An expressway deep foundation pit excavation intelligent monitoring system and method

Abstract

This invention belongs to the field of digital twin technology, specifically relating to an intelligent monitoring system and method for deep foundation pit excavation on highways. The system constructs a digital twin model of the deep foundation pit based on the pit's design outline, soil parameters, excavation stages, and support boundary conditions. It maps on-site monitoring data into a foundation pit deformation state vector according to the excavation stage and spatial location, and performs residual calculations with the simulation state vector output by the digital twin model. Within a preset physical constraint range, it performs parameter inversion on the soil's equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient. Based on the inversion results, it updates the calculation parameters of the digital twin model and outputs the corrected foundation pit deformation state distribution. This invention can reduce the cumulative deviation between design parameters and actual excavation states, enabling monitoring data to participate in the model correction process, and improving the continuity and consistency of deep foundation pit deformation state calculations.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology, specifically relating to an intelligent monitoring system and method for deep foundation pit excavation on highways. Background Technology

[0002] Deep foundation pit excavation for highways is typically characterized by long route distances, multiple excavation stages, significant variations in geological conditions, and substantial influence of construction sequence on support boundaries. Existing monitoring schemes generally establish a foundation pit calculation model based on design documents and collect monitoring data such as horizontal displacement, vertical settlement, and support deformation at the construction site, then compare the monitored values ​​with preset alarm values. When the monitored values ​​exceed the limits, the system outputs an alarm result; when the monitored values ​​do not exceed the limits, the system generally considers the foundation pit to be in an acceptable state. Some schemes establish a foundation pit visualization model in the monitoring platform to display the foundation pit outline, monitoring point locations, monitoring curves, and construction stages. However, such models primarily serve a data display function, and the model parameters often use geological parameters, support parameters, and excavation condition parameters from the survey and design stage. Soil disturbance and boundary changes during construction are difficult to reflect synchronously in the model calculation process.

[0003] In relatively similar existing technologies, digital twin models are used for monitoring deep foundation pit construction. These models typically correlate foundation pit geometry, geological stratification, support structure information, and monitoring point data to create a visualized or semi-simulated monitoring environment. Such solutions can update the model appearance according to the excavation stage and load monitoring data at corresponding locations. They can also extrapolate trends based on historical curves. However, the calculation parameters of digital twin models are mostly determined once at the initial modeling stage, and subsequent adjustments rely mainly on manual adjustments or fixed rule corrections. When discrepancies exist between on-site monitoring data and simulation calculation results, existing solutions usually only determine whether the difference exceeds an empirical threshold or use methods such as data smoothing and mean compensation to handle anomalies. It is difficult to transform these discrepancies into online correction bases for calculation parameters such as soil equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient.

[0004] Therefore, the core technical problem in existing intelligent monitoring of deep foundation pit excavation for highways lies in the continuous deviation between the design parameters of the digital twin model and the actual excavation state. Although on-site monitoring data can reflect local deformation results, it is difficult to directly correct the internal calculation parameters of the model, resulting in cumulative residuals between the simulated state output by the digital twin model and the actual deformation state of the foundation pit. When these residuals are superimposed across different excavation stages, spatial locations, and monitoring types, existing technologies struggle to distinguish between model parameter deviations, construction disturbance deviations, and monitoring anomaly deviations. This lack of reliable parameter basis for subsequent deformation state distribution calculations affects the continuity and consistency of foundation pit deformation evolution judgment. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent monitoring system and method for deep foundation pit excavation on highways, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A smart monitoring system for deep foundation pit excavation on highways includes a digital twin modeling component, a monitoring state construction component, a simulation residual calculation component, a parameter inversion component, and a model correction component. The digital twin modeling component constructs a digital twin model of the deep foundation pit based on the pit's design outline, geological parameters, excavation stage, and support boundary conditions. The monitoring state construction component converts on-site monitoring data into a foundation pit deformation state vector. The simulation residual calculation component calculates the residual between the foundation pit deformation state vector and the simulation state vector output by the digital twin model. The parameter inversion component inverts the model's parameters to be corrected within a preset physical constraint range. The model correction component updates the calculation parameters of the digital twin model based on the inversion results and outputs the corrected foundation pit deformation state distribution, thus forming a smart monitoring system for deep foundation pit excavation on highways.

[0007] Preferably, the monitoring status construction component includes an excavation stage index subcomponent, a spatial mapping subcomponent, and a status vector generation subcomponent; the excavation stage index subcomponent is used to generate stage identifiers according to the excavation depth, excavation zones, and construction sequence of the foundation pit; the spatial mapping subcomponent is used to map the field monitoring data to the corresponding calculation positions in the deep foundation pit digital twin model; the status vector generation subcomponent is used to generate a foundation pit deformation status vector containing horizontal displacement, vertical settlement, support deformation increment, deformation rate, and stage code based on the stage identifiers and the corresponding calculation positions.

[0008] Preferably, the parameter inversion component includes a parameter candidate space construction sub-component, a constraint optimization sub-component, and a parameter combination output sub-component; the parameter candidate space construction sub-component is used to set the soil equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient as parameters to be corrected, and to generate corresponding physical constraint intervals based on the stratum type, excavation stage, and support boundary conditions; the constraint optimization sub-component is used to search for combinations of parameters to be corrected within the physical constraint intervals with the normalized value of the residuals as the optimization objective; the parameter combination output sub-component is used to output parameter inversion results that satisfy the constraint conditions.

[0009] Preferably, the simulation residual calculation component includes a residual matrix generation subcomponent, a residual decomposition subcomponent, and a residual type marking subcomponent; the residual matrix generation subcomponent is used to generate a multidimensional residual matrix according to the monitoring type, spatial location, and excavation stage; the residual decomposition subcomponent is used to decompose the residuals into spatially continuous residuals, stage-accumulated residuals, and local abrupt residuals based on the multidimensional residual matrix; the residual type marking subcomponent is used to mark the spatially continuous residuals, the stage-accumulated residuals, and the local abrupt residuals as candidate types of model parameter deviations, construction disturbance deviations, or monitoring anomaly deviations, respectively.

[0010] Preferably, the state vector generation subcomponent includes a time-series window configuration subcomponent and a multi-scale feature splicing subcomponent; the time-series window configuration subcomponent is used to configure short-term windows and cross-stage windows according to the continuous monitoring period before the current excavation stage; the multi-scale feature splicing subcomponent is used to extract deformation increment and deformation rate from the short-term window, extract deformation accumulation and stage change trend from the cross-stage window, and splice the deformation increment, deformation rate, deformation accumulation, stage change trend and stage code into a multi-scale foundation pit deformation state vector.

[0011] Preferably, the spatial mapping subcomponent includes a coordinate normalization subcomponent, a computational grid binding subcomponent, and a monitoring gap completion subcomponent; the coordinate normalization subcomponent is used to transform the coordinates of the field monitoring points to the unified coordinate system of the deep foundation pit digital twin model; the computational grid binding subcomponent is used to bind the transformed field monitoring point coordinates to adjacent twin computational grids; the monitoring gap completion subcomponent is used to generate complete monitoring values ​​and corresponding confidence levels for state vector construction based on spatial adjacency relationships and deformation trends at the same stage when there are monitoring gaps between adjacent twin computational grids.

[0012] Preferably, the constrained optimization subcomponent includes a hierarchical inversion subcomponent and a coupled constraint subcomponent; the hierarchical inversion subcomponent is used to first perform grouped inversion on the equivalent stiffness of soil corresponding to different strata, then perform staged inversion on the excavation unloading coefficients corresponding to different excavation stages, and perform local inversion on the support constraint reduction coefficients corresponding to the support boundary; the coupled constraint subcomponent is used to apply parameter continuity constraints, excavation stage progression constraints, and support boundary consistency constraints between the grouped inversion, the staged inversion, and the local inversion, to generate hierarchically coupled parameter inversion results.

