Intelligent test method and system for mechanical parameters of foundation pit solidified soil based on impact response

Abstract

The application discloses a foundation pit solidified soil mechanics parameter intelligent test system based on impact response, relates to the geotechnical engineering test technical field, and is used for solving the problem that the existing method generally assumes that the soil body is an isotropic layered medium and is difficult to process the anisotropy problem commonly existing in the solidified soil body. Interference of an excitation and a sensing contact link is reduced through incident deconvolution and coupling correction. On a k-ω azimuthal dispersion surface, weighting is performed according to a joint weight of signal-to-noise ratio, modal degree and phase closed calculation, so as to stabilize dispersion extraction and modal separation. A persistent coherence is used to select a candidate layer position, and L2,1 group sparsity and depth TV selection are combined with cross R / L and multi-azimuth sharing, so that the number of layers and layer boundary data are driven to be determined and controlled and constrained. In the layered anisotropy joint inversion, a distinguishability penalty dominated by a sensitivity kernel and a multi-start Gauss-Newton are introduced, and after parameters and the number of layers are obtained by using RJ-MCMC, parameter estimation and uncertainty quantification are formed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geotechnical engineering testing, more specifically, the present application relates to an intelligent test method and system for foundation pit solidified soil mechanical parameters based on impact response. BACKGROUND

[0002] Foundation pit engineering plays an important role in urban construction, and its stability is directly related to the safety of the surrounding environment and structures. In order to ensure the stability of the foundation pit, it is often necessary to solidify the soil around the foundation pit, such as cement mixing piles, high-pressure jet grouting piles, etc. Accurate and efficient testing and evaluation of the mechanical parameters of the solidified soil is a key link in the design, construction and safety monitoring of the foundation pit.

[0003] Traditional soil mechanical parameter testing methods, such as laboratory tests (triaxial test, direct shear test) and field tests (standard penetration test, static cone penetration test, wave velocity test, etc.), have many limitations. Laboratory tests are difficult to sample and the sample is disturbed, making it difficult to truly reflect the mechanical properties of the field soil; field tests can reflect the in-situ properties, but usually require a lot of time and effort, and have high requirements for the testing environment, and it is difficult to achieve comprehensive and detailed evaluation of the complex mechanical parameters of the solidified soil. In particular, for solidified soil, its mechanical properties may exhibit significant anisotropy, and there are differences in different depths and directions, making it difficult for traditional methods to effectively capture these complex characteristics.

[0004] The existing technology has the following shortcomings:

[0005] Wave velocity testing methods based on impact response, such as the surface wave method (MASW), can to some extent obtain the shear wave velocity information of the soil. However, these methods usually assume that the soil is an isotropic layered medium, making it difficult to handle the anisotropy problem that is common in solidified soil. In addition, existing methods often process or simply weight and fuse data from different wave modes (such as Rayleigh waves and Love waves), different frequencies, and different directions independently, resulting in information loss and uncertainty in the inversion results. In the presence of problems such as unknown number of layers, uneven parameter identifiability, and complex field conditions (such as weak lateral variation and slight inclination), existing methods often face ill-conditioning, resulting in unstable and inaccurate inversion results, making it difficult to meet the needs of foundation pit engineering for detailed and intelligent evaluation of mechanical parameters. Therefore, how to develop an intelligent testing system and method that can comprehensively, accurately and efficiently obtain the complex mechanical parameters of the solidified soil in the foundation pit and effectively solve the problem of inversion ill-conditioning is a technical problem that needs to be solved.

[0006] In view of the above problems, the present application provides a solution. SUMMARY

[0007] In order to overcome the above-mentioned defects of the prior art, embodiments of the present application provide an impact response-based foundation pit solidified soil mechanics parameter intelligent testing system to solve the problems raised in the above background art.

