A method and system for predicting the length of engineering piles

Abstract

This invention discloses a method and system for estimating the length of engineering piles. The method includes: acquiring multi-source detection signals reflecting the pile hammer impact response and prior geological information; based on the prior geological information, performing physical-driven adaptive analysis on the acoustic components in the multi-source detection signals to extract preliminary acoustic reflection features; using the vibration components in the multi-source detection signals to perform correlation verification on the preliminary acoustic reflection features and generate corrected reflection features; evaluating the signal quality of the corrected reflection features and generating intelligent hammer impact control commands to execute closed-loop enhanced signal excitation; and finally fusing the multi-modal information before and after enhancement to calculate the final estimated pile length. This invention improves the accuracy, reliability, and detection depth of pile length estimation under conditions of high noise, complex geological formations, and great depth through acoustic-vibration coupling verification and intelligent active excitation.

Description

Technical Field

[0001] This invention belongs to the field of building engineering testing, and in particular to a method and system for estimating the length of engineering piles. Background Technology

[0002] The pile length is one of the core parameters determining the bearing capacity of a single pile, directly affecting whether the pile tip can be embedded into a predetermined, stable stratum with sufficient bearing capacity. In engineering practice, insufficient pile length will result in the foundation's bearing capacity failing to meet design requirements, potentially leading to uneven settlement or even structural failure. Conversely, excessively long piles not only waste materials and labor but also increase project costs. Especially in areas with complex and variable geological conditions, the actual pile depth often differs from the initial design. Therefore, developing a technology capable of real-time, accurate, and reliable prediction of pile length during pile foundation construction is of significant engineering value and economic importance for ensuring project quality, optimizing construction decisions, and controlling construction costs.

[0003] Currently, methods for detecting the length of engineering piles are mainly divided into two categories: static detection and dynamic detection. Static detection is mostly carried out after pile formation, such as the low-strain reflection wave method, which is one of the most widely used non-destructive testing techniques. Dynamic detection techniques, which estimate in real time during construction, are mainly based on stress wave theory. The basic principle is: the impact of the pile hammer generates a stress wave at the pile top, which propagates downwards along the pile. When it encounters the pile bottom or a ground interface with a significant difference in acoustic impedance, a reflected wave is generated. Sensors (such as accelerometers) placed on the pile collect signals containing the initial wave and the reflected wave. By measuring the time difference Δt between the initial wave and the reflected wave at the pile bottom, and combining this with the known wave propagation velocity c in the pile material, the pile length can be estimated using the formula L = 1 / 2cΔt. To extract the weak reflected signal from the strong construction noise, existing technical solutions typically employ digital filtering, Fourier transform, and other preprocessing techniques. A more advanced approach involves using wavelet transform technology from the field of signal processing to denoise the signal, or using matched filtering technology, which involves setting a theoretically standard reflection waveform as a template and finding the segment in the acquired signal that best matches it in order to identify the arrival time of the reflected wave.

[0004] However, when faced with complex field conditions, the existing dynamic detection technologies still have bottlenecks that urgently need to be addressed. These problems mainly focus on two aspects: first, insufficient ability to identify weak effective signals against strong noise backgrounds; and second, a lack of effective means to enhance deep signals and a cross-validation mechanism for analysis results. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for estimating the length of engineering piles, so as to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution: A method for estimating the length of engineering piles, comprising:

[0007] Acquire prior information about the formation that reflects the impact response of the pile hammer, as well as multi-source detection signals including acoustic and vibration components;

[0008] Based on prior information about the formation, adaptive analysis of the acoustic wave components is performed to extract preliminary acoustic wave reflection characteristics.

[0009] By using vibration components, the preliminary acoustic wave reflection characteristics are correlated and verified, spurious features are suppressed, and corrected reflection characteristics are generated.

[0010] The signal quality of the corrected reflection characteristics is evaluated, and intelligent hammering control commands are generated accordingly. Enhanced signal excitation is executed, and information before and after enhancement is fused to calculate the estimated pile length.

[0011] A system for predicting the length of engineering piles, comprising:

[0012] The sensor assembly is configured to acquire multi-source detection signals that reflect the impact response of the pile hammer;

[0013] The electrically controlled hammering system is configured to receive control commands and execute hammering actions with adjustable parameters;

[0014] A signal processing unit, operatively coupled to the sensor assembly and the electrically controlled hammer impact system, is configured as follows:

[0015] Based on the acoustic wave components in the multi-source detection signals obtained from the sensor assembly and the pre-stored a priori information of the formation, preliminary acoustic wave reflection characteristics are extracted.

[0016] By utilizing the vibration components in the multi-source detection signals, the preliminary acoustic wave reflection characteristics are correlated and verified to generate corrected reflection characteristics;

[0017] The modified reflection characteristics are evaluated, and based on this, intelligent hammering control commands are generated and sent to the electronically controlled hammering system to execute enhanced hammering excitation.

[0018] By integrating information before and after enhanced hammering, the final estimated pile length is calculated.

[0019] Beneficial effects: This invention improves the accuracy, reliability, and detection depth of pile length estimation under conditions of high noise, complex strata, and great depth. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the steps of a method for estimating the length of an engineering pile as provided in this application embodiment.

[0021] Figure 2A flowchart illustrating the steps for calculating the estimated pile length in this embodiment of the application.

[0022] Figure 3 A flowchart illustrating the steps for generating modified reflection features provided in an embodiment of this application.

[0023] Figure 4 A flowchart illustrating the steps for calculating the correlation between acoustic and vibration time delays provided in this application embodiment. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0026] The study found that at critical interfaces such as the boundary between soft and hard strata, the reflection coefficient of sound waves is extremely low (typically only 0.05-0.15), coupled with on-site construction noise of up to 80-120 dB, resulting in a signal-to-noise ratio often below -20 dB. Under these extreme conditions, traditional fixed-parameter filters, while filtering out noise, easily erase weak, real reflected signals. Even wavelet transforms or matched filters using fixed basis functions cannot adapt to the waveform distortion, dispersion, and attenuation that occur when sound waves propagate in complex strata. This leads to a severe mismatch between the preset template and the actually received, varied reflected waves, resulting in a significant decrease in interface recognition rate. Furthermore, sound wave energy attenuates exponentially with propagation depth. For deep piles exceeding 20 meters, the reflected signal at the pile bottom may be completely submerged in noise, creating blind spots for existing passive detection methods at depth. On the other hand, existing methods rely entirely on analyzing a single acoustic signal. When complex signal processing algorithms introduce seemingly real pseudo-reflection peaks (e.g., pseudo-peaks generated at the boundaries by segmented processing), the system lacks information from a second physical dimension for cross-validation. It cannot determine whether a detected peak is a real physical reflection or an artifact generated by the algorithm, which directly reduces the reliability of pile length determination.

[0027] This embodiment provides a system for predicting the length of an engineering pile. The system includes a sensor assembly, an electrically controlled hammer impact system, and a signal processing unit. The sensor assembly is configured to acquire multi-source detection signals reflecting the pile hammer impact response. Specifically, the sensor assembly may include a hammer force sensor for acquiring the pile hammer impact force signal, an accelerometer array (e.g., one every 2 meters along the pile) arranged on the pile shaft for acquiring acoustic wave signals, and a MEMS vibration sensor (e.g., a triaxial MEMS vibration sensor with a range of ±50g) installed at the pile tip for acquiring the soil response at the pile tip. Through a hardware triggering mechanism, microsecond-level synchronous acquisition of signals from each sensor channel can be achieved, resulting in a time-synchronized raw signal set. Optionally, in addition to the accelerometer, the pile shaft sensor array may also include strain gauges or fiber optic grating sensors to directly measure stress waves or strain. The pile tip MEMS vibration sensor may also be replaced by a high-sensitivity detector.

