standard deviation σ of the current number of penetration hammer blows N value sequence. If any N value Ni satisfies |Ni-μ|>3σ, then mark the Ni as suspicious data and repeat the penetration verification of the upper and lower preset sections corresponding to the depth of the suspicious data. S7. Based on the sequence of penetration hammer blows N corresponding to the most recent m hammer blows, calculate the coefficient of variation Cv of the sequence, and the test is completed and terminated when the preset conditions are met.

2. The discrete control method for a standard penetrator based on multimodal feedback and dynamic correction according to claim 1, characterized in that, The specific steps in step S1 for processing the data collected by the sensor and calculating the energy transfer efficiency η are as follows: S1.1 Calculate the temperature compensation factor CT, CT = 1 + α(T T0), where α is the coefficient of thermal expansion of the sensor material, T0 is the calibration reference temperature, and T is the real-time temperature; S1.

2. Perform temperature compensation on the raw time-domain signal acquired by the sensor to obtain the corrected impact force and velocity: Fh1(t)=Fh(t) / CT, Fp1(t)=Fp(t) / CT; Wherein, Fh1(t) is the original waveform of the hammer impact force at time t after correction, Fp1(t) is the original waveform of the bottom reaction force of the probe at time t after correction, Fh(t) is the original waveform of the hammer impact force at time t, and Fp(t) is the original waveform of the bottom reaction force of the probe at time t. vh1(t)=vh(t) / CT, vp1(t)=vp(t) / CT; Where vh1(t) is the corrected original velocity signal of the hammer head at time t, vp1(t) is the corrected original velocity signal of the bottom end of the probe at time t, and vh(t) is the original velocity signal of the hammer head at time t. S1.3 Perform Fourier transform on the corrected time-domain signals Fh1(t), vh1(t), Fp1(t), and vp1(t) to obtain the frequency-domain signals Fh(f), Vh(f), Fp(f), and Vp(f), and retain the effective frequency bands through low-pass filtering; Where Fh(f) is the signal obtained by performing a Fourier transform on the modified Fh1(t), Vh(f) is the signal obtained by performing a Fourier transform on the modified vh1(t), Fp(f) is the signal obtained by performing a Fourier transform on the modified Fp1(t), and Vp(f) is the signal obtained by performing a Fourier transform on the modified vp1(t). S1.4 Calculate the hammerhead input energy E based on the frequency domain signal obtained from the Fourier transform. h and soil layer absorbs energy E p : Where fmax is the highest frequency, and Vh'(f) is the complex conjugate of Vh(f). Let Fh1(t) be the Fourier transform of the data, and Re denote the real part. Where fmax is the highest frequency, and Vp'(f) is the complex conjugate of Vp(f). The data is obtained after the Fourier transform of Fp1(t), and Re represents taking the real part; S1.5 Calculate the energy transfer efficiency η, η=E p / E h *100%.

3. The discrete control method for a standard penetrator based on multimodal feedback and dynamic correction according to claim 1, characterized in that, In step S2, the dynamic response matrix M is used to perform single / multi-parameter discreteness analysis, and the corresponding correction strategy is triggered according to the threshold setting: S2.1 Based on the calculated energy transfer efficiency η, and combined with the standard deviation σ of the number of penetration hammer blows N and the penetration rebound amount Δh obtained from the sensor data, construct the dynamic response matrix M, M=[η,σ,Δh]; S2.2 Perform single / multi-parameter discreteness analysis on the dynamic response matrix M; S2.3 Trigger the corresponding correction strategy based on the discreteness analysis results of single / multi-parameter parameters.