[0008] To achieve the above object, the present application provides the following technical solutions:

[0009] The impact response-based foundation pit solidified soil mechanics parameter intelligent testing system comprises the following steps:

[0010] The impact excitation is triggered along the azimuth angle set Θ within the test point domain Ωx, and three-direction acceleration time histories are acquired by the multi-channel seismic array, deconvolution and analog-digital processing are performed, unit impulse responses and their time-frequency representations and R / L dispersion curves are obtained, and the triggered force spectrum is recorded synchronously;

[0011] Two-dimensional fast Fourier transform is performed on the unit impulse responses to obtain k-ω spectrum Azimuth ring method and ridge line tracking are combined with phase continuity to decompose modes, dispersion surfaces are constructed according to azimuth, and observation weights w are generated based on signal-to-noise ratio, mode separability and phase closure statistics;

[0012] The mode intersection of the azimuth dispersion surface in the stable frequency band and the sensitivity core extreme value are used to determine the candidate interface set Zcand, topological robustness is evaluated by persistent homology, depth is discretized and a shear modulus jump variable is set, group sparse optimization shared across R / L and all azimuths is solved to obtain layer boundary support set and layer number;

[0013] Layered anisotropy parameterization is established under the layer number and layer boundary constraints, and a joint cost function containing observation residual, anisotropy prior and sensitivity identifiable penalty is constructed, Gauss-Newton and multi-start are used for solving, and reversible jump MCMC is combined to obtain the posterior of parameters and layer number and output layered parameters and uncertainty.

[0014] In a preferred embodiment, the data acquisition and preprocessing comprises: synchronously acquiring the three-direction acceleration time histories at a sampling rate fs, using a window function to shape the original signal, recording the impact force spectrum according to the trigger, and performing detrending, bandpassing and resampling after analog-to-digital conversion to obtain unit impulse responses , and cross-calibrating and time-aligning the three-direction channels.

[0015] In a preferred embodiment, the incident deconvolution and coupling correction comprises: calculating response spectrum ) and deconvolving according to , constructing a multi-channel cross-power spectrum matrix , the sensor-soil contact transfer function H(ω) is estimated by minimum phase reconstruction and blind deconvolution, based on which the unit impulse response is corrected, and the corrected .

[0016] In a preferred embodiment, the k-ω imaging and directional decomposition comprises: performing two-dimensional fast Fourier transform on the unit impulse response to obtain , and azimuthal angle set is discretized to construct an azimuthal dispersion surface, based on ridge tracking and phase continuity, R / L modes are separated, and is zero-padded and de-aliased, while imposing a band mask and leakage suppression.

[0017] In a preferred embodiment, the generation of observation weights w comprises: calculating a signal-to-noise ratio from the time-domain signal, and defining a resolvable from the variance of closed three-phase sums, and normalizing to obtain, and mirror-continuing in the k-domain and setting a zero-padding multiple.

[0018] In a preferred embodiment, the horizon candidate and support set selection comprises: determining a candidate interface set Zcand from the modal bending and crossing of the azimuthal dispersion surface in a stable frequency band, and calculating a Fréchet derivative of the Thomson-Haskell transfer matrix to form a sensitivity kernel , based on which a depth grid is discretized and an extreme depth set is determined, and Δz and Δω sampling intervals are set, Δz and Δω representing a discrete step in the depth direction and a sampling interval in the angular frequency direction, respectively.

[0019] In a preferred embodiment, the support set solving comprises: setting a shared support for the shear modulus jump in R / L modes and all azimuths, constructing an objective function with linearized dispersion mapping and observation residuals as terms, combining L2,1 group sparsity and depth total variation regularization, and obtaining the layer boundary support set and the number of layers and outputting the depth positions.

[0020] In a preferred embodiment, the layered anisotropy joint inversion comprises: parameterizing each layer and constructing a joint cost function containing observation residuals, anisotropy priors, and an identifiability penalty derived from the sensitivity kernel under the constraints of the number of layers and the layer boundary support set, and solving by using automatic differentiation to generate Jacobians and weight matrices.