[0013] Preferably, the parameter combination output subcomponent includes a parameter stability evaluation subcomponent and a boundary proximity recording subcomponent; the parameter stability evaluation subcomponent is used to evaluate the sequence consistency of the parameter inversion results output from multiple consecutive excavation stages and generate parameter fluctuation characteristics; the boundary proximity recording subcomponent is used to record the degree of proximity of the parameter inversion results to the upper or lower limit of the physical constraint interval, and write the parameter fluctuation characteristics and the degree of proximity into the stage correction record of the deep foundation pit digital twin model.

[0014] Preferably, the residual type marking subcomponent includes a sensitivity perturbation subcomponent and an uncertain region calibration subcomponent; the sensitivity perturbation subcomponent is used to apply single-parameter perturbations to the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient, respectively, to generate the residual sensitivity of each parameter to be corrected to the multidimensional residual matrix; the uncertain region calibration subcomponent is used to calibrate the calculation location where the residual sensitivity is lower than the preset sensitivity condition and the residual continuously increases as a high uncertainty evolution region, and write the high uncertainty evolution region into the digital twin calculation boundary of the subsequent excavation stage.

[0015] A method for intelligent monitoring of deep foundation pit excavation on highways includes: constructing a digital twin model of the deep foundation pit based on the pit design outline, soil parameters, excavation stage, and support boundary conditions; mapping field monitoring data to the digital twin model of the deep foundation pit according to the excavation stage and spatial location to generate a deformation state vector of the foundation pit; obtaining the simulation state vector output by the digital twin model of the deep foundation pit and calculating the residual between the deformation state vector of the foundation pit and the simulation state vector; within a preset physical constraint range, inverting the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient with the residual as the target; updating the calculation parameters of the digital twin model of the deep foundation pit based on the inversion results, and outputting the corrected distribution of the foundation pit deformation state, thus forming an intelligent monitoring method for deep foundation pit excavation on highways.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a deep foundation pit digital twin model using a digital twin modeling component, which integrates the foundation pit design outline, soil parameters, excavation stages, and support boundary conditions. A monitoring state construction component converts on-site monitoring data into a foundation pit deformation state vector, enabling the on-site monitoring results to correspond with the model simulation state within the same computational object. A simulation residual calculation component calculates the residual between the foundation pit deformation state vector and the simulation state vector. A parameter inversion component inverts the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient within a preset physical constraint range. The model calibration component updates the calculation parameters of the deep foundation pit digital twin model accordingly. Thus, monitoring data is no longer merely used as alarm comparison data but participates in the parameter calibration process of the digital twin model, allowing model parameters to be updated according to changes in the excavation stage and on-site deformation state. This reduces the cumulative deviation between design parameters and actual construction conditions, ensuring that the output foundation pit deformation state distribution has continuous monitoring and simulation constraints.

[0017] 2. The excavation stage indexing, spatial mapping, and state vector generation of this invention enable field monitoring data to be entered into the digital twin model according to excavation depth, excavation zone, construction sequence, and calculation location, reducing state expression deviations caused by mixing data from different stages. The multi-scale foundation pit deformation state vectors formed by short-time windows and cross-stage windows can retain the correspondence between deformation increments, deformation rates, cumulative deformation, and stage change trends. Coordinate normalization, computational grid binding, and monitoring gap completion enable monitoring point data to be mapped to the twin computational grid, and complete monitoring values ​​with confidence are generated at monitoring gap locations. Residual matrix generation, residual decomposition, and residual type labeling can distinguish between spatially continuous residuals, stage-accumulated residuals, and locally abrupt residuals according to monitoring type, spatial location, and excavation stage. Hierarchical inversion, coupling constraints, parameter stability evaluation, boundary proximity recording, sensitivity perturbation, and high-uncertainty evolution region calibration enable the model calibration process to retain stage records, parameter boundary information, and residual sensitivity information, providing data basis for the digital twin calculation boundaries of subsequent excavation stages. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the overall monitoring and digital twin correction process of the intelligent monitoring system for deep foundation pit excavation on highways according to the present invention. Figure 2 This is a flowchart illustrating the stage indexing, spatial mapping, and state vector construction of field monitoring data in this invention. Detailed Implementation

[0019] refer to Figure 1In one embodiment, the intelligent monitoring system for deep foundation pit excavation on highways is implemented using a computational program within an engineering monitoring platform. The system includes a digital twin modeling component, a monitoring state construction component, a simulation residual calculation component, a parameter inversion component, and a model calibration component. The digital twin modeling component constructs a digital twin model of the deep foundation pit based on the pit's design outline, geological parameters, excavation stage, and support boundary conditions. The monitoring state construction component converts on-site monitoring data into a foundation pit deformation state vector that can be used for model calculation. The simulation residual calculation component calculates the residual between the foundation pit deformation state vector and the simulation state vector output by the deep foundation pit digital twin model. The parameter inversion component inverts the model's parameters to be corrected within a preset physical constraint range. The model calibration component updates the computational parameters of the deep foundation pit digital twin model based on the inversion results and outputs the corrected foundation pit deformation state distribution. All of the above components can be implemented through computer program modules, data tables, model call interfaces, and computational task scheduling logic. The components communicate with each other through a unified data object, transmitting information such as the foundation pit location, excavation stage, monitoring type, simulation state, residual results, and parameter inversion results.

[0020] In this embodiment, the digital twin model of the deep foundation pit is not merely used to display the pit's outline or monitoring curves, but rather to transform the pit's design outline, soil parameters, excavation stages, and support boundary conditions into computational objects that can participate in residual calculation and parameter inversion. Specifically, the digital twin modeling component first reads the design outline of the pit along the highway alignment direction, dividing the pit's length, excavation depth, and lateral width directions into twin computational grids; then, it binds the soil parameters to the corresponding computational grids according to their spatial location; subsequently, it writes the excavation stages into the model's timeline, ensuring that each excavation stage has a corresponding excavation depth, excavation zone, and construction sequence identifier; and it writes the support boundary conditions into the model boundary in the form of constraint state fields. Thus, the digital twin model can establish a correspondence between spatial location, soil attributes, excavation stages, boundary constraints, monitoring status, and simulation status. Subsequent residual calculations no longer rely on manual comparison but are completed under the same computational coordinates and the same stage index.

[0021] In this embodiment, the data object mapping of the deep foundation pit digital twin model is shown in Table 1.

[0022] Table 1 Data Object Mapping Table for Digital Twin Model of Deep Foundation Pit

[0023] In one embodiment, the digital twin modeling component unifies the field coordinates and model coordinates to avoid mismatches in residual calculation objects caused by inconsistencies between the locations of field monitoring points and the twin computational grid. The field monitoring point coordinates are represented as field coordinate vectors, and the digital twin model uses a unified coordinate system. The coordinate transformation relationship is expressed by the following formula:

[0024] in, This represents the field coordinate vector of the nth field monitoring point. This represents the transformed digital twin coordinate vector. Represents the coordinate rotation and scaling unification matrix. This represents a coordinate translation vector, and the operator indicates multiplication of a matrix and a vector followed by vector addition. The physical meaning lies in transforming field coordinates from different sources to the unified computational space of the digital twin model. It should be noted that all example values ​​of monitoring data in this invention are actual measurements under the corresponding physical dimensions. The units of each physical quantity are specified as follows: horizontal displacement, vertical settlement, and support deformation are all in millimeters (mm); the unit of support deformation increment is also millimeters (mm); the unit of deformation rate is millimeters / day (mm / d); and the time stamp... The unit is day (d); the unit of soil equivalent stiffness is megapascal (MPa); the excavation unloading factor and the support constraint reduction factor are both dimensionless proportionality coefficients, ranging from 0 to 1. Normalized scale Having the same units as the corresponding monitoring type, it is used to transform residuals with different units to a unified scale for use in objective function calculation. All values ​​in the following examples are given in the above unit system. For example, when... Take the identity matrix, The value can be (10m, 5m, -3m). When the value is (2m, -1m, 0), the converted value is... The value (12m, 4m, -3m) indicates that the field monitoring point has been translated to its corresponding computational location in the twin model. This process ensures that the same monitoring point can be stably bound to the corresponding twin computational grid in subsequent stages.