[0028] The electrically controlled hammering system is configured to receive control commands and execute hammering actions with adjustable parameters. This system can precisely adjust parameters such as the frequency, interval, and force of the hammering based on instructions sent by the signal processing unit.

[0029] The signal processing unit is operatively coupled to the sensor assembly and the electrically controlled hammer impact system. Optionally, the physical implementation of this unit can be an edge computing device deployed on the construction site, an embedded module integrated into the pile driver control system, or a remote server that uploads data to the cloud for computation. The signal processing unit is configured to perform actions such as... Figure 1 The method for estimating the length of engineering piles, as shown, includes the following steps:

[0030] Acquire prior information about the formation reflecting the impact response of the pile hammer, as well as multi-source detection signals including acoustic and vibration components.

[0031] In this embodiment, the multi-source detection signal is a set of time-series signal data synchronously collected by different physical sensors under a single hammer impact excitation, reflecting different physical phenomena (such as pile stress wave propagation and pile tip soil vibration). The signal processing unit receives the time-synchronized raw signal set from the sensor assembly and reads pre-stored geological exploration borehole data, standard penetration test data, etc., to construct a stratigraphic reference model as prior stratigraphic information. Before processing, the signal processing unit can also perform quality assessment on the raw signal, filter out noise, and perform baseline correction to obtain pre-processed acoustic signal and pre-processed vibration signal.

[0032] Based on prior information about the geological formation, adaptive analysis is performed on the acoustic wave components to extract preliminary acoustic wave reflection characteristics.

[0033] This embodiment aims to address the problem that traditional methods struggle to identify weak ground interface reflection signals in environments with strong construction noise (e.g., 80-120 dB) and extremely low signal-to-noise ratios (e.g., <-20 dB). In this embodiment, the signal processing unit employs a physically driven dynamic signal processing architecture, abandoning the traditional fixed basis function method and achieving adaptive decomposition of the acoustic signal.

[0034] By using vibration components, the preliminary acoustic wave reflection characteristics are correlated and verified, spurious features generated during the analysis process are suppressed, and corrected reflection characteristics are generated.

[0035] Specifically, this step aims to address the problem of spurious reflection peaks introduced at data boundaries due to algorithmic reasons such as segmented processing. In this embodiment, the signal processing unit establishes a physical correlation model between the acoustic wave reflection signal and the vibration signal of the pile tip soil, and uses the vibration signal to verify and correct the analysis results of the acoustic wave signal, effectively identifying and suppressing spurious features.

[0036] The signal quality of the corrected reflection characteristics is evaluated, and intelligent hammering control commands are generated based on the signal quality. Enhanced signal excitation is executed, and information before and after enhancement is fused to calculate the estimated pile length.

[0037] In this embodiment, this step aims to address the low detection success rate caused by rapid attenuation of deep signals (e.g., exceeding 20 meters). The signal processing unit dynamically optimizes the parameters of the next hammer blow by evaluating the energy attenuation state of the current signal and instructs the electronically controlled hammering system to execute a tailored enhancement excitation. This enables targeted enhancement of weak signals at depth, improving detection depth and reliability. Finally, by fusing multimodal information before and after enhancement, a high-precision final pile length estimate is calculated.

[0038] According to one aspect of this application, before extracting preliminary acoustic reflection features, a step of generating a depth-adaptive atomic dictionary is included. Specifically, this step includes:

[0039] By combining the acoustic components in the multi-source detection signals with prior information about the formation, a real-time physical parameter field varying with depth is estimated. This parameter field includes information on acoustic wave velocity and acoustic impedance.

[0040] The purpose of this step is to ensure that subsequent signal analysis models can match the actual physical properties of the formation in real time, rather than relying on fixed, unrealistic mathematical models. Specifically, the steps for estimating the real-time physical parameter field that varies with depth include:

[0041] The acoustic components in the multi-source detection signals generated by the initial hammer blows are analyzed, the arrival time difference Δt(z) of the first wave at adjacent depths is calculated, and the real-time wave velocity field v(z) that varies with depth is determined.

[0042] For example, the local wave velocity can be estimated using the sliding window method (window length 1 meter, step 0.2 meters). The calculation formula is: v(z) = Δz / Δt(z); where v(z) is the real-time wave velocity at depth z, Δz is the distance between adjacent sensors (e.g., 2 meters), and Δt(z) is the time difference between the arrival of the first wave of the sound wave at the two sensors.

[0043] The dynamic impedance field Z(z) is constructed by multiplying the real-time wave velocity field v(z) with the density data ρ(z) extracted from prior formation information. The calculation formula is: Z(z) = ρ(z) × v(z); where ρ(z) is the formation density at depth z, which can be obtained from the geological survey report. The real-time physical parameter field is jointly constituted by the real-time wave velocity field v(z) and the dynamic impedance field Z(z).

[0044] Based on the values ​​of the real-time physical parameter field at different depths, a depth-adaptive atomic parameter set is calculated; the frequency, phase, and attenuation coefficient of the atom are determined by the acoustic impedance abrupt change characteristics and wave velocity at that depth, respectively; using the depth-adaptive atomic parameter set, atomic functions with specific mathematical forms are generated for each depth, which together constitute a depth-adaptive atomic dictionary.

[0045] Specifically, an atomic dictionary is an overcomplete signal representation method composed of a set of basic waveforms (i.e., atoms) with specific mathematical forms. Complex signals can be sparsely represented through linear combinations of these atoms. The frequency, phase, and attenuation coefficient of the atomic functions are determined by the acoustic impedance abrupt change characteristics and wave velocity at that depth, respectively. For example, the atomic parameters for each depth can be calculated based on the impedance gradient ΞZ / Ξz and the wave velocity v(z), where Ξ is the partial derivative. The resulting dictionary of atoms can better match the waveform distortion and attenuation caused by geological changes during actual propagation, exhibiting stronger adaptability and matching accuracy compared to the fixed wavelet basis or preset dictionaries used in existing technologies. For example, the atomic function ψ(t, z) can be expressed as: ψ(t, z) = A(z) × exp(-α(z) × t) × sin(2πf(z) × t + Φ(z)); where t is time, z is depth, A(z) is amplitude, α(z) is attenuation coefficient, f(z) is frequency, and Φ(z) is phase. These parameters are all adaptively adjusted with depth z. In some implementations, in addition to calculating the atomic parameters through physical formulas, machine learning methods can be used to learn a deep neural network model capable of generating optimal atoms by training on a large amount of known stratigraphic data and corresponding reflection signals.

[0046] Based on this, preliminary sound wave reflection characteristics are extracted, specifically including:

[0047] The acoustic component in the multi-source detection signal is divided into a segmented signal sequence with overlapping beginnings and ends along the depth dimension.

[0048] For example, the signal can be divided into multiple intervals based on depth, such as [0-5 meters], [4-10 meters], [9-15 meters], etc., with adjacent intervals having a 1-meter overlap. Segmented processing can decompose a large-scale global optimization problem into multiple local subproblems with smaller computational requirements, thus meeting the requirements of real-time processing.

[0049] Configure a local projection operator for each signal segment, and perform cascaded projection calculations on the segmented signal sequence in sequence to obtain the projection result; wherein the projection calculation of the later signal segment includes the processing residual generated by the projection calculation of the previous segment.