[0021] ​In a preferred embodiment, wherein the solving process comprises: employing Gauss-Newton iteration and multi-start global-local hybrid strategy, setting modal energy cutoff, line search or damping factor and trust region radius, updating the Jacobian and weight matrix according to free surface, interlayer continuity and bottom impedance boundary, limiting maximum step number and minimum step length, and terminating according to preset convergence criterion.

[0022] The intelligent test system for foundation pit solidified soil mechanical parameters based on impact response is used for realizing the method, and comprises:

[0023] The impact excitation device is configured to apply impact along the azimuth angle set Θ in the test point domain Ωx.

[0024] The multi-channel vibration pickup array is configured to acquire three-direction acceleration time histories and output corresponding electrical signals.

[0025] The data acquisition and transmission module is configured to amplify, filter and analog-to-digital convert the electrical signals and transmit them.

[0026] The data processing and analysis module is configured to: acquire unit impulse response and R / L dispersion curves; generate k-ω spectrum and construct azimuth dispersion surface and observation weight; determine layer boundary support set and layer number according to candidate interface and group sparse optimization; solve joint cost function under layer anisotropy parameterization and perform reversible jump MCMC to obtain layered parameters and uncertainty.

[0027] The intelligent test system for foundation pit solidified soil mechanical parameters based on impact response has the following technical effects and advantages:

[0028] The present application reduces the interference of excitation and sensing contact link through incident deconvolution and coupling correction, improves the reliability of dispersion extraction and modal separation by using k-ω azimuth dispersion surface and "signal-to-noise ratio - modal separability - phase closure" joint weight; the number of layers and layer boundary data are determined and controlled by using persistent homology to select candidate layer positions and combining L2,1 group sparse and depth TV layer selection shared across R / L and multiple azimuths; in the joint inversion of layer anisotropy, the sensitivity kernel dominant distinguishability penalty and multi-start Gauss-Newton are introduced to obtain parameter and layer number posteriori by RJ-MCMC, forming stable parameter estimation and uncertainty quantification; the "layering / direction" confusion is decoupled by azimuth consistency and principal axis cooperation, and the segmented consistent 2.5D profile is obtained by cross-point domain spatial regularization; PINNs physical residual back-feeding is used to absorb slight forward bias, and SNR, cond(J), layer number posteriori entropy and other indicators are used as convergence and quality control basis, and the results are output accordingly, realizing reliable inversion and traceable quality control of layered parameters of foundation pit solidified soil. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1The structural block diagram of the intelligent test system for the mechanical parameters of the foundation pit solidified soil based on impact response is provided in the present application.

[0030] Figure 2 The simple flow chart of the intelligent test method for the mechanical parameters of the foundation pit solidified soil based on impact response is provided in the present application.

[0031] Figure 3 The schematic diagram of k-omega spectral imaging, azimuthal dispersion surface and weight generation is provided in the present application.

[0032] Figure 4 The horizon inversion result graph is provided in the present application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0034] Please refer to Figure 1 The present application provides an intelligent test system for the mechanical parameters of the foundation pit solidified soil based on impact response, which comprises an impact excitation device, a multi-channel vibration pickup array, a data acquisition and transmission module, and a data processing and analysis module.

[0035] The impact excitation device is used to apply impact excitation to the foundation pit solidified soil body to generate elastic waves. The impact excitation device can be a built-in force hammer or other controllable impact source. The multi-channel vibration pickup array is used to receive the vibration signals generated in response to the soil body and convert them into electrical signals. The vibration pickup array is usually composed of several three-axis acceleration sensors, so as to receive the three-axis acceleration time histories in the test point domain Ωx (several adjacent test points) along several azimuth angles The measured three-axis acceleration time histories The data acquisition and transmission module is used to amplify, filter, and analog-digital convert the electrical signals collected by the vibration pickup array, and transmit the digital signals to the data processing and analysis module.