[0025] In this embodiment, the monitoring status construction component converts on-site monitoring data into a foundation pit deformation status vector. The on-site monitoring data includes data types related to the safety of deep foundation pit excavation, such as horizontal displacement, vertical settlement, and support deformation. To avoid a single monitoring value failing to represent stage changes, the monitoring status construction component simultaneously writes the deformation amount, deformation increment, deformation rate, and stage code when generating the status vector. The foundation pit deformation status vector adopts the following form:

[0026] in, This represents the foundation pit deformation state vector at the nth calculation location during the hth excavation stage. Indicates horizontal displacement. Indicates vertical settlement. Indicates the amount of support deformation. Indicates the incremental support deformation between adjacent stages Indicates the deformation rate of the support. This indicates the code for the excavation stage. This represents the transpose of a vector. Physically, it compresses different monitoring quantities into a calculable state representation for the same stage and location. For example, if the horizontal displacement of the nth calculated location in the 3rd excavation stage is 12mm, the vertical settlement is 5mm, the support deformation is 8mm, the support deformation increment is 2mm, the support deformation rate is 1.5mm / d, and the stage code is 3, then the state vector is written as the transpose of [12,5,8,2,1.5,3].

[0027] For example, when the monitored state vector is the transpose of [12mm, 5mm, 8mm, 2mm, 1.5mm / d, 3], and the simulated state vector is the transpose of [10mm, 4mm, 7mm, 1mm, 1.2mm / d, 3], the residual vector is the transpose of [2mm, 1mm, 1mm, 1mm, 0.3mm / d, 0]. A stage-coded residual of 0 indicates that the two sets of data are in the same excavation stage, while the residuals of the other fields are used for parameter inversion and residual type determination.

[0028] For example, when It is 8mm. 5mm, When it is 2d, The calculated result is 1.5 mm / d.

[0029] For example, in a certain stage, only two monitoring types are considered: the residual of the first type of horizontal displacement is 2 mm, and the normalized scale is... 1mm, weight The value is 1, the second type (vertical settlement) residual is 3mm, and the normalized scale is 1.

[0030] 3mm, weight If the value is 2, then the target value is 1 × (2 / 1) of the first term. 2 =4 and the second term 2×(3 / 3) 2 =2, summed to obtain a result of 6 (dimensionless). This result can be used as a basis for comparing candidate parameter combinations.

[0031] For example, when h is 4, a is 1, and b is 3, It is 11mm. It is 9mm. It is 5mm.

[0032] It is 2d. When the value is 4, the multi-scale feature vector is the transpose of [2mm, 1mm / d, 6mm, 2mm / stage, 4]. This vector can be combined with the aforementioned foundation pit deformation state vector and input into the residual calculation process. For example, when For 2, 3. =1, It is 10mm. It is 11mm. When the value is 8mm, the completed monitoring value is 10.17mm. For example, when the Frobenius norm of the difference between the residual matrix before and after the disturbance is 3mm, When the absolute value of is 0.5 (dimensionless, representing the proportion of the perturbation applied to the normalized parameter), the residual sensitivity is 6 mm. This vector can be directly used for subsequent residual calculations.

[0033] The horizontal displacement variation across stages is used to place a single numerical result back into the operational context of the excavation stage, spatial grid, and monitoring type. The horizontal axis uses the stage numbers generated by the excavation stage index sub-component as the stage sequence, arranged sequentially according to the actual construction progress. For example, S1, S2, S3...Sn correspond to the first excavation stage, the second excavation stage, and so on up to the nth excavation stage, respectively. The vertical axis represents the average horizontal displacement within the monitoring area corresponding to each stage. Adjacent horizontal axes represent adjacent excavation stages, rather than continuous time scales, thus providing a clear visual representation of the trend of horizontal displacement variation as construction progresses.

[0034] Furthermore, the support deformation rate is not based on a fixed empirical value, but is calculated from the support deformation at the same calculation location in adjacent excavation stages and the time difference between the stages. The calculation formula is as follows:

[0035] in, This represents the support deformation rate at the nth calculation location during the hth excavation stage. This represents the support deformation during the h-th excavation stage. This indicates the amount of support deformation in the previous excavation stage. This represents the time marker corresponding to the h-th excavation stage. This represents the time marker corresponding to the previous excavation stage. The numerator represents the stage deformation increment, and the denominator represents the stage time interval. Its physical meaning lies in characterizing the intensity of change in support deformation as the excavation stage progresses. For example, when... It is 8. It is 5. When it is 2, The calculation result is 1.5. After this calculation result is entered into the state vector, the system can distinguish between a state with a large deformation but a slow change, and a state with a small deformation but a rapid change.

[0036] In one embodiment, the simulation residual calculation component reads the simulation state vector from the deep foundation pit digital twin model. The simulation state vector and the foundation pit deformation state vector use the same field order and the same stage index, allowing for bitwise subtraction. The residual calculation formula is:

[0037] in, This represents the residual vector at the nth calculation location during the hth excavation stage. This represents the deformation state vector of the foundation pit formed by on-site monitoring data. This represents the simulation state vector output by the digital twin model of the deep foundation pit. The minus sign indicates a subtraction of each item within the same field, location, and stage. Its physical meaning lies in describing the deviation between the on-site monitoring state and the simulation calculation state. For example, when the monitoring state vector is the transpose of [12,5,8,2,1.5,3] and the simulation state vector is the transpose of [10,4,7,1,1.2,3], the residual vector is the transpose of [2,1,1,1,0.3,0]. A stage-coded residual of 0 indicates that both sets of data are in the same excavation stage; the residuals of the other fields are used for parameter inversion and residual type determination.

[0038] In this embodiment, the parameter inversion component uses the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient as parameters to be corrected. These parameters are not arbitrarily adjusted, but rather searched within a preset physical constraint range, so that the simulation results of the deep foundation pit digital twin model gradually approach the on-site monitoring state. The objective function of the parameter inversion adopts the form of a normalized residual sum of squares:

[0039] in, This indicates that the h-th excavation stage is in the parameter vector The residual target value is below. Represents the vector of parameters to be corrected. This represents the set of locations involved in the calculation during the h-th excavation stage. This represents the set of monitoring types involved in the calculation. This represents the weight of the r-th monitoring type. In the parameter vector The residuals corresponding to the nth calculation location, the hth excavation stage, and the rth monitoring type are as follows: This represents the normalized scale for the r-th monitoring type. The summation operation in this formula represents the accumulation of all locations and monitoring types involved in the calculation, and the squaring operation is used to preserve the residual amplitude. The physical significance lies in avoiding inversion bias caused by different dimensions of different monitoring types. For example, if only two monitoring types are considered in a certain stage, with the first type residual being 2, the normalized scale being 1, and the weight being 1, and the second type residual being 3, the normalized scale being 3, and the weight being 2, then the target value is obtained by accumulating the first term 4 and the second term 2, resulting in a value of 6. This result can be used as a basis for comparing candidate parameter combinations.

[0040] objective function The descent curve illustrates how the normalized objective value changes during the parameter inversion process as the candidate parameter combination search progresses. The horizontal axis represents the candidate parameter combination index, numbered sequentially from 1, 2, 3...N according to the order in which the candidate parameter combinations are generated or searched; the vertical axis represents the normalized objective value calculated for the corresponding candidate parameter combination. The objective function descent curve reflects the gradual decrease in the objective value during parameter inversion, providing a visual basis for the selection of candidate parameter combinations and the parameter optimization process.