[0050] Specifically, for each interval, the atoms most relevant to that depth range are selected from the depth-adaptive atom dictionary (e.g., through principal component analysis dimensionality reduction), and a local, optimized projection operator P is constructed. k Then, cascaded projection is performed. Cascaded projection is a fast algorithm that decomposes a large-scale global optimization problem into a series of interconnected local projection subproblems, where the residuals calculated in the previous stage are propagated and incorporated into the calculation of the next stage. For example: x k =P k (x k-1 +r k-1 ); where x k This is the projection result of the kth level, x k-1 It is the input signal of the previous stage, r k-1 This is the residual signal generated after the previous stage of projection calculation. By passing the residual from the previous stage to the calculation of the next stage, the effective transfer of signal energy and information between different segments is ensured, thus improving computational efficiency.

[0051] A weighted fusion method is used to smooth the overlapping areas of adjacent projection results to generate a continuous projection signal.

[0052] Due to segmented processing, a smooth transition is needed between the processing results of adjacent segments in the overlapping area to avoid abrupt changes. Specifically, the steps for generating a continuous projection signal include: for any data point within the overlapping area, calculating a set of position-related weights based on the data point's relative position within the overlapping area, where the closer the data point is to an endpoint, the greater the weight of the segment result containing that endpoint. Multiplying the corresponding values ​​of adjacent projection results within the overlapping area by the position-related weights and summing them yields the fused value of the data point in the continuous projection signal. For example, the fused value y... 融合 Calculate using the following formula: y 融合 =w1·y k +w2·y k+1 ; where y k and y k+1 The projection results are for two adjacent segments, where w1 and w2 are the position-related weights. Preferably, w1 = (1 - d / L).2 w2=(d / L) 2 , where L is the total length of the overlapping region (e.g., 1 meter), and d is the distance of the data point from the end of segment k. Optionally, in addition to quadratic weighted fusion, linear weighted fusion (equivalent to applying a triangular window) or sine / cosine cross fusion that can guarantee constant power can also be used to achieve a smooth transition.

[0053] Multi-scale reflection information is extracted and assembled from continuous projection signals to form preliminary acoustic wave reflection characteristics.

[0054] In this embodiment, the stratigraphic interface can be identified by finding abrupt changes in the reflection coefficient from the generated continuous projection signal, thereby obtaining preliminary information on the position and intensity of the reflection peak and constituting preliminary acoustic wave reflection characteristics.

[0055] Furthermore, after obtaining the preliminary acoustic wave reflection characteristics, the signal processing unit can also analyze the preliminary acoustic wave reflection characteristics to generate a preliminary pile length estimate.

[0056] Specifically, the reflection coefficient at each depth point is calculated, and abrupt changes in the reflection coefficient (e.g., a reflection coefficient value greater than 0.15 and its depth gradient exceeding a preset threshold) are identified to preliminarily identify stratigraphic interfaces. Abrupt changes in the reflection coefficient are points on the depth-reflection intensity mapping function that simultaneously satisfy both amplitude and gradient thresholds, physically corresponding to a stratigraphic interface with potentially significant acoustic impedance differences. The depth corresponding to the deepest identified significant stratigraphic interface is recorded as a preliminary pile length estimate. This preliminary pile length estimate is then passed to the subsequent intelligent hammering parameter optimization step, serving as one of the input bases for calculating the target resonant frequency and dynamic hammering force.

[0057] This embodiment improves the ability to identify and extract weak but effective signals in environments with extremely low signal-to-noise ratios (SNR). It constructs an atomic dictionary that adapts to depth by using real-time estimated wave velocity field v(z) and dynamic impedance field Z(z), abandoning the fixed mathematical basis functions used in existing technologies that are disconnected from actual physical processes. This means that the template used to match the signal (i.e., dictionary atoms) can simulate in real-time the real physical changes (such as dispersion and attenuation) that occur when sound waves propagate in complex strata. Therefore, when using this dynamic dictionary for sparse decomposition, it can accurately capture extremely weak real reflected signals that highly match the physical characteristics of the current depth from strong construction noise. This solves the problem of insufficient formation interface recognition rate caused by excessively low SNR (<-20dB) and mismatch between fixed templates and distorted waveforms, thus improving the recognition rate.

[0058] According to one aspect of this application, while cascaded projection improves computational efficiency, it inevitably produces pseudo-reflection peaks at the boundaries of segmented processing (e.g., depths of 4-5 meters, 9-10 meters, etc.), severely interfering with pile length determination. Therefore, by establishing a cross-modal time-delay correlation model between acoustic wave reflection and soil vibration, physical constraints are used to suppress artifacts generated by mathematical processing. Figure 3 As shown, the modified reflection features are generated, specifically including:

[0059] The spectral abrupt changes of vibration components in multi-source detection signals are detected, and the time-frequency features corresponding to the spectral abrupt changes are extracted to form vibration abrupt change features.

[0060] For example, the time-varying spectrum of a vibration signal can be calculated using a short-time Fourier transform (window length 100ms, step size 10ms). By calculating the spectral distance between adjacent time windows, D(t) = ||S(f,t) - S(f,t - Δt)||2, abrupt changes in the spectrum can be detected, thus obtaining the vibration abrupt change time sequence. Here, S(f,t) is the frequency distribution obtained by the short-time Fourier transform at time t, and Δt is the time difference.

[0061] For each reflection peak in the preliminary acoustic wave reflection characteristics, the acoustic-vibration time delay correlation between each reflection peak and the vibration abrupt change characteristics is calculated based on the theoretical time delay model of acoustic wave propagation.

[0062] The acoustic-vibration time delay correlation is a quantitative statistical indicator used to measure the strength of the linear relationship between a sound wave reflection event and a subsequent pile-end vibration event, after considering the theoretical propagation time delay. Specifically, such as... Figure 4 As shown, the calculation of acoustic-vibration time delay correlation includes:

[0063] For each reflection peak, the theoretical propagation delay is calculated by integrating the real-time wave velocity that varies with depth along the depth path of the reflection peak.

[0064] In this embodiment, for each detected reflection peak depth L, the theoretical time delay τ is calculated by considering the integral form of the wave velocity changing with depth. 理论 To improve accuracy 理论 =2∫0 L (dz / v(z))+t 系统 Where v(z) is the real-time wave velocity field that varies with depth, and t 系统 This is the inherent system latency (e.g., 3.5ms).

[0065] A delay search window is set for each theoretical propagation delay, where the size of the window is proportional to the value of the theoretical propagation delay, in order to adaptively compensate for the uncertainty of deep signals.

[0066] Preferably, the range of the search window can be set to [τ]. 理论 -Δτ,τ 理论 +Δτ], where the formula for calculating the window half-width Δτ is: Δτ = 0.2·τ 理论 +5ms; this allows for a wider search window as the depth and latency increase, thus better adapting to situations where deep signal propagation paths are more complex and uncertain.

[0067] Within the time delay search window, the narrowband correlation coefficient is calculated in the characteristic frequency band of each reflection peak and vibration abrupt change pair, and the narrowband correlation coefficient is energy-weighted and synthesized, with its maximum value being used as the acoustic-vibration time delay correlation.

[0068] For example, first calculate the narrowband correlation coefficient spectrum, then calculate the frequency weight w(f) based on the signal energy distribution, and finally obtain the correlation coefficient ρ. 综合 (τ)= ∑ f We find the maximum value of w(f) ·ρ(f, τ) as the final acoustic-vibration time delay correlation ρ. max This describes how to quantitatively calculate the correlation strength between sound wave reflection and vibration abrupt changes. Here, ρ(f, τ) is the narrowband correlation coefficient at frequency f and time delay τ. Optionally, in addition to calculating the energy-weighted correlation coefficient, the phase consistency or mutual information of the two signals in the time-frequency domain can also be calculated as alternative or supplementary indicators to measure the degree of correlation.