[0036] The data processing and analysis module is the core of the application, and its function is to process and invert the collected data in depth to obtain the mechanical parameters of the solidified soil of the foundation pit. The data processing and analysis module can be composed of one or more processors, memories and programs running on the processors, and specifically includes the following functional units: incident deconvolution and coupling correction unit, k-ω wave number-frequency imaging and direction decomposition unit, horizon candidate and support set selection unit, layered anisotropy joint inversion unit, Bayesian layered modeling unit with variable number of layers, PINNs physical constraint network unit, azimuth consistency and principal axis solving unit, cross-point domain continuity and spatial regularization unit, and convergence criterion, quality metric and engineering output unit.

[0037] Please refer to Figure 2 The application also provides an intelligent test method for mechanical parameters of solidified soil of a foundation pit based on impact response, which specifically includes the following steps:

[0038] S1: data acquisition and preprocessing:

[0039] In the test point domain Ωx (several adjacent test points), along several azimuth angles , an impact is applied by an impact excitation device, and three-direction acceleration time histories are measured by using a multi-channel seismic array. After deconvolution processing, unit impulse responses and their time-frequency representations are obtained. On this basis, the dispersion curves dominated by Rayleigh waves (R) and Love waves (L) are further extracted.

[0040] S2: incident deconvolution and coupling correction:

[0041] The impact force spectrum is obtained by using a built-in force hammer or a reference sensor. By dividing the measured response spectrum by the impact force spectrum , the unit impulse response spectrum is obtained. In order to eliminate the disturbance of the sensor-soil contact transfer function, the application corrects it by using the minimum phase reconstruction of the multi-channel mutual power spectrum matrix and the blind deconvolution technology. This step ensures the accuracy of subsequent data processing and avoids errors caused by sensor coupling effects.

[0042] S3: k-ω wave number-frequency imaging and direction decomposition:

[0043] The unit impulse response obtained in step S2 is subjected to 2DFFT (two-dimensional fast Fourier transform) to obtain the wave number-frequency spectrum . By ridge line tracking and phase continuity analysis, the dispersion curves of Rayleigh waves and Love waves are accurately extracted from the spectrum and To comprehensively capture the anisotropic characteristics of soil, this invention employs the azimuth ring method (along several discrete azimuth angles θ) to construct azimuth dispersion surfaces. ,like Figure 3 As shown. This step maximizes the consistency of phase closure error, modal separability (R / L non-contamination), and azimuth smoothness in the observation domain, and generates observation uncertainty weights based on the azimuth dispersion surface. This weight is not a subjective weighting, but is generated by the following formula:

[0044] ;

[0045] in, Signal-to-noise ratio (SNR) represents the ratio of signal strength to noise level. To achieve modal separability, the degree of separation between Rayleigh and Love waves in the frequency domain is quantified; Phase closure error The reciprocal of the variance, The results, obtained through analysis of the multi-channel cross-power spectrum matrix, reflect the self-consistency of the observation data. , , Z is the normalization coefficient and Z is the normalization factor. This weight directly reflects the confidence level of the observed data, providing reliable input for subsequent inversion.

[0046] S4: Selection of candidate stratigraphic levels and support sets:

[0047] This step aims to determine the number of soil layers and layer boundaries using a data-driven approach. First, based on the stable occurrence frequency bands of higher-order mode intersections / bending points on the multi-directional dispersion surface, combined with sensitivity kernels... The extreme depth is used to obtain the candidate interface set Zcand. Sensitivity kernel The sensitivity of wave velocity at a specific frequency to soil parameters at different depths can be calculated using the Fréchet derivative of the Thomson-Haskell transfer matrix or the Born approximation, followed by numerical integration and discretization. Secondly, to filter out false layers triggered by random noise, this invention employs the Persistent Homology method to evaluate the topological robustness of candidate interfaces. Persistent Homology can identify topological features at different scales in the data, thereby distinguishing between true layer interfaces and noise.