[0041] After obtaining the objective function, the parameter inversion component determines the parameter inversion results through constraint optimization. The generation of the candidate parameter set adopts a strategy combining Latin hypercube sampling and orthogonal design: First, the physical constraint intervals of the soil equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient are discretized into several equally spaced initial sampling layers in the three-dimensional parameter space. Then, a sample point is randomly selected from each sampling layer, ensuring that the sample points are evenly distributed throughout the parameter space, forming an initial candidate parameter set. Based on this, local sampling points are generated for each parameter to be corrected according to a preset disturbance ratio, using the inversion results from the previous stage as the center. Specifically, multiple candidate parameter values ​​are generated based on the parameter values ​​from the previous stage, with disturbance amplitudes of ±2%, ±5%, and ±10%, respectively. When a generated candidate parameter exceeds the physical constraint interval of the corresponding parameter in the current stage, it is truncated to the corresponding constraint boundary. Subsequently, a locally refined candidate parameter subset is formed according to the arrangement and combination of each parameter to be corrected, and this subset is merged with the initial candidate parameter set to form a complete candidate parameter set for subsequent constraint optimization calculations.

[0042] The optimization process is performed using the particle swarm optimization algorithm, which uses the normalized residual objective function. As a fitness function, it searches for the optimal combination of parameters within the constraints of the candidate parameter set. The position vector of each particle in the particle swarm corresponds to a set of parameters to be corrected. ,in Indicates the equivalent stiffness of the soil. Indicates the excavation unloading coefficient. This represents the support constraint reduction coefficient. The initial particle swarm size is 30 particles, the maximum number of iterations is set to 200, the inertia weight decreases linearly from 0.9 to 0.4 during iteration, and the acceleration factor... and All values ​​are set to 1.5. In each iteration, each particle updates its velocity and position based on its individual best historical position and the group's best historical position. If the updated position exceeds...

[0043] The range is then trimmed to the nearest constraint boundary to ensure that all candidate parameter combinations remain within the physical constraint space. When the fitness value change is less than a preset convergence threshold of 10 over 20 consecutive iterations... −4 When the algorithm terminates early, it outputs the optimal parameter inversion result that minimizes the objective function, calculated as follows:

[0044] in, This represents the optimal parameter inversion result for the h-th excavation stage. This represents the physical constraint space corresponding to the h-th excavation stage. In the parameter vector The residual objective value is given by arg min, which represents the value of the residual objective value under the given conditions. The parameter vector that minimizes the objective function is selected within the specified range. Physically, this ensures that the parameter inversion remains within a reasonable range defined by geological properties, excavation stage, and support boundary conditions. For example, candidate parameter combinations A, B, and C correspond to objective values ​​of 7, 4, and 6, respectively, and all three sets of parameters are within the specified range. When the time comes, parameter combination B is selected as If the target value of a candidate parameter combination is small but exceeds... If it does not pass the final output, it will not be included.

[0045] In this embodiment, after receiving the parameter inversion results, the model calibration component writes the equivalent soil stiffness, excavation unloading coefficient, and support constraint reduction coefficient back to the stage parameter table of the deep foundation pit digital twin model. To prevent historical stages from being overwritten by subsequent stages, the model calibration component binds and stores each calibration result with the excavation stage, calculation area, and residual target value, forming a stage calibration record. After the deep foundation pit digital twin model completes parameter updates in the current stage, it recalculates the deformation state distribution of the current stage and uses the calibrated model parameters as the initial parameters for the simulation calculation of the next excavation stage. Through this process, the design stage parameters, field monitoring status, simulation residuals, and inversion parameters are closed in the same data chain, reducing the problem of long-term disconnect between model parameters and actual excavation status. This implementation transforms the digital twin model from a static design model into a dynamic calculation model constrained by field monitoring through online inversion and stage calibration.

[0046] refer to Figure 2 In a preferred embodiment, the monitoring status construction component further includes an excavation stage index subcomponent, a spatial mapping subcomponent, and a state vector generation subcomponent. The excavation stage index subcomponent generates stage identifiers according to the excavation depth, excavation zones, and construction sequence, and writes these stage identifiers into each monitoring data entry. The spatial mapping subcomponent maps the field monitoring data to the corresponding computational location in the deep foundation pit digital twin model, prioritizing the nearest computational grid after coordinate normalization during mapping; when a monitoring point is located between multiple computational grids, the primary and secondary binding grids are determined based on spatial distance and grid adjacency. The state vector generation subcomponent generates a foundation pit deformation state vector containing horizontal displacement, vertical settlement, support deformation increment, deformation rate, and stage code based on the stage identifier and corresponding computational location. Through the combined constraints of the stage index and spatial mapping, field monitoring data will not be incorrectly written into adjacent excavation stages, nor will it be incorrectly bound to non-corresponding twin computational grids.

[0047] Furthermore, when generating stage identifiers, the excavation stage index subcomponent not only records the current excavation stage number but also the temporal relationship between the current stage and the previous and next planned stages. Stage codes can be represented by consecutive integers or by a combined code associated with the excavation partitions. For long linear excavation scenarios of deep foundation pits in highways, different partitions may be at different excavation depths at the same time. Therefore, the stage index subcomponent combines the excavation depth, excavation partition, and time sequence into a stage identifier, rather than using a single time point as the basis for stage determination. After this processing, two data points collected at the same monitoring time, if located in different excavation partitions, can still be mapped to different excavation stages; data collected at different times from the same excavation partition can also form a continuous state sequence according to the stage order. This implementation enables the state vector to express the actual characteristics of segmented excavation of deep foundation pits in highways.

[0048] In a preferred embodiment, the state vector generation subcomponent further includes a temporal window configuration subcomponent and a multi-scale feature splicing subcomponent. The temporal window configuration subcomponent configures short-time windows and cross-stage windows according to the continuous monitoring periods preceding the current excavation stage. Short-time windows are used to express the deformation increment and deformation rate within adjacent stages or adjacent time periods, while cross-stage windows are used to express the cumulative deformation and stage change trend within multiple excavation stages. The multi-scale feature splicing subcomponent splices the short-time window features, cross-stage window features, and stage codes into a multi-scale foundation pit deformation state vector, calculated as follows:

[0049] in, This represents the multi-scale feature vector of the nth calculated location at the hth excavation stage. This represents the monitoring value for the selected monitoring type at the h-th excavation stage. 'a' represents the stage interval corresponding to the short-time window, and 'b' represents the stage interval corresponding to the cross-stage window. This indicates the time interval corresponding to the short-time window. This indicates the excavation stage code. The first item in parentheses represents the short-term deformation increment, the second represents the short-term deformation rate, the third represents the cumulative deformation across stages, the fourth represents the average change trend across stages, and the fifth represents the stage code. For example, when h is 4, a is 1, and b is 3... It is 11. It is 9. It is 5. It is 2. When the value is 4, the multi-scale feature vector is the transpose of [2,1,6,2,4]. This vector can be combined with the aforementioned foundation pit deformation state vector and input into the residual calculation process.

[0050] Multi-scale feature trends are used to place single numerical results back into the operational context of the excavation stage, spatial grid, and monitoring type. The horizontal axis represents the line grid number, and the vertical axis represents the multi-scale feature value.

[0051] In this embodiment, the underlying logic of the multi-scale feature splicing sub-component lies in simultaneously retaining the adjacent changes of the current stage and the cumulative changes of multiple stages. When the short-term deformation rate increases while the cross-stage cumulative amount is still small, the system can use it as an early change feature to participate in the residual calculation; when the short-term deformation rate slows down while the cross-stage cumulative amount is large, the system can use it as a stage cumulative feature to participate in the parameter inversion. If only the current monitoring value is used, the two states may be confused; if only the cumulative value is used, the stage abrupt change may be smoothed. Multi-scale feature splicing enables the simulation residual calculation component to obtain a more complete state input, giving the subsequent parameter inversion a more stable data foundation for changes in excavation unloading and support constraints. This implementation combines short-term windows and cross-stage windows to adapt the state construction process to the non-uniform deformation process in the staged excavation of deep foundation pits.