[0069] Based on the calculated correlation between acoustic and vibration time delays, the intensity of each reflection peak in the preliminary acoustic reflection characteristics is weighted and corrected to suppress false features and enhance true features, thus obtaining corrected reflection characteristics.

[0070] Specifically, the intensity of each reflection peak in the preliminary acoustic wave reflection characteristics is weighted and corrected to obtain the corrected reflection characteristics, including:

[0071] For each reflection peak in the preliminary acoustic wave reflection characteristics, a pseudo-peak discrimination index is calculated by combining the acoustic vibration time delay correlation, spectral bandwidth and phase continuity of each reflection peak, so as to identify the pseudo-peaks generated at the boundary of cascaded processing.

[0072] In this embodiment, a pseudo-peak is a peak that appears in the signal processing result in a shape similar to a real reflection, but it is essentially generated by mathematical operations (such as boundary effects or algorithm artifacts) and does not correspond to any real physical event. For example, a reflection peak may simultaneously satisfy the following: acoustic vibration time delay correlation ρ max A peak is considered a spurious peak if it meets two or more of the following three conditions: <0.3; spectral bandwidth >100Hz (the bandwidth of a true reflection is usually <50Hz); and phase discontinuity >π / 2.

[0073] The S-shaped nonlinear weighting function takes the acoustic-vibration time delay correlation as input and applies a penalty factor in combination with the pseudo-peak discrimination index to generate an adaptive correction weight for each reflection peak.

[0074] Preferably, the sigmoid weighting function W(ρ) can be designed as: W(ρ) = 1 / (1+exp(-k(ρ-ρ0))); where ρ is the correlation between the input acoustic and vibration time delays, parameter k controls the steepness of the transition curve of the function (e.g., k=10), and ρ0 is the inflection point (e.g., ρ0=0.5). For reflection peaks identified as spurious peaks, an additional penalty factor (e.g., 0.1) can be applied to their weights to further increase the suppression effect. Optionally, the sigmoid nonlinear weighting function can be replaced by other functions with similar monotonically increasing characteristics, such as the hyperbolic tangent function (tanh), or a piecewise linear function customized based on empirical data.

[0075] The original intensity of each reflection peak in the preliminary acoustic wave reflection characteristics is multiplied by the adaptive correction weight, and a weighted correction is performed to obtain the corrected reflection characteristics.

[0076] For example, the corrected reflection intensity R 修正 (z) can be calculated using the following formula: R 修正 (z)=R 原始 (z) [1+α (W(ρ)-0.5)]; where, R 原始 (z) represents the original reflection intensity, and α is the enhancement coefficient (e.g., α = 2.5). This formula can enhance the true reflection peak with high correlation (W(ρ) → 1) while effectively suppressing the spurious peak with low correlation (W(ρ) → 0).

[0077] It should be noted that during the weighted correction of each reflection peak, the calculated acoustic vibration time delay correlation ρ max Each value was recorded. After verifying the correlation of all preliminary reflection peaks, these correlation values ​​were compiled into a data structure corresponding one-to-one with each reflection peak, namely the correlation weight matrix. This matrix will be passed to the subsequent multimodal feature fusion step as the direct basis for calculating the confidence weight of each pile length estimate.

[0078] This embodiment introduces an independent, physically constrained cross-validation dimension for judging the authenticity of acoustic detection results, which can accurately identify and suppress spurious reflection peaks generated by the signal processing algorithm itself. It utilizes the physical coupling effect that real physical interface reflections are inevitably accompanied by a predictable vibration response of the pile tip soil. By calculating the time-delay correlation coefficient between the acoustic reflection peak and subsequent vibration abrupt events, the system obtains an objective, quantitative reliability index: real reflection peaks have strong correlation (e.g., ρ>0.7), while algorithm artifacts are almost uncorrelated (e.g., ρ<0.3). Based on this, a weighted correction of the reflection intensity can selectively enhance the real signal and suppress spurious signals. This not only solves the boundary spurious peak problem introduced by cascaded projection processing (spurious peak identification accuracy >95%), but also overcomes the inherent defects of single signal source detection, such as difficulty in distinguishing between real and spurious signals and susceptibility to algorithm interference, thus improving the reliability of the detection results.

[0079] like Figure 2 As shown, according to one aspect of this application, the estimated pile length is calculated by the following steps:

[0080] The corrected reflection characteristics are analyzed to assess the signal energy attenuation state and obtain the energy attenuation characteristics.

[0081] For example, by calculating the sound wave energy attenuation curve E 声 (z) and vibrational energy curve E 振 (z) assess the energy decay rate and identify weak signal depth regions with insufficient energy (e.g., E 振 <0.1g 2 (area).

[0082] By combining energy attenuation characteristics with preliminary pile length estimation, intelligent hammering control parameters are optimized and generated.

[0083] In this embodiment, the intelligent hammer control parameters include at least:

[0084] Optimize hammering frequency: Based on the preliminary pile length estimation and the preset geophysical model, the target resonance frequency in the target depth range is estimated. Based on the target resonance frequency and energy attenuation characteristics, the optimized hammering frequency for sweeping around the target resonance frequency is calculated.

[0085] Specifically, for the target depth L target Calculate the fundamental resonant frequency f0=v(L) target ) / (4L target The target resonant frequency f is obtained by considering soil damping and making corrections. 共振 Then, a frequency modulation strategy is designed, for example, to make the hammer frequency f 锤 In f 共振 The scanning is performed within ±10% of the range to effectively excite a wideband resonant response.

[0086] Adaptive hammering interval: The adaptive hammering interval is adaptively adjusted and determined based on the signal energy strength reflected by the energy attenuation characteristics.

[0087] For example, it can be based on the vibrational energy E 振 The hammering interval T is calculated by the ratio of the energy to the reference energy E0. 间隔 :T 间隔 =T0+ΔT×(1-E 振 / E0); where T0 is the baseline interval (e.g., 2 seconds) and ΔT is the maximum adjustment (e.g., 0.5 seconds). This strategy allows for a longer hammer interval when the signal energy is weaker, thus accumulating more energy for the next excitation.

[0088] Dynamic hammering force: Taking into account the target depth and energy attenuation characteristics, the dynamic hammering force is dynamically adjusted and determined.

[0089] For example, the hammering force F can be adjusted by a coefficient that combines the effects of depth and energy attenuation: F = F 标准 ×[1+β 深度 ×(L / 10)+β 衰减 [×(1-E / E0)]; where F 标准 This is the standard hammering force, L is the target depth, and β is... 深度 and β 衰减 These are adjustment coefficients (e.g., 0.3 and 0.5 respectively). This strategy effectively enhances deep signals while avoiding signal overload on shallow formations.

[0090] Based on the intelligent hammering control parameters, the electric hammering system performs directional hammering excitation and simultaneously collects the enhanced signal after resonance enhancement.

[0091] In this embodiment, intelligent hammering control parameters (including optimized hammering frequency, adaptive hammering interval, and dynamic hammering force) are loaded into the PLC (Programmable Logic Controller) of the electric hammering system. The electric hammering system precisely controls the hammering frequency (e.g., with a control accuracy of ±0.1Hz) through a servo motor and precisely adjusts the hammering force (e.g., with a control accuracy of ±5%) through a hydraulic system, thereby enabling high-fidelity physical execution of the enhanced excitation command.

[0092] Within a critical time window following each enhanced hammer strike (e.g., before and after the expected response time calculated based on a preliminary estimate of the target depth), the signal processing unit synchronously and rapidly acquires acoustic and vibration signals at a higher sampling rate (e.g., 50 kHz). Fast Fourier Transform (FFT) analysis is performed on the real-time acquired response data to monitor whether the target resonance peaks appear as expected and their intensity changes. Furthermore, the system calculates two key enhancement effect evaluation metrics: signal-to-energy ratio G: G=E 增强 / E 原始 Peak-to-peak ratio P: P = P 增强 / P 原始 Among them, E 增强 and P 增强 This refers to the energy and key spectral peaks of the signal after this excitation, E. 原始 and P 原始 This is the baseline value before enhancement.