4. The discrete control method for a standard penetrator based on multimodal feedback and dynamic correction according to claim 3, characterized in that, In step S2.2, the single-parameter discreteness analysis includes: When the energy transfer efficiency η is greater than the first limit value, it is judged as an energy transfer anomaly; When the standard deviation σ of the number of penetration hammer blows N exceeds the second limit value, it is judged as soil inhomogeneity; When the instantaneous rate of change of the penetration rebound amount Δh is greater than the third limit value, it is judged as a sudden change in the formation interface; The multi-parameter discreteness analysis includes: Correlation analysis was performed on the data in the dynamic response matrix M: When η is not related to σ and Δh, it is judged as an energy transfer anomaly; When σ is positively correlated with Δh and both are negatively correlated with η, it is judged that the soil layer is heterogeneous; When η, σ, and Δh deviate from the specified values, it is judged as a sudden change in the formation interface.

5. The discrete control method for a standard penetrator based on multimodal feedback and dynamic correction according to claim 4, characterized in that, The method for triggering the corresponding correction strategy based on the discreteness analysis results of single / multi-parameter parameters in step S2.3 is as follows: S2.3.1 Perform single-parameter correction: In case of abnormal energy transmission, adjust the drop hammer height or the hammer weight of the mercury liquid counterweight chamber through the electromagnetic-hydraulic drive system to bring the energy transmission efficiency η back to the preset efficiency range. S2.3.2 Perform multi-parameter correction: Utilize the LSTM network and Q-learning reinforcement learning algorithm for collaborative optimization. Input the energy transfer efficiency η of the first n hammer blows, the standard deviation σ of the number of hammer blows N, and the amount of springback Δh to predict the optimal combination of energy En and probe parameters, so that the matrix norm ∥M∥ of the corrected dynamic response matrix M converges to the preset range.

6. The discrete control method for a standard penetrator based on multimodal feedback and dynamic correction according to claim 5, characterized in that, In step S3, an LSTM network is used to analyze the data from the previous n hammer blows, outputting the preset energy E1 for the next hammer blow and the probe adjustment parameters. The specific process for achieving adaptive formation adjustment by combining optimization strategies is as follows: S3.1 Time-series data processing: Input the time-series data sequence Xt=[Fh(t),Fp(t),vp(t),η(t),Rs] of the first n hammer blows into the trained LSTM network; where Fh(t) is the original waveform of the impact force of the hammer head at time t, Fp(t) is the original waveform of the bottom reaction force of the probe at time t, vp(t) is the original velocity signal of the bottom of the probe at time t, η(t) is the energy transfer efficiency at time t; Rs is the ratio of side friction resistance to total penetration resistance; S3.

2. Parameter Prediction: Based on the time series data sequence, the LSTM network outputs the preset energy E1 for the next hammer strike and the probe adjustment parameters, where the probe adjustment parameters include the sleeve shrinkage rate and the damping coefficient; S3.

3. Strategy Optimization: Optimize and correct the preset energy E1 and the probe adjustment parameters based on the strategy optimized by the Q-learning reinforcement learning algorithm. The Q-learning reinforcement learning algorithm selects a combination of sleeve shrinkage rate and lubricant release according to the current state to reduce the proportion Rs of the corrected side friction resistance. The reward function R is designed as follows: when ΔE < z1 and Cv < z2, a positive reward coefficient R1 is given; when Rs > z3 or Cv > z2, a negative penalty R2 is given. Here, Cv is the coefficient of variation of the N value, and z1, z2, and z3 are preset coefficient of variation thresholds; S3.

4. Cooperative Execution: Based on the optimized parameters, adjust the drop hammer height through the electromagnetic-hydraulic drive system and adjust the damping characteristics of the probe through the damping adjustment device.

7. The discrete control method for a standard penetrator based on multimodal feedback and dynamic correction according to claim 6, characterized in that, Step S7 also includes: If the coefficient of variation Cv of the N value sequence corresponding to the most recent m hammer strikes is ≤ z2 and the penetration depth reaches the preset depth, it is determined that the test is completed, and the final N value sequence and related penetration parameters are output; If the coefficient of variation Cv of the N value sequence corresponding to the most recent m hammer strikes is > z2, or the penetration depth does not reach the preset depth, penetration verification is performed, and steps S4 to S7 are repeated.

Images