[0052] In a preferred embodiment, the spatial mapping subcomponent includes a coordinate normalization subcomponent, a computational grid binding subcomponent, and a monitoring gap completion subcomponent. The coordinate normalization subcomponent transforms the coordinates of the field monitoring points to the unified coordinate system of the deep foundation pit digital twin model according to the aforementioned coordinate transformation relationship. The computational grid binding subcomponent binds the transformed field monitoring point coordinates to adjacent twin computational grids and records the distance from the monitoring point to the center of each adjacent twin computational grid. If the distance between the monitoring point and the center of a twin computational grid is the smallest, then that twin computational grid is used as the primary binding grid; if the monitoring point is located between two or more twin computational grids, the binding weight is determined by the inverse of the distance and the grid adjacency relationship. When there are monitoring gaps between adjacent twin computational grids, the monitoring gap completion subcomponent generates complete monitoring values ​​and corresponding confidence levels for state vector construction based on spatial adjacency relationships, deformation trends at the same stage, and historical samples from the same stage.

[0053] The monitoring gap completion sub-component does not directly use a single mean for filling, but instead combines the monitoring trends of adjacent computational grids, the predicted values ​​of the digital twin model, and historical samples from the same period for calculation. Among these, the estimated values ​​from adjacent twin computational grids... The interpolation value is obtained by performing inverse distance-weighted interpolation on adjacent grids with bound monitoring data around the missing calculation location. The interpolation weight is inversely proportional to the square of the distance from the monitoring point to the center of the missing grid. The number of adjacent grids participating in the interpolation does not exceed 8, and at least 3 valid monitoring values ​​are required. (Digital twin model predicted value) The values ​​are provided by the monitoring type corresponding to the simulation state vector output by the current deep foundation pit digital twin model at the missing mesh locations. Historical values ​​from samples at the same stage are also included. Retrieve from the calibration records of completed excavation phases, selecting samples from the same phase that share the same geological stratum type, excavation depth range, and support boundary conditions as the current phase. Take the weighted average of the monitoring values ​​at the corresponding grid locations, assigning weights based on the geological similarity between the sample and the current phase. This similarity is determined by the degree of geological stratum type matching and the proportion of overlap in depth ranges. The supplementary monitoring values ​​are completed using the following formula:

[0054] in, This represents the supplementary monitoring value at the nth calculation location during the hth excavation stage. This represents the inferred value from adjacent twin computational grids. This represents the predicted value from the digital twin model. This represents the estimated value from samples from the same historical period. , and These represent the weights corresponding to adjacent trends, twin predictions, and historical samples, respectively. The adjacent trend weight is determined by the number of adjacent grids participating in the interpolation and their average distance from the missing grids. First, the quantity factor and the distance factor are calculated separately. The quantity factor is the ratio of the number of adjacent grids currently participating in the interpolation to the maximum allowed number of grids (8). The distance factor is the normalized reciprocal between the average distance and the preset maximum influence distance. Then, the result of 0.5 × quantity factor + 0.5 × distance factor is linearly mapped to the range of 1 to 3, and the result outside the range is limited to 1 or 3. The twin prediction weights are determined by the confidence level of the digital twin model at the corresponding grid position in the current stage. The parameter version stability flag is calculated based on the parameter inversion results of multiple consecutive excavation stages. When the change rate of the corresponding parameter in three or more consecutive excavation stages is lower than the preset change threshold of 5%, and the residual target value continues to decrease, the parameter version stability flag is assigned a value of 1. When the change rate of the parameter in any stage exceeds the preset change threshold or the residual target value rebounds, the parameter version stability flag is assigned a value of 0. The historical fluctuation amplitude of the residual is calculated using the standard deviation of the residual values ​​of the most recent five consecutive excavation stages in the current computing grid. The parameter version stability flag and the historical fluctuation amplitude of the residual are then normalized, and the confidence level is calculated according to the ratio of 60% for the parameter version stability flag and 40% for the historical fluctuation amplitude of the residual. The twin prediction weights are then obtained by linear mapping based on the confidence level. The weight of historical samples is determined by the number of historical samples retrieved from the same period and the average stratigraphic similarity. First, the sample quantity factor and the stratigraphic similarity factor are calculated. The sample quantity factor is the ratio of the number of valid historical samples retrieved from the same period to the preset maximum number of samples. The stratigraphic similarity factor is the average stratigraphic similarity of all valid historical samples from the same period. Then, the result of the calculation of 0.5 × sample quantity factor + 0.5 × stratigraphic similarity factor is linearly mapped to a value between 0.5 and 2 as the weight of historical samples.

[0055] All three weights are dimensionless proportional coefficients. The completed monitoring value is calculated as a weighted average of the three estimated values, where the numerator is the sum of the products of each estimated value and its corresponding weight, and the denominator is the sum of the three weights. For example, when the adjacent trend weight is 2, the twin prediction weight is 3, the historical sample weight is 1, the adjacent twin calculation grid estimated value is 10 mm, the digital twin model prediction value is 11 mm, and the historical same-stage sample estimated value is 8 mm, the calculated completed monitoring value is 10.17 mm. The completed monitoring value and the corresponding completed confidence level are written into the state vector. The completed confidence level is calculated based on the proportion of the three weights and the dispersion among the data involved in the completed data. The smaller the dispersion of the data involved in the completed data and the higher the corresponding weight, the higher the completed confidence level. The values ​​for the three weights are determined based on the following: The adjacent trend weight is determined by the number of neighboring grids participating in the interpolation and their average distance from missing grids. The adjacent trend weight is calculated using both the number of neighboring grids and the average distance. First, a quantity factor and a distance factor are calculated separately, then a weighted sum is applied. The quantity factor is the ratio of the number of neighboring grids currently participating in the interpolation to the maximum allowed number of grids (8). The distance factor is the normalized reciprocal of the average distance and the preset maximum influence distance. Finally, the result of 0.5 × quantity factor + 0.5 × distance factor is linearly mapped to a value between 1 and 3, with results outside this range limited to 1 or 3. Therefore, the more neighboring grids and the closer the average distance, the closer the adjacent trend weight is to 3; conversely, the closer it is to 1. The confidence level of the digital twin model at the current grid location is determined by the parameter version stability flag, which is calculated based on the parameter inversion results of multiple consecutive excavation stages. When the rate of change of the corresponding parameter is below a preset threshold of 5% in three or more consecutive excavation stages, and the residual target value continues to decrease, the parameter version stability flag is assigned a value of 1; when the rate of change of the parameter exceeds the preset threshold or the residual target value rebounds in any stage, the parameter version stability flag is assigned a value of 0. The historical fluctuation amplitude of the residual is calculated using the standard deviation of the residual values ​​of the most recent five consecutive excavation stages in the current computational grid; a smaller standard deviation indicates more stable historical fluctuations. Subsequently, the parameter version stability flag and the historical fluctuation amplitude of the residual are normalized separately, with the parameter version stability flag accounting for 60% and the historical fluctuation amplitude of the residual accounting for 40%, to calculate the confidence level. The twin prediction weight is then linearly mapped according to the confidence level; the higher the confidence level, the larger the twin prediction weight. The weight of historical samples is determined by the number of historical samples retrieved from the same period and the average stratigraphic similarity. Specifically, the sample quantity factor and the stratigraphic similarity factor are calculated first. The sample quantity factor is the ratio of the number of valid historical samples retrieved from the same period to the preset maximum sample size, and the stratigraphic similarity factor is the average stratigraphic similarity of all valid historical samples from the same period. Then, a linear mapping is performed using the result of the calculation of 0.5 × sample quantity factor + 0.5 × stratigraphic similarity factor, and the result is limited to between 0.5 and 2 as the weight of historical samples. The more samples and the higher the average stratigraphic similarity, the closer the weight of historical samples is to 2; conversely, the closer it is to 0.5.