[0093] The system makes an immediate decision based on the enhancement effect evaluation index obtained in the previous step. If the evaluation result shows G < 1.5 or P < 2, the system determines that the actual effect of this enhancement excitation has failed to achieve the preset enhancement target. In this case, the system will automatically fine-tune the hammering control parameters (e.g., fine-tune the hammering frequency by ±2Hz, or fine-tune the hammering force by ±10%), and immediately instruct the electronically controlled hammering system to perform a secondary excitation. The process of evaluation-decision-fine-tuning-re-excitation constitutes a fast-response closed-loop feedback control. Closed-loop feedback control is a type of control system in which the system output (signal enhancement effect) is continuously measured and fed back to the input, compared with the desired target, and the difference is used to adjust the control action (hammering parameters), so that the system output can automatically tend towards the desired target. This cycle can be repeated until the enhancement effect is detected to be up to standard (i.e., both G and P exceed their thresholds), or the preset maximum number of attempts is reached. Optionally, in addition to using simple threshold judgment and fixed step fine-tuning, more mature automatic control algorithms, such as proportional-integral-derivative (PID) controllers, can also be used. The PID controller can dynamically calculate the optimal hammering parameter adjustment based on the magnitude of the error between the enhancement effect and the target, the cumulative error, and the rate of change of error, thereby achieving faster and more stable closed-loop control.

[0094] After the closed-loop control confirms that the enhancement effect meets the target, the system performs a complete, long-term signal acquisition to obtain the final enhanced acoustic and vibration signals. It is important to note that the preprocessing performed on the enhanced signals is specially adjusted. Specifically, the filtering step employs a protective bandpass strategy to ensure that the newly excited valuable signal components located within the target resonant frequency band (e.g., target resonant frequency ±20Hz) are not accidentally attenuated or filtered out during noise removal. The final output includes enhanced reflection and vibration characteristics.

[0095] By comprehensively considering the corrected reflection characteristics, enhanced signals, and correlation weight matrices inherited from acoustic-vibration coupling analysis, multimodal fusion is performed using an acoustic-vibration resonance model to calculate the final estimated pile length. Multimodal fusion is an information processing technique that aims to obtain more accurate, robust, and comprehensive conclusions than any single mode by integrating data or analysis results from multiple different sensors or different physical dimensions (modalities).

[0096] This embodiment revolutionizes the traditional passive detection paradigm into an active, closed-loop feedback control intelligent detection system, effectively solving the detection blind zone problem caused by severe signal attenuation at deep depths. Unlike existing technologies that use fixed-parameter hammering, this embodiment assesses energy attenuation in the initial detection results, actively identifying target depth ranges where the signal is weak. Instead of abandoning the target, it precisely calculates the optimal hammering parameters (frequency, interval, force) to achieve energy focusing based on the resonance characteristics of that depth. This customized excitation is then executed by the electronically controlled hammering system, achieving directional and resonant enhancement of the target depth signal. This closed-loop control of detection-evaluation-optimization-re-excitation improves the detection success rate and expands the effective working range of dynamic detection methods.

[0097] According to one aspect of this application, a final estimated pile length is calculated by multimodal fusion using an acoustic-vibration resonance model, including:

[0098] Based on the modified reflection characteristics and enhanced signal, a set of multi-source pile length estimates containing independent physical meanings is calculated and generated through acoustic wave reflection, vibration abrupt change, and acoustic resonance relationship models. For example, this set may include: an estimate L based on acoustic wave reflection time delay. 声 The estimated value L based on the vibration change time delay 振 ; Estimated value L based on resonant frequency 共振 The estimated value L based on energy decay characteristics 能量 .

[0099] Based on the correlation weight matrix inherited from the acoustic-vibration coupling analysis, a corresponding confidence weight is calculated for each estimated value in the multi-source pile length estimation set, forming a confidence weight set. Preferably, the confidence weight wi Based on the acoustic vibration time delay correlation coefficient ρ i The calculation yields: w i =ρ i 2 / ∑ j ρ j 2 Among them, the estimates with higher relevance are given greater weight.

[0100] Weighted least squares fusion is performed on the multi-source pile length estimates using a set of confidence weights to obtain the final pile length estimate. The detection confidence level is then calculated based on the residuals after weighted fusion. The final pile length estimate L... final The calculation formula is: L final =∑ i w i L i Simultaneously, the weighted standard deviation σ can be calculated as a measure of confidence level, thus providing reliable data support for engineering decisions. Alternatively, in addition to using the weighted least squares method, a Kalman filter framework can be employed, treating the pile length estimate as a state variable updated with each hammer blow. Alternatively, a Bayesian inference method can be used to fuse the various estimates as evidence with uncertainty in a probabilistic sense.

[0101] This embodiment improves the accuracy and confidence level of the final pile length estimate by intelligently fusing information from multiple sources and multiple physical dimensions. Instead of obtaining an estimate from a single physical phenomenon (such as sound wave transit time), it simultaneously calculates multiple independent pile length estimates from various physical dimensions, including sound wave reflection, vibration abrupt changes, and acoustic resonance. Instead of simply averaging these estimates, it uses a correlation weight matrix as an objective and quantifiable confidence index. Estimates with stronger correlation are considered to have higher physical authenticity and are therefore given greater weight in the final weighted fusion calculation. This ensures that the final result is most likely to align with the most reliable physical evidence, effectively avoiding the systematic bias and random errors of single detection methods, improving the accuracy of the final pile length estimate, and outputting a clear detection confidence level, providing data support for engineering decisions.

[0102] In one specific embodiment, the correlation between acoustic vibration time delay is utilized to distinguish between reflection peaks and pseudo-reflection peaks generated by the algorithm processing, and these are then enhanced and suppressed respectively. In this case, the following initial conditions are set: In the preliminary acoustic wave reflection characteristics, two reflection peaks from different depths are considered: Reflection peak A: A reflection from a real geological interface located at a deeper depth, with a depth L... A =15.2 meters, initial reflection intensity R 原始，A =0.8. Reflection peak BA suspected spurious peak located at the boundary of the cascaded projection processing (in the 9-10 meter range), with a depth L B =9.5 meters, initial reflection intensity R 原始，B =0.5. Available data also includes: real-time wave velocity field v(z), and a sequence of vibration abrupt change times extracted from the vibration signal. For a depth of L... B The reflection peak B is 9.5 meters deep. By integrating the real-time wave velocity field v(z) along the depth path, the two-way propagation time of the sound wave is calculated to be 18.0 ms. Considering the inherent system delay t... 系统 =3.5ms, the final theoretical propagation delay τ 理论，B For: τ 理论，B =18.0ms + 3.5ms = 21.5ms.