[0048] After determining the candidate interface set, Zcand is discretized into depth grid points, and then... This represents a shear modulus jump. This invention obtains the sparse jump support set Z* by solving a set of sparse-fusion problems, thereby determining the final layer interface. The optimization problem can be expressed as:

[0049] ;

[0050] where F is the linearized dispersion map, and G is the depth group shared across azimuths / modes. The group sparse variable The Rayleigh / Love modes are shared across all azimuths, i.e. group g is defined as the shared variable at the same depth grid point across R / L modes and all azimuth angles. denotes the L2,1 norm, i.e. first take the L2 norm of the elements within a group, and then take the L1 norm of all groups, which enforces the shear modulus jumps to be sparse in the depth direction, i.e. only a few depth positions have jumps, thus determining the layer interfaces. This step takes the “whether a certain layer exists” as a shared decision variable across the observation dimension, avoiding subjective weighting, and achieving the sparsity of the structure domain.

[0051] S5: Joint inversion of layered anisotropy:

[0052] After determining the number of layers , this step performs parameter inversion. Let the parameters of each layer j be , where is the thickness, is the density, is the isotropic base velocity, is the anisotropy amplitude, is the principal axis direction, is the dissipation. The present application constructs a joint cost function J(Θ) for optimization solution:

[0053] ;

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] where is the theoretical dispersion calculated based on the Thomson-Haskell transfer matrix extended to weakly anisotropic layered media. The observation residual term contains the observation uncertainty weight generated by step S3.

[0059] Forward model boundary conditions: Free surface boundary condition (top), interlayer continuity condition (displacement and stress continuity), and half-space / impedance boundary condition (bottom) are considered when calculating theoretical dispersion. The numbering of Rayleigh / Love multimodal and the truncation criterion are based on the energy contribution, and high-order modes with energy contribution less than 1% are usually ignored.

[0060] The second term is an anisotropy prior term to limit unreasonable extreme anisotropy, for example, to limit and penalize large deviation from the known construction direction (such as the construction reinforcement direction / surrounding purlin direction), ensuring mechanical reasonableness.

[0061] The third term is a sensitivity-guided identifiability penalty term where is the depth sensitivity kernel. This term is used to suppress the free drift of low-identifiability parameters and avoid the ill-conditioned nature of inversion, i.e., when the parameters are less sensitive to the observed data, automatically "reduce the degrees of freedom".

[0062] The fourth term is an interface sharpness / overfitting suppression term where represents the shear modulus of the jth layer, and this term promotes layer interface sharpness and suppresses excessive fluctuations of parameters in the depth direction.

[0063] The fifth term is a dissipation slow-varying prior term to promote smooth changes of the dissipation parameter in the depth direction.

[0064] This joint cost function is solved by the Gauss-Newton algorithm combined with automatic differentiation, and a multi-start global-local hybrid optimization (such as temperature annealing) is used to avoid falling into local minimum. This step couples the physical equation, observation uncertainty, sensitivity identifiability, and structural sparsity in the same optimization problem, and introduces physical constraints and statistical terms in the same optimization problem.

[0065] S6: Bayesian hierarchical modeling with variable number of layers:

[0066] Around the optimal solution obtained in step S5, a reversible jump Markov chain Monte Carlo (RJ-MCMC) algorithm is performed. RJ-MCMC allows model jumps of plus / minus interfaces to the sparse jump support set Z* so as to explore the possibility of different numbers of layers N in the model space. The optimization initial value of the model jump is updated reversely from the priori by the results of the layer candidate and support set selection in step S4, and the layer finding result and the posteriori estimation are updated mutually. The jump operators of RJ-MCMC include: birth (adding a layer), death (deleting a layer), split (splitting a layer into two layers), merge (merging two layers into one layer) and drift (fine-tuning parameters). Each jump operator defines a corresponding parameterization, prior distribution and acceptance rate (calculated by Hastings ratio). In order to ensure physical rationality, a minimum layer thickness (for example, 0.1 meters) and a merging threshold are set. The likelihood function uses the residual term in step S5, and the priori contains , Ψ, TV, Q, etc. The output of this step is the posteriori distribution p(N|data) of the number of layers N and the posteriori distribution p( |data, N) of each layer parameter . This makes the number of layers no longer externally artificially set, but jointly determined with parameter estimation and physical consistency through the same posteriori, avoiding misjudgment of a multi-layer structure as overfitting, and simultaneously statistically evaluating the number of layers and parameters, giving a joint estimation of structure, parameters and uncertainty.