[0056] In this embodiment, the data sources and constraints for monitoring missing data completion are shown in Table 2.

[0057] Table 2. Data Sources and Constraints for Monitoring Gap Filling

[0058] In this embodiment, the completion confidence level is not used as an alarm result, but rather as an auxiliary field for residual calculation and parameter inversion. When the confidence level corresponding to the completed monitoring value is low, the simulation residual calculation component can reduce the weight of this data in the objective function; when the confidence level corresponding to the completed monitoring value is high, this data can participate in the residual matrix generation as part of the state vector. In this way, monitoring omissions will not directly cause model calculation interruption, nor will simple completion lead to parameter inversion being driven by erroneous data. This implementation method enables on-site monitoring data to continuously enter the deep foundation pit digital twin model through spatial mapping, mesh binding, and constrained completion.

[0059] In a preferred embodiment, the parameter inversion component includes a parameter candidate space construction subcomponent, a constraint optimization subcomponent, and a parameter combination output subcomponent. The parameter candidate space construction subcomponent sets the soil equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient as parameters to be corrected, and generates corresponding physical constraint intervals based on the stratum type, excavation stage, and support boundary conditions. The soil equivalent stiffness is used to express the equivalent deformation resistance of the stratum in the current excavation stage; the excavation unloading coefficient is used to express the impact of excavation unloading on the deformation field in the current stage; and the support constraint reduction coefficient is used to express the equivalent constraint change of the support boundary conditions in the current stage. The parameter candidate space does not use a fixed global range, but generates local constraints according to stratum properties and construction stages, allowing the same parameter to have different search ranges in different strata and different excavation stages.

[0060] In this embodiment, the parameter inversion object and the source of constraints are shown in Table 3.

[0061] Table 3. Parameter Inversion Objects and Constraint Sources

[0062] Furthermore, the constrained optimization subcomponent includes a hierarchical inversion subcomponent and a coupled constraint subcomponent. The hierarchical inversion subcomponent first performs grouped inversion of the equivalent soil stiffness corresponding to different strata, then performs staged inversion of the excavation unloading coefficients corresponding to different excavation stages, and finally performs local inversion of the support constraint reduction coefficients corresponding to the support boundary. The order here is computational dependency, not a fixed manual operation sequence. The equivalent soil stiffness, as a stratum attribute layer parameter, has a large spatial impact due to its variation; the excavation unloading coefficient, as a stage time axis parameter, mainly affects the current stage and subsequent stages; and the support constraint reduction coefficient, as a boundary constraint layer parameter, mainly affects the calculation location near the support boundary. The coupled constraint subcomponent applies parameter continuity constraints, excavation stage progression constraints, and support boundary consistency constraints between the above grouped inversion, staged inversion, and local inversion to prevent abrupt changes in a local parameter that do not conform to the relationship between adjacent areas in order to reduce residuals.

[0063] In this embodiment, parameter continuity constraints are used to ensure that the changes in equivalent soil stiffness between adjacent computational grids within the same stratum maintain an interpretable spatial transition; excavation stage progression constraints are used to ensure that the excavation unloading coefficient matches the excavation stage sequence, preventing subsequent stage inversion results from negatively impacting completed stages; and support boundary consistency constraints are used to ensure that the reduction coefficient on the same continuous support boundary does not exhibit unfounded local abrupt changes. All of these constraints are implemented through parameter candidate space pruning and objective function penalty terms. If a candidate parameter combination can reduce local residuals but causes discontinuities in parameter changes between adjacent grids, the constraint optimization sub-component removes the candidate combination from the available space or increases its objective function value. This process ensures that the parameter inversion results are driven by monitoring residuals and constrained by stratum spatial relationships, stage progression relationships, and boundary continuity relationships.

[0064] In a preferred embodiment, the parameter combination output subcomponent includes a parameter stability evaluation subcomponent and a boundary proximity recording subcomponent. The parameter stability evaluation subcomponent performs a sequence consistency evaluation on the parameter inversion results output from multiple consecutive excavation stages and generates parameter fluctuation characteristics. Parameter fluctuation characteristics include the direction of change, magnitude of change, and consistency markers between adjacent stages for the same parameter within consecutive stages. The boundary proximity recording subcomponent records the degree of proximity of the parameter inversion results to the upper or lower limit of the physical constraint interval and writes the parameter fluctuation characteristics and proximity degree into the stage correction record of the deep foundation pit digital twin model. If a parameter continuously approaches the physical constraint boundary within consecutive stages, the system does not directly use it as the final risk conclusion, but instead submits this state to the residual type marking subcomponent for joint judgment with the sensitivity perturbation results. Through this design, the parameter inversion results have traceable stage records, and the model correction process can trace back the parameter source of each stage.

[0065] In this embodiment, parameter stability evaluation does not use a single threshold judgment, but adopts a stage sequence recording. For the equivalent stiffness of soil within the same stratum, if the direction of the inversion results in consecutive stages is consistent and the residual decreases accordingly, the parameter combination output sub-component records it as a stable correction; if the direction of the inversion results changes repeatedly and the residual decreases unstablely, it is recorded as a fluctuation correction; if the parameter inversion results continuously approach the boundary of the physical constraint interval but the residual still does not converge, it is recorded as a boundary-constrained correction. This record then enters the residual type judgment and high uncertainty evolution region calibration process. This implementation, through hierarchical coupled inversion, parameter stability evaluation, and boundary proximity recording, provides a staged, spatialized, and traceable data foundation for the parameter updates of the digital twin model.

[0066] In a preferred embodiment, the simulation residual calculation component includes a residual matrix generation subcomponent, a residual decomposition subcomponent, and a residual type labeling subcomponent. The residual matrix generation subcomponent generates a multidimensional residual matrix according to the monitoring type, spatial location, and excavation stage. Each element in the multidimensional residual matrix corresponds to a specific monitoring type, a twin computation location, and an excavation stage. The residual decomposition subcomponent, based on the multidimensional residual matrix, decomposes the residuals into spatially continuous residuals, stage-accumulated residuals, and locally abrupt residuals. Spatially continuous residuals represent the situation where adjacent computational grids exhibit the same direction of deviation in the same stage; stage-accumulated residuals represent the situation where residuals continuously accumulate at the same computational location across multiple consecutive excavation stages; and locally abrupt residuals represent the situation where abrupt changes occur at a single location or a few locations within a short time window. The residual type labeling subcomponent labels the above residuals as candidate types for model parameter deviation, construction disturbance deviation, or monitoring anomaly deviation, respectively.

[0067] In this embodiment, the residual decomposition and candidate type labeling are shown in Table 4.

[0068] Table 4 Residual Decomposition and Candidate Type Labeling Table

[0069] Furthermore, when processing spatially continuous residuals, the residual decomposition subcomponent reads the adjacent grid sets for each calculation location and calculates whether the residual directions of adjacent grids are consistent. If the residual directions of adjacent grids are consistent and the residual amplitude changes continuously with spatial location, the residual is closer to the model parameter deviation; if only a single calculation location shows a sudden jump, and adjacent calculation locations do not show a corresponding response, the residual is closer to the monitoring anomaly deviation or local construction disturbance deviation. When processing stage-accumulated residuals, the residual decomposition subcomponent reads the residual sequence of the same calculation location within consecutive stages. If the residual direction remains consistent and gradually accumulates with the progress of the excavation stage, the residual is included in the excavation unloading coefficient inversion; if the residual direction changes repeatedly, it is verified by combining the completion confidence and monitoring data reliability. This processing transforms the residuals from being used as a single difference value into structured data with spatial and stage relationships.