[0103] According to the theoretical time delay τ 理论，B Calculate the half-width Δτ of the adaptive search window. B ΔτB = 0.2 × τ 理论，B +5ms = 0.2 × 21.5ms + 5ms = 4.3ms + 5ms = 9.3ms; therefore, the time delay search window for finding the corresponding vibration event is [21.5ms - 9.3ms, 21.5ms + 9.3ms], i.e., [12.2ms, 30.8ms]. Within this time delay search window, the characteristics of reflection peak B are correlated with the vibration abrupt change time sequence. The calculations show that no significant vibration event corresponds to this reflection peak in terms of time and physical cause. The final maximum energy-weighted comprehensive correlation coefficient ρ is obtained. max，B Only 0.25. The reflection peak B was evaluated based on the pseudo-peak discrimination index: Index 1 (correlation): ρ max，B =0.25, less than the threshold of 0.3, thus meeting this condition. Indicator 2 (Spectral Bandwidth): Spectral analysis shows the peak's bandwidth is 110Hz, greater than the threshold of 100Hz, thus meeting this condition. Indicator 3 (Phase Continuity): Phase analysis shows a large phase discontinuity, thus meeting this condition. Since multiple spurious peak discrimination conditions are met, reflection peak B is highly certain to be a spurious peak generated during processing. Its corrected weight is calculated using the S-shaped nonlinear weighting function W(ρ), where parameter k=10 and ρ0=0.5. W(ρ B =W(0.25)=1 / (1+exp(-10(0.25-0.5)))=1 / (1+exp(2.5))≈0.076; After obtaining the weights, calculate the corrected reflection intensity R according to the correction formula. 修正，B Where the enhancement coefficient α = 2.5: R 修正，B =R 原始，B ×[1+α×(W(ρ B)-0.5)]=0.5×[1+2.5×(0.076-0.5)]=0.5×[1-1.06]=-0.03. The calculation results show that the original intensity of reflection peak B, 0.5, is effectively suppressed to near zero, and thus basically eliminated in the corrected reflection characteristics.

[0104] In contrast, for a depth of L A The same processing procedure is applied to the actual reflection peak A, which is 15.2 meters in diameter. Assume the calculated theoretical time delay τ... 理论，A =35.0ms. Window half-width Δτ A =0.2·35.0ms+5ms=12.0ms, the search window is [23.0ms, 47.0ms]. Within this window, a corresponding strong abrupt change in pile end vibration was found, and the maximum correlation coefficient ρ was calculated. max，A =0.85. This correlation coefficient is much greater than 0.7, which does not meet the criteria for identifying a false peak, and it is confirmed as a true reflection peak. Weight calculation and weighted correction are then performed: W(ρ A )=W(0.85)=1 / (1+exp(-10(0.85-0.5)))=1 / (1+exp(-3.5))≈0.97; R 修正，A =R 原始，A ×[1+α×(W(ρ A The calculation results show that the original intensity of the true reflection peak A, 0.8, is enhanced to 1.74, making it more prominent in the feature map. It can be seen that this embodiment can distinguish between real physical reflection and algorithmic artifacts through quantitative physical correlation analysis, and apply diametrically opposed correction effects to both: suppressing the pseudo-peak by more than 90% and enhancing the real signal by more than 100%. This makes the final corrected reflection features cleaner and more reliable, laying a solid foundation for subsequent high-precision pile length estimation.

[0105] According to one aspect of this application, in actual engineering projects, information such as borehole columnar sections and test data obtained directly from geological exploration units are diverse in format and have discrete data points, making them unsuitable for the adaptive analysis requiring continuous physical parameter input as described in this application. Therefore, it is necessary to obtain prior formation information, including:

[0106] Geological exploration borehole data, standard penetration test data, and wave velocity test data are read, and a stratigraphic reference model containing expected wave velocity, density, and impedance parameters is constructed through interpolation and data fusion processing to serve as the stratigraphic precursor information.

[0107] In this embodiment, the stratigraphic reference model is a structured, digitized three-dimensional geological body data model. It stores the expected distribution of geophysical parameters (such as density and wave velocity) within the target area in the form of a continuous data field by spatially interpolating and fusing discrete exploration data. The signal processing unit is first configured to read raw geological exploration data from multiple sources. This data may include, but is not limited to: geological exploration borehole data: typically borehole columnar diagrams or textual descriptions depicting the lithology, thickness, color, and state of soil layers at different depths at a specific location; Standard Penetration Test (SPT) data: providing the SPT blow count (N value) at different depths, which can be used to indirectly assess the compaction and strength of the soil; and wave velocity test data: acoustic or shear wave test results from the field or adjacent sites, providing P-wave or S-wave velocities at specific depths. After reading the discrete borehole data, the signal processing unit expands this point-like or linear information into a continuous three-dimensional stratigraphic model using a spatial interpolation algorithm. Preferably, geostatistical methods such as Kriging interpolation or inverse distance weighted interpolation can be used. This step aims to infer the distribution of soil and rock layers at any location and depth within the entire area where the engineering piles are located, based on limited borehole data. Optionally, in addition to Kriging interpolation and inverse distance weighted interpolation, other modeling methods commonly used in geostatistics, such as sequential Gaussian simulation or triangular mesh irregular grid (TIN) models, can be used to accommodate different types and distributions of geological data.

[0108] Furthermore, based on the three-dimensional stratigraphic model, the signal processing unit further integrates various experimental data to calculate key physical parameters for each grid point in the model. Specifically, the unit can utilize empirical formulas or data fusion algorithms from the field of geotechnical engineering. For example, based on the N value and soil type from the Standard Penetration Test (SPT), it can estimate the density ρ and initial wave velocity v at that point. If direct wave velocity test data is available, it is prioritized and locally corrected. Based on this, the acoustic impedance Z = ρ × v can be further calculated. After the above processing, a structured stratigraphic reference model that can be directly queried and called by computer programs is finally output. This model, in the form of a three-dimensional data field, stores the expected wave velocity, density, and acoustic impedance, among other physical parameters, which continuously vary with spatial location and depth throughout the target area. As a standardized set of stratigraphic prior information, this model provides initial, depth-varying physical parameter inputs for the generation of the depth-adaptive atomic dictionary, forming the basis for realizing physics-driven adaptive analysis.

[0109] According to one aspect of this application, parameters in a general theoretical physical model (e.g., parameters reflecting soil damping characteristics) can vary due to differences in soil conditions at specific construction sites. Directly using theoretical values ​​can introduce systematic errors into the final pile length estimation. Therefore, by utilizing high signal-to-noise ratio measured data from the field for online model calibration, the model's applicability to specific cases and the accuracy of the final estimation are improved. Online calibration is a model parameter optimization process that uses a high-reliability subset of data collected in real-time from the actual operation site to adjust the parameters of a general theoretical model, enabling it to better reflect the personalized physical characteristics of the specific object. Specifically, before fusing the information before and after enhancement, the process also includes:

[0110] By combining the modified reflection characteristics and vibration abrupt change characteristics, the parameters of the preset acoustic-vibration resonance relationship are calibrated using the least squares method to generate the calibrated acoustic-vibration resonance model.

[0111] In this embodiment, the signal processing unit filters and prepares a high-confidence dataset for model calibration from existing analysis results. Preferably, the high-confidence dataset comes from the detection results of the shallow part of the pile (e.g., less than 10 meters deep). The reason for choosing shallow data is that the signal-to-noise ratio of the acoustic and vibration signals at this depth is high, the energy attenuation is small, and the identified corrected reflection feature positions are accurate, which can be regarded as the true labels or anchors of this calibration process. This dataset specifically includes: the accurate depth L of each true reflection peak in the shallow part, the corresponding vibration abrupt change characteristics, and the resonance frequency f resonant based on the vibration signal analysis. In some large projects, optionally, one or several dedicated test piles can be used for model calibration. The length and geological conditions of these test piles are precisely controlled or measured, and their detection data can provide a more reliable and unified calibration model for the subsequent detection of all engineering piles.