[0067] S7: PINNs physical constraint network residual convergence and direction consistency correction:

[0068] The present application introduces a physical information neural network (PINNs) for constraining physical residuals. A network is constructed to approximate the field and with depth z and azimuth angle θ as inputs. The anisotropic elastic wave equation and boundary conditions (such as free surface / half space) are explicitly embedded in the loss function of PINNs, and the parameterized solution obtained in steps S5 / S6 is used as a soft constraint initialization. The PINNs physical constraint network only alternately couples PDE residuals and analytical forward, and updates the parameters Θ by feeding back the PDE residuals RPDE. The loss function L of PINNs can be represented as:

[0069] ;

[0070] wherein the first term is a PDE residual term, representing the residual after substituting the network output into the anisotropic elastic wave equation; the second term is a boundary condition residual term; and the third term is the data fitting term. The PDE form is elastic wave equation, and the boundary term B includes free surface stress zero, interlayer displacement and stress continuity, etc. The number of collocation points (depth / azimuth) is set according to the calculation resources and accuracy requirements. , , is the weight coefficient, and the default range is [0.1,]. The parameter adjustment order is usually to adjust the weight of the data fitting term first, and then adjust the weight of the physical constraint term. This alternating minimization strategy avoids attributing the system error to the material parameters, and improves the robustness of the inversion result.

[0071] S8: Azimuth consistency and principal axis solving:

[0072] In order to improve the robustness and interpretability of the direction parameters, the present application uses the establishes a shared principal axis prior, for example, the construction of the direction of the reinforcing bar / enclosure purlin can be referred to , and the is punished for too large deviation, while allowing local drift.

[0073] This makes the principal axis direction shared across layers and allows local fine-tuning; non-layer-by-layer independent fitting and then averaging. In addition, the is fitted with Von Mises distribution and Watson U2 circular consistency test. If the azimuth consistency test is insufficient (for example, the U2 statistic does not pass the significance test), the model jump of step S6 is triggered for layering or rotation, which is used to distinguish the jump of layer interface and the change of direction parameters. The direction parameters and the layering parameters are jointly constrained in the estimation.

[0074] S9: continuity and spatial regularization across point domains:

[0075] If there are multiple adjacent test points in the same work area Ωx, in order to improve the point measurement result to a continuous profile, the present application adds a spatial coupling term to the joint cost function J(Θ) in step S5: This term promotes the smooth change of parameters between adjacent test points, forming a segmented consistent 2.5D profile. At the same time, the is subjected to field smoothing and geological member constraint, so that the principal axis changes slowly near the member direction. This makes the parameters, layer interfaces, and principal axes simultaneously spatially regularized, avoiding the problem of "point-by-point" and consistently constraining the results of adjacent test points in space.

[0076] S10: convergence criterion, quality metric and engineering output:

[0077] The present application judges the convergence of the algorithm by the following criteria: the rate of dispersion residual error reduction tends to be stable, the PDE residual RPDE is stable, the number of layers posterior entropy H(N|data) converges, and the azimuth consistency statistic exceeds the threshold.

[0078] At the output of the results, the present application provides quality labels, including signal-to-noise ratio (SNR), Jacobian condition (cond(J)), layer boundary persistence index, principal axis consistency score, and local anisotropy confidence interval. These quality labels provide users with a quantitative assessment of the reliability of the inversion results.

[0079] The final engineering output includes: stratified parameter table and its uncertainty interval. Based on these parameters, the stiffness modulus and the directional equivalent stiffness modulus can be further derived for foundation pit support / backfill design. The output parameters and their uncertainties are provided for subsequent design.