[0070] In a preferred embodiment, the residual type labeling subcomponent includes a sensitivity perturbation subcomponent and an uncertain region calibration subcomponent. The sensitivity perturbation subcomponent applies single-parameter perturbations to the soil equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient, respectively, generating the residual sensitivity of each parameter to be corrected to the multidimensional residual matrix. The residual sensitivity is calculated as follows:

[0071] in, This represents the residual sensitivity of the g-th parameter to be corrected in the h-th excavation stage. This represents the multidimensional residual matrix at the h-th excavation stage when the g-th type of parameter takes its current value. This indicates the increase in perturbation amount for the g-th type parameter. The resulting multidimensional residual matrix Denotes the Frobenius norm of a matrix. This represents the absolute value of the perturbation. Its physical meaning lies in measuring the strength of the effect of a certain type of parameter change on the residual matrix. For example, when the Frobenius norm of the difference between the residual matrices before and after the perturbation is 3, When the absolute value of is 0.5, the residual sensitivity is 6. If the residual in a certain region increases continuously but is not sensitive to the perturbations of each parameter to be corrected, then this region is identified as a candidate region for high uncertainty evolution region.

[0072] The spatial distribution of high-uncertainty areas is used to put single numerical results back into the operational context of excavation stage, spatial grid, and monitoring type. The horizontal axis represents the line grid number, and the vertical axis represents the stage or depth index.

[0073] In this embodiment, the uncertain region calibration sub-component writes residual sensitivity, records of continuous residual increase, parameter boundary proximity, and completion confidence into the high uncertainty evolution region calibration table. If the residual of a certain region continuously increases and the parameter inversion result continues to approach the physical constraint boundary, it indicates that the current parameter space to be corrected may be insufficient to explain the deformation of that region; if the residual continuously increases but the sensitivity perturbation result is low, it indicates that it is difficult to reduce the residual by continuing to adjust the existing parameters; if the residual changes abruptly and the data confidence is low, it is given priority as a candidate for monitoring abnormal deviations. After the high uncertainty evolution region is calibrated, the on-site control object is not directly changed, but is written into the digital twin calculation boundary of the subsequent excavation stage, so that the subsequent simulation calculation retains a higher uncertainty mark and a more complete stage correction record in this region. This implementation transforms the residual from a simple alarm basis into a data source for model correction and subsequent calculation boundary updates through residual decomposition, sensitivity perturbation, and uncertain region calibration.

[0074] In a preferred embodiment, the model calibration component employs a parameter version management approach when updating the calculation parameters of the deep foundation pit digital twin model. After parameter inversion is completed at each excavation stage, the model calibration component generates a parameter version record. This record includes the excavation stage identifier, the set of monitoring locations involved in the calculation, the residual target value, the parameter inversion results, parameter stability markers, boundary proximity records, and high-uncertainty evolution zone markers. When performing simulation calculations for the next stage, the deep foundation pit digital twin model reads the latest parameter version from the previous stage as the initial calculation parameters, while retaining the initial parameters from the design stage as a backtracking benchmark. If subsequent stages reveal conflicts between the parameter inversion results and historical stages, the model calibration component can call upon the stage calibration record to recalculate the residual source of the conflicting stage without overwriting the original monitoring data and original simulation output. This process ensures data chain continuity in the model calibration process.

[0075] Furthermore, the corrected excavation pit deformation state distribution output by the model correction component includes the horizontal displacement distribution, vertical settlement distribution, support deformation distribution, residual distribution, and uncertainty markers for each twin computational grid at the current excavation stage. The corrected state distribution does not output from a single monitoring point, but rather from a continuous twin computational grid, enabling areas without monitoring points to obtain state estimates based on adjacent grids, digital twin simulation results, and stage correction records. For grids with completed monitoring values, the output results are simultaneously labeled with the completion confidence level; for parameter regions with abnormal boundary proximity, the output results are simultaneously labeled with the parameter constraint state; for locations identified as high-uncertainty evolution regions, the output results are simultaneously labeled with the computational boundary update information for subsequent excavation stages. This implementation allows the corrected deformation state distribution to express monitoring constraints, simulation constraints, and parameter uncertainties.

[0076] In one embodiment, the intelligent monitoring method for deep foundation pit excavation on highways can be executed by the aforementioned system. The execution process is as follows: A digital twin model of the deep foundation pit is constructed based on the pit design outline, soil parameters, excavation stage, and support boundary conditions; field monitoring data is mapped to the digital twin model according to the excavation stage and spatial location, generating a foundation pit deformation state vector; the simulation state vector output by the digital twin model is obtained, and the residual between the foundation pit deformation state vector and the simulation state vector is calculated; within a preset physical constraint range, the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient are inverted using the residual as the target; the calculation parameters of the digital twin model are updated based on the inversion results, and the corrected foundation pit deformation state distribution is output. The above processes are completed within the same data chain, and the field monitoring data, simulation output data, residual matrix, and parameter inversion results all carry excavation stage and spatial location identifiers.

[0077] In this embodiment, the method first reads the design outline of the foundation pit and generates a twin computational grid, then initializes the deep foundation pit digital twin model based on the stratum parameters and support boundary conditions. After the field monitoring data enters the system, the method does not directly issue a threshold alarm, but instead writes each monitoring data point into the excavation stage index and spatial mapping results. If the coordinates of the monitoring point do not completely coincide with the center of the twin computational grid, the method maps it to the corresponding calculation location through coordinate normalization and grid binding; if some calculation locations lack monitoring data, the method generates supplementary monitoring values ​​with confidence through adjacent grid trends, digital twin predictions, and historical samples from the same stage. Subsequently, the method converts the monitoring data or supplementary monitoring values ​​into a foundation pit deformation state vector, and reads the simulation state vector output by the deep foundation pit digital twin model to perform residual calculation.

[0078] Furthermore, after obtaining the residuals, the method generates multidimensional residual matrices according to monitoring type, spatial location, and excavation stage, and decomposes the residuals based on spatial continuity, stage accumulation, and local abrupt changes. For spatially continuous residuals, the method uses them as important inputs for inverting the equivalent stiffness of the soil or the reduction coefficient of the support constraint; for stage-accumulated residuals, the method uses them as important inputs for inverting the excavation unloading coefficient; for locally abrupt residuals, the method combines the completion confidence, the response of adjacent grids, and the results of sensitive disturbances to label candidate types. During parameter inversion, the method searches for parameter combinations that minimize the target value of the normalized residual within the physical constraint space, and restricts unreasonable jumps in parameters between the geological space, excavation stage, and support boundary through hierarchical inversion and coupling constraints.

[0079] In this embodiment, after outputting the parameter inversion results, the method writes the inversion results into the stage parameter table of the digital twin model and simultaneously generates parameter stability evaluation and boundary proximity records. If the parameter inversion results of the current stage are stable, the method uses them as the initial parameters for the next stage of simulation calculation; if the parameter inversion results of the current stage are in a boundary-constrained state, the method writes the relevant calculation location into the high-uncertainty evolution region; if the residual of the current stage is not sensitive to the parameters to be corrected, the method writes the sensitive perturbation results into the digital twin calculation boundary of the subsequent excavation stage. This method enables the deep foundation pit digital twin model to continuously receive on-site monitoring constraints during the excavation process and update the calculation parameters through parameter inversion, maintaining a continuous correspondence between the model calculation state and the on-site excavation state.

[0080] In a preferred embodiment, the system can also use historical samples from the same stage as auxiliary data for model calibration. These historical samples are not generalized engineering experience data, but rather consist of records of completed excavation stages in the same or similar projects. Sample fields include stage code, geological attributes, excavation depth range, support boundary status, monitoring state vector, simulation state vector, residual matrix, and parameter inversion results. When using historical samples from the same stage, the system first filters samples based on stage code, geological attributes, and support boundary status, and then uses the filtered samples to complete monitoring value calculations or set the initial candidate space for parameters. If a historical sample does not match the geological attributes or support boundary status of the current stage, the sample will not participate in the calculation for the current stage. This process prevents historical samples from deviating from the current excavation state and affecting parameter inversion.