[0112] The least squares method was used for parameter calibration. The signal processing unit adopted a preset acoustic-vibration resonance relationship, which established the pile length L, wave velocity c, and resonant frequency f. 共振 And the relationship between parameters such as acoustic vibration time delay correlation ρ. An exemplary relationship is: L=(c / 4f 共振 The formula is: β × (1 - β × ρ); where β is an undetermined coefficient reflecting the influence of soil on the acoustic-vibration coupling effect, and its theoretical value is difficult to predict accurately. Therefore, β is treated as an unknown, and the optimal value of parameter β that best matches the current site conditions is obtained by fitting a high-reliability calibration dataset using the least squares method. After the parameter solution is completed, the system will obtain a calibrated acoustic-vibration resonance model containing the optimized parameter β after online calibration. This model is then used in the multimodal feature fusion step as an estimate of the pile length (i.e., the estimate L based on the resonance frequency) for calculating one of the pile length estimates. 共振This provides the basis for the estimation of this path, ensuring its accuracy and thus providing a strong guarantee for the overall accuracy of the final fusion result.

[0113] According to one aspect of this application, at the pile foundation construction site, the raw signals acquired by sensors are inevitably affected by strong environmental noise interference and their own hardware limitations. Without effective preprocessing, these defects will severely affect the accuracy of all subsequent analysis results. Specifically, refined preprocessing of the raw signals includes:

[0114] After receiving the raw, time-synchronized signal set acquired by the sensor assembly, the signal processing unit first performs a comprehensive quality assessment on the signal of each channel.

[0115] Specifically, the evaluation process includes calculating key quality metrics for each signal channel, such as: Signal-to-Noise Ratio (SNR): assessing the relative strength of the effective component and noise in the signal; Zero-drift: detecting whether the signal baseline undergoes an undesirable slow shift over time; and Saturation: checking whether the signal amplitude exceeds the sensor's range limit, causing the top or bottom to be clipped. Based on these evaluation metrics, the system automatically identifies and marks various anomalies, including but not limited to sensor hardware failures, signal saturation caused by excessive single hammer impact, and abnormal spikes caused by sudden electromagnetic interference. Two key intermediate data sets are output: a signal quality evaluation report and a valid signal period marker. These two sets of data will guide subsequent filtering and analysis steps, ensuring the algorithm operates only on high-quality, valid signal segments and avoiding interference from contaminated data in the final results.

[0116] After completing the quality assessment, the signal processing unit performs adaptive noise filtering and baseline correction on the valid signal segments based on the valid signal time period markings.

[0117] Unlike traditional filters that use fixed parameters, this embodiment recognizes that noise at the construction site (such as power frequency interference and other mechanical vibration noise) is not constant in the time domain. Therefore, an adaptive filtering strategy is crucial. This embodiment employs an adaptive filter based on the local statistical characteristics of the signal. This filter can analyze the statistical characteristics of the signal (such as variance and kurtosis) in different time windows in real time, and automatically adjust its filtering parameters (such as filter order and cutoff frequency) based on these statistical characteristics. This allows the filter to effectively remove various types of noise while preserving as much useful weak signal detail as possible from reflections from the formation interface. The adaptive filter is a digital filter whose transfer function (filter parameters) can be self-adjusted according to the statistical characteristics of the input signal through an optimization algorithm, enabling it to achieve optimal filtering performance in non-stationary noise environments. After filtering, zero-point drift compensation and baseline correction are performed on the signal to accurately correct the signal baseline to zero. After all the above preprocessing steps, the system finally outputs two high-quality data streams: a preprocessed acoustic signal and a preprocessed vibration signal. These two signals are the direct inputs for all subsequent core analyses. In complex scenarios with multiple independent noise sources, blind source separation (BSS) techniques, such as independent component analysis (ICA), can optionally be used to separate and remove noise. Alternatively, a deep learning denoising autoencoder model can be pre-trained for nonlinear denoising of the signal.

[0118] According to one aspect of this application, after extracting preliminary acoustic wave reflection features, the method further includes:

[0119] The preliminary acoustic reflection characteristics are analyzed to identify abrupt changes in the reflection coefficient, and the depth of the deepest significant abrupt change is used as a preliminary estimate of the pile length.

[0120] In this embodiment, the continuous projection signal is transformed into an intuitive, depth-varying reflection intensity function R(z). Specifically, the signal processing unit extracts the signal amplitude corresponding to each depth point z from the continuous projection signal y(z), and this amplitude is regarded as the reflected wave amplitude A at that depth. 反射 (z). Simultaneously, the initial hammer impact signal amplitude measured at the pile top is acquired as the reference incident wave amplitude A. 入射 The reflection coefficient at each depth point is calculated using the following formula, thus constructing a continuous depth-reflection intensity mapping function R(z): R(z) = |A| 入射 |A 反射The function R(z) intuitively reflects the reflection of sound wave energy at different depths. To accurately identify reflections caused by significant stratigraphic interfaces from a continuous R(z) function, a screening method based on both amplitude and gradient criteria is adopted. Using only the amplitude criterion may misclassify some wide, non-abrupt reflection regions as interfaces; using only the gradient criterion may be sensitive to noise. Combining the two criteria ensures that the identified abrupt change points have both sufficient reflection energy and clear interface characteristics. Specifically, the signal processing unit traverses all points on the R(z) curve. A point is only identified as a significant abrupt change point if it simultaneously meets the following two conditions: the reflection intensity value of the point must be greater than a preset minimum reflection threshold, i.e., R(z) > 0.15. This aims to filter out invalid reflections caused by minor stratigraphic inhomogeneities or background noise. The depth gradient of the reflection intensity at the point must be greater than a preset gradient threshold, i.e., ΞzΞR > threshold G. The depth gradient ΞR / Ξz can be numerically calculated by applying the first-order finite difference method to the R(z) function. The aim is to ensure that the selected point is a sharp, abrupt peak, rather than a gentle energy bulge. Alternatively, more complex peak detection algorithms from signal processing can be employed, such as using continuous wavelet transforms to find ridges in scale maps at different scales, or applying the top-hat transform from mathematical morphology to highlight sharp peaks, thus achieving more robust abrupt change detection.

[0121] After selecting the set of all significant abrupt change points using the aforementioned dual criteria, the following algorithm is executed to generate the final preliminary pile length estimate. A variable L is defined. preliminary And its initial value is set to 0. Traverse all data points identified as significant mutation points. During the traversal, the depth z corresponding to each significant mutation point is... point With variable L preliminary Compare with the current value of z. point >L preliminary Then update L preliminary =z point After the iteration is complete, variable L... preliminary The value represents the maximum depth among all significant abrupt change points. This value is output as a preliminary pile length estimate and passed to the subsequent energy-adaptive mutual-excitation enhancement module for calculating and optimizing smart hammering parameters.

[0122] According to one aspect of this application, generating the modified reflection features also includes:

[0123] The magnitude of the acoustic vibration time delay correlation calculated for each preliminary acoustic wave reflection feature is compiled into a correlation weight matrix.

[0124] In this embodiment, the correlation weight matrix is ​​a structured data array that stores a corresponding acoustic-vibration time-delay correlation coefficient value that quantifies the physical authenticity of each initially identified reflection peak. Functionally, it is equivalent to a confidence lookup table that can be queried by subsequent fusion algorithms. Specifically, the correlation weight matrix is ​​a two-dimensional data structure with a size of N×2, where N is the total number of reflection peaks identified from the initial acoustic reflection features. Matrix dimensions: N rows, 2 columns. Element meaning: First column (index / key): stores a unique identifier for each reflection peak; preferably, the depth value z corresponding to each reflection peak is used. i (where i = 1, 2, ..., N) serves as its unique identifier; the second column (value): stores the reflection peak corresponding to the identifier in the first column, in the final acoustic-vibration time delay correlation coefficient ρ max，i Alternatively, in addition to using an N×2 two-dimensional matrix, a hash table (HashMap) or a dictionary object can also be used. This key-value pair structure, with the reflection peak depth as the key and the correlation coefficient as the value, usually has higher computational efficiency when querying data.