[0080] As shown in Figure 4 , the present application forms a complete intelligent testing system and method through the close cooperation of the above steps. Among them, the pretreatment and directional dispersion (S2, S3) provide high-quality input for subsequent inversion; the topological persistence identifies the candidate layer, and then the group sparse constraint determines the number of layers and layer boundaries; the joint inversion (S5) optimizes the parameters in R / L wave, frequency, and direction; RJ-MCMC (S6) realizes the Bayesian uncertainty quantification of the number of layers and parameters; PINNs residual backfill (S7) handles the slight deviation between the model and the actual situation; the principal axis layer consistency (S8) ensures the robustness of the directional parameters; the spatial coupling (S9) improves the point measurement results to continuous profiles; and the final quality control convergence and output (S10) provides reliable basis for engineering decision-making. The absence of any step will affect the consistency of stratification, direction, and dissipation parameters.

[0081] The present application accurately solves the ill-conditioned problem of the prior art:

[0082] Ill-conditioned source 1: unknown number of layers: through the topological persistence, group sparsity, and cross-model Bayesian linkage (S4, S6), the data-driven determination of the number of layers is realized.

[0083] Ill-conditioned source 2: azimuth anisotropy and layering confusion: through the construction and sharing of principal axis prior of multi-azimuth dispersion surface (S3, S8), the change of azimuth angle and the change of layer boundary are separated and identified.

[0084] Ill-conditioned source 3: forward simplification and field deviation: the physical residual term in PINNs is used to handle the slight model and field differences.

[0085] Ill-conditioned source 4: uneven distinguishability: a sensitivity-guided distinguishability penalty term is introduced in the joint cost function (S5) to reduce the parameter degrees of freedom in the frequency band or depth with lower recognition degree, thereby reducing the drift.

[0086] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product.

[0087] Those skilled in the art can realize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and the constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0088] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0089] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0090] Finally, the above is merely preferred embodiments of the present application, and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for intelligent testing of soil mechanics parameters of a foundation pit based on impact response, characterized in that, Comprising the steps of: Along the azimuthal angle set Triggered impulse excitations and three-component acceleration time histories were acquired by a multi-channel seismic array. Deconvolution and digital processing were performed to obtain the unit impulse response and its time-frequency representation and R / L dispersion curves. The trigger force spectrum was also recorded simultaneously. A two-dimensional fast Fourier transform is performed on the unit impulse response to obtain a k-ω spectrum , combined with the ridge tracking phase continuity decomposition mode, according to the azimuth to construct the dispersion surface and based on the signal-to-noise ratio, modal degree of separation and phase closure statistics to generate observation weights w; Determine the candidate interface set Zcand according to the modal crossing of the azimuthal dispersion surface in the stable frequency band and the extremum of the sensitivity kernel, and evaluate the topological robustness with persistent homology, discretize the depth and set the shear modulus jump, solve the group sparse optimization shared by R / L and all azimuths to obtain the layer boundary support set and the number of layers; Under the constraints of the number of layers and the layer boundary, establish the layered anisotropy parameterization and construct a joint cost function containing observation residuals, anisotropy priors and sensitivity identifiable penalties, solve with Gauss-Newton and multi-start, and obtain the posterior of parameters and the number of layers with reversible jump MCMC to output the layered parameters and uncertainties.

2. The intelligent test method for the mechanical parameters of the solidified soil of a foundation pit based on impact response according to claim 1, characterized in that: Data acquisition and pre-processing includes: synchronous acquisition of three-direction acceleration time history at sampling rate fs, shaping of original signal with window function and pre-emphasis, recording of impact force spectrum according to trigger , and performing detrending, band-passing and resampling to obtain unit impulse response after analog-digital conversion , cross-calibration and time alignment of three-direction channels, azimuth angle.

3. The method for intelligent testing of soil mechanics parameters of foundation pit consolidation based on impact response according to claim 1, characterized in that: wherein the incident deconvolution and coupling correction comprises calculating a response spectrum and deconvolution, constructing a multi-channel cross-power spectrum matrix using minimum phase reconstruction and blind deconvolution to estimate the sensor-soil contact transfer function H(ω), based on which the unit impulse response is corrected, and the corrected is output.

4. The method for intelligent testing of soil mechanics parameters of a foundation pit based on impact response according to claim 1, characterized in that ； where k-ω imaging and directional decomposition includes: performing a two-dimensional fast Fourier transform on the unit impulse response to obtain , discretizing the azimuthal angle set to construct an azimuthal dispersion surface, separating R / L modes based on ridge tracking and phase continuity, and performing zero padding and de-aliasing while applying a band mask and leakage suppression. ​ 5. The intelligent test method for the mechanical parameters of the solidified soil of a foundation pit based on impact response according to claim 1, characterized in that: The generation of the observation weights w comprises: Computing the signal-to-noise ratio with the time-domain signal Defining the separable degree of the bandwidth of the modal cross Defining the phase closure statistics with the variance of the closed ternary phase sum , and the mirror extension is made in the k domain and the zero padding multiple is set. Normalization is obtained, and the mirror extension is made in the k domain and the zero padding multiple is set.

6. The method for intelligent testing of soil mechanics parameters of a foundation pit consolidation according to the impact response of claim 4, characterized in that, The horizon candidates and support set selection includes: determining a candidate interface set Zcand according to mode bending and intersection of the azimuth dispersion surface in a stable frequency band, calculating a Fréchet derivative of a Thomson-Haskell transfer matrix to form a sensitivity kernel , and setting a Δz and a Δω sampling interval, the Δz and the Δω respectively representing a discrete step length in a depth direction and a sampling interval in an angular frequency direction.

7. The method for intelligent testing of soil mechanics parameters of foundation pit consolidation based on impact response according to claim 6, characterized in that, The support set solving includes: setting the shear modulus jump variable to share the support on the R / L mode and all azimuths, constructing an objective function with linearized dispersion mapping and observation residuals as terms, combining L2,1 group sparsity and total variation regularization to obtain the layer boundary support set and the number of layers and output the depth position.

8. The method for intelligent testing of soil mechanics parameters of a foundation pit based on impact response according to claim 1, characterized in that, Wherein the layered anisotropy joint inversion includes: under the constraints of the number of layers and the layer boundary support set, parameterizing each layer and constructing a joint cost function containing observation residuals, anisotropy priors and identifiable penalties derived from sensitivity kernels, solving with automatic differentiation to generate Jacobian and weight matrix.

9. The method for intelligent testing of soil mechanics parameters of a foundation pit based on impact response according to claim 8, characterized in that, Wherein the solving process includes: using Gauss-Newton iteration and multi-start global-local hybrid strategy, setting mode energy cutoff, line search or damping factor and trust region radius, updating Jacobian and weight matrix according to free surface, interlayer continuity and bottom impedance boundary, limiting the maximum number of steps and the minimum step length, and terminating according to the preset convergence criterion.

10. An intelligent test system for soil mechanical parameters of foundation pit consolidation based on impact response, for implementing the method of any one of claims 1-9, characterized in that, Comprising: An impact excitation device configured to apply an impact along a set of azimuthal angles Θ within a test point domain Ωx; A multi-channel seismic array configured to obtain three-dimensional acceleration time histories and output corresponding electrical signals; A data acquisition and transmission module configured to amplify, filter and analog-to-digital convert the electrical signals and transmit them; A data processing and analysis module configured to: obtain unit impulse responses and R / L dispersion curves; Generate k-ω spectrum and construct azimuthal dispersion surface and observation weight; Determine the layer boundary support set and the number of layers according to the candidate interface and group sparse optimization; solve the joint cost function under the layered anisotropy parameterization and perform reversible jump MCMC to obtain the layered parameters and uncertainties.

Images