[0125] The signal processing unit executes the following algorithm to construct the matrix: obtain a depth list of all N reflection peaks {z1, z2, ..., z...} N Based on this, an empty matrix of size N×2 is created and named CorrWeightMatrix. The system starts a loop, iterating from i=1 to N. In each loop, the system performs the following operations: The depth z of the current reflection peak is... i Fill the matrix to the i-th row and 1-th column, i.e., CorrWeightMatrix[i, 0] = z i According to depth z i As an index, the acoustic-vibration time delay correlation coefficient ρ corresponding to the reflection peak at that depth is queried and retrieved. max，i The retrieved relevance coefficient value ρ max，i Fill the matrix into the i-th row and 2-th column, i.e., CorrWeightMatrix[i, 1] = ρ max，i After the loop ends, the CorrWeightMatrix is ​​fully populated. This matrix, as structured data, is then output as the correlation weight matrix. The correlation weight matrix is ​​subsequently passed to the multimodal fusion module to calculate the corresponding confidence weights for different pile length estimates, ensuring a complete closed loop in the data flow.

[0126] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for estimating the length of engineering piles, characterized in that, include: Acquire prior information about the formation that reflects the impact response of the pile hammer, as well as multi-source detection signals including acoustic and vibration components; Based on prior information about the formation, adaptive analysis of the acoustic wave components is performed to extract preliminary acoustic wave reflection characteristics. By using vibration components, the preliminary acoustic wave reflection characteristics are correlated and verified, spurious features are suppressed, and corrected reflection characteristics are generated. The signal quality of the corrected reflection characteristics is evaluated, and intelligent hammering control commands are generated accordingly. Enhanced signal excitation is executed, and information before and after enhancement is fused to calculate the estimated pile length.

2. The method according to claim 1, characterized in that, The estimated pile length is calculated, including: Based on the corrected reflection characteristics, the signal energy attenuation state is evaluated, and the energy attenuation characteristics are obtained. By combining energy attenuation characteristics with a preset preliminary pile length estimate, intelligent hammering control parameters are optimized and generated. Directional hammering is performed based on intelligent hammering control parameters, and the enhanced signal after resonance enhancement is acquired simultaneously. By comprehensively correcting the reflection characteristics, enhancing the signal, and using the pre-defined correlation weight matrix inherited from the acoustic-vibration coupling analysis, multi-modal fusion is performed through the acoustic-vibration resonance model to calculate the final estimated pile length.

3. The method according to claim 2, characterized in that, The intelligent hammer control parameters shall include at least the following three parameters: Based on the preliminary pile length estimation and the preset geotechnical model, the target resonance frequency in the target depth range is estimated. Combined with the energy attenuation characteristics, the optimized hammering frequency for sweeping around the target resonance frequency is calculated. Based on the signal energy strength reflected by the energy attenuation characteristics, the adaptive hammering interval is adaptively adjusted and determined. Taking into account both the target depth and energy attenuation characteristics, the dynamic hammering force is dynamically adjusted and determined.

4. The method according to claim 2, characterized in that, Multimodal fusion was performed using an acoustic-vibration resonance model to calculate the final estimated pile length, including: Based on the modified reflection characteristics and enhanced signal, a set of multi-source pile length estimates containing independent physical meanings is calculated and generated through acoustic wave reflection, vibration abrupt change and acoustic resonance relationship models. Based on the correlation weight matrix, the corresponding confidence weight is calculated for each estimated value in the multi-source pile length estimation set, thus forming a confidence weight set; The weighted least squares fusion of the multi-source pile length estimates is performed using the confidence weight set to obtain the final pile length estimate.

5. The method according to claim 1, characterized in that, Generate corrected reflection features, including: Detect spectral abrupt changes in vibration components and extract their corresponding time-frequency features to form vibration abrupt change features; For each reflection peak in the preliminary acoustic wave reflection characteristics, calculate the acoustic-vibration time delay correlation between it and the vibration abrupt change characteristics; Based on the magnitude of the acoustic vibration time delay correlation, the intensity of each reflection peak in the preliminary acoustic wave reflection characteristics is weighted and corrected to suppress false features and enhance true features, thus obtaining the corrected reflection characteristics.

6. The method according to claim 5, characterized in that, Calculating the correlation between acoustic and vibration time delays includes: For each reflection peak, the theoretical propagation delay is calculated by integrating the real-time wave velocity that varies with depth along its depth path. Set a delay search window for each theoretical propagation delay, where the size of the delay search window is proportional to the value of the theoretical propagation delay; Within the time delay search window, the narrowband correlation coefficient is calculated in the characteristic frequency band of each reflection peak and vibration abrupt change pair, and then energy-weighted synthesis is performed on it. The maximum value of the energy-weighted synthesis result is taken as the acoustic-vibration time delay correlation.

7. The method according to claim 5, characterized in that, The intensity of each reflection peak in the preliminary acoustic wave reflection characteristics is weighted and corrected to obtain the corrected reflection characteristics, including: For each reflection peak in the initial acoustic wave reflection characteristics, a pseudo-peak discrimination index is calculated by comprehensively considering its acoustic vibration time delay correlation, spectral bandwidth, and phase continuity, and pseudo-peaks generated at the boundary of cascaded processing are identified. Based on the correlation between pseudo-peaks and acoustic vibration time delay, an adaptive correction weight is generated for each reflection peak through an S-shaped nonlinear weighting function. The original intensity of each reflection peak in the preliminary acoustic wave reflection characteristics is multiplied by the adaptive correction weight to obtain the corrected reflection characteristics.

8. The method according to claim 1, characterized in that, Extract preliminary acoustic wave reflection features, including: The sound wave components are divided into segmented signal sequences with overlapping beginnings and ends along the depth dimension; Configure a local projection operator for each signal segment, and perform cascaded projection calculations on the segmented signal sequence in sequence to obtain the projection result; wherein, the projection calculation of the later signal segment includes the processing residual generated by the projection calculation of the previous segment; A weighted fusion method is used to smooth the overlapping areas of adjacent projection results to generate a continuous projection signal; Multi-scale reflection information is extracted and assembled from continuous projection signals to form preliminary acoustic wave reflection characteristics.

9. The method according to claim 8, characterized in that, Generating a continuous projection signal includes: For any data point within the overlapping region, a position-related weight is calculated based on the relative position of the data point within the overlapping region; the closer the data point is to a preset endpoint, the greater the weight of the segment result containing that endpoint. The corresponding values ​​of adjacent projection results in the overlapping area are multiplied by the position-related weights and summed to obtain the fused value of the data point in the continuous projection signal.

10. A system for predicting the length of engineering piles, characterized in that, include: The sensor assembly is configured to acquire multi-source detection signals that reflect the impact response of the pile hammer; The electrically controlled hammering system is configured to receive control commands and execute hammering actions with adjustable parameters; A signal processing unit, operatively coupled to the sensor assembly and the electrically controlled hammer impact system, is configured as follows: Based on the acoustic wave components in the multi-source detection signals obtained from the sensor assembly and the pre-stored a priori information of the formation, preliminary acoustic wave reflection characteristics are extracted. By utilizing the vibration components in the multi-source detection signals, the preliminary acoustic wave reflection characteristics are correlated and verified to generate corrected reflection characteristics; The modified reflection characteristics are evaluated, and based on this, intelligent hammering control commands are generated and sent to the electronically controlled hammering system to execute enhanced hammering excitation. By integrating information before and after enhanced hammering, the final estimated pile length is calculated.

Images