[0081] Furthermore, historical samples from the same stage are only considered as one of the multi-source constraints when participating in the calculation of supplementary monitoring values. The system will not directly replace field monitoring data with historical samples, nor will it cover the simulation output of the current stage of the digital twin model with historical samples. When field monitoring data is sufficient and has a high confidence level, the weight of historical samples is reduced; when there are gaps in field monitoring data and insufficient information from adjacent grids, historical samples can provide a reference for the stage change trend. This approach ensures that the data supplementation process is simultaneously constrained by the current monitoring state, the current simulation state, and the patterns of historical stages, reducing the bias caused by a single data source. This implementation method enables the deep foundation pit excavation digital twin model to maintain continuous calculation even when data is missing, while retaining the source record of each supplementary value.

[0082] In a preferred embodiment, the system verifies the candidate types of monitoring anomalies. The verification process does not rely on manual experience descriptions but is determined jointly by data confidence, adjacent grid responses, and residual sensitivity. If a monitoring point exhibits a local abrupt residual change, but adjacent grids do not show a corresponding response in the same direction, and the deformation field output by the digital twin model at the same stage does not show a corresponding continuous spatial change, and the historical data for that monitoring point shows short-term interruptions or drift records, then the system marks it as a candidate for monitoring anomalies. If, after a local abrupt residual change occurs, adjacent grids show a corresponding change in the same direction in subsequent stages, and the residual is sensitive to the excavation unloading coefficient or support constraint reduction coefficient, then the system changes it from a candidate for monitoring anomalies to a candidate for construction disturbance deviation or a candidate for model parameter deviation. This processing allows the anomaly type to be adjusted as subsequent data is updated.

[0083] In this embodiment, candidate abnormal deviations are not directly deleted; instead, the original value, substitute value, confidence level, and reason for the deviation are retained. During residual calculation, both the original and substitute values ​​can be entered into the record table, but the values ​​used for parameter inversion are determined based on the confidence level and candidate type. If the original value is marked as having low confidence, the completed value has independent weight when participating in the objective function calculation. If subsequent stages prove that the original value's evolution is consistent with that of adjacent regions, the system can restore the original value's participation status in the stage correction record. This approach avoids the loss of true early deformation information due to simply removing abnormal data, and also avoids incorrect correction of model parameters due to directly using abnormal data. This implementation ensures consistency between the monitoring data governance process and the digital twin model calculation process.

[0084] In a preferred embodiment, the computational grid of the deep foundation pit digital twin model can be segmented according to the route direction of the highway foundation pit, maintaining the grid state corresponding to the excavation stage within each segment. The route direction segmentation represents the spatial continuity of long-distance excavation of the highway deep foundation pit, while the excavation stage represents the impact of construction sequence on the deformation field. When calculating the residual matrix, the system treats adjacent grids within the same segment as spatially adjacent objects, and grids with support continuity relationships between different segments as boundary adjacent objects. If adjacent segments are at different excavation stages, the residual matrix generation sub-component calculates the residual according to their respective stage codes and records the stage differences in the cross-segment relationship. This process avoids the mixing of data from different construction zones in long-distance foundation pits.

[0085] Furthermore, when a segment is designated as a high-uncertainty evolution region, the system does not mark the entire line as being in the same state. Instead, it transmits the uncertainty boundary to adjacent segments based on spatial adjacency. The transmission range is jointly defined by residual continuity, support boundary consistency, and stage differences. If an adjacent segment has a continuous support boundary with the high-uncertainty region and the excavation stages are close, the residual record density of that adjacent segment is increased in subsequent simulation calculations. If the geological properties of adjacent segments are different or the stage differences are significant, only the adjacency relationship is recorded, and the high-uncertainty label is not directly inherited. This processing ensures that the labeling of the high-uncertainty evolution region has spatial boundaries, preventing overly broad risk expansion judgments. This implementation is adapted to the engineering data structure of long linear, segmented excavation of deep foundation pits for highways.

[0086] In a preferred embodiment, the simulation state vector of the digital twin model can be output by a deformation calculation program based on computer-aided engineering or by a calibrated proxy calculation model. Regardless of the calculation method used, the simulation state vector must have the same field order, the same stage index, and the same spatial location index as the monitoring state vector. If the proxy calculation model is involved in the output, the training samples come from stage calibration records and historical simulation outputs. The input fields include geological properties, excavation stage, support boundary state, and calibration parameters of the previous stage. The output fields include simulated values ​​of horizontal displacement, vertical settlement, and support deformation. After the proxy calculation model outputs, it still needs to enter the residual calculation and parameter inversion process and is not used as an independent judgment criterion. This process maintains a controllable relationship between calculation speed and calculation consistency.

[0087] In this embodiment, training the proxy computation model does not alter the basic data structure of the digital twin model. Before training, the system filters out abnormal candidate data in the stage correction records according to confidence level, excluding low-confidence data from the primary training samples. After training, the system compares the output of the proxy computation model with the original simulation output under the same input conditions. If the difference between the output of the proxy computation model and the original simulation output exceeds the allowable model consistency range, the output of the proxy computation model will not participate in the parameter inversion of the current stage. This approach ensures that the proxy computation model serves only as one implementation of simulation state generation, without altering the core digital twin correction logic centered on monitoring state, simulation residuals, and parameter inversion. This implementation extends the computational implementation path of the deep foundation pit digital twin model while maintaining consistency in the residual correction chain.

[0088] In a preferred embodiment, the system's human-computer interface only displays the structured results generated by the computational chain, and does not directly accept manually input risk conclusions overriding the model output. The interface can display the excavation stage, twin computational grid, monitoring state vector, simulation state vector, residual matrix, parameter inversion results, parameter stability records, boundary proximity records, and high uncertainty evolution regions. If engineers need to correct a data field, the system requires that the correction be bound to the correction source, correction time, and correction field, and retain the data before correction. The corrected data must undergo stage indexing and spatial mapping verification before entering the state vector construction. This design ensures that the human-computer interaction process does not disrupt the original data chain, nor bypass the parameter calibration process of the digital twin model.

[0089] Furthermore, the system can display the corrected foundation pit deformation state distribution in the interface, organized by computational grid. Each computational grid displays monitored or supplemented values, simulated values, residual values, inversion status of participating parameters, and uncertainty markers. For grids with monitoring gaps that need to be filled, the interface displays the combination of supplementation sources; for areas where parameter boundaries are close to records, the interface displays the corresponding parameter category; for areas where the residual type is not yet stable, the interface displays candidate types rather than definitive conclusions. In this way, the interface display content remains consistent with the underlying computational data, avoiding the substitution of specific technical information with a single color or level. This implementation provides the system output with a traceable data structure, facilitating the review of the digital twin model's phased correction process.

[0090] In a preferred embodiment, the system can also be configured with stage playback and recalculation logic. Stage playback reads the original monitoring data, simulation state vector, residual matrix, and parameter inversion results of a specific historical excavation stage, and reconstructs the correction process according to the original stage parameter version. The recalculation logic allows parameter inversion to be performed again using a new physical constraint space or an updated historical sample library without changing the original records, and the recalculation results are saved as an independent version. Both the original stage version and the recalculated version can be compared, but they will not overwrite each other. This method is suitable for verifying the data chain of historical stages when a monitoring anomaly candidate or a high-uncertainty evolution region is subsequently discovered in a certain stage. Through stage playback and recalculation, the system can maintain the traceability of historical data and prevent subsequent correction results from overwriting previous calculations.

[0091] In this embodiment, stage replay can also be used to verify the stability of the parameter inversion process. If a stage yields similar parameter inversion results using both the original parameter space and the recalculated parameter space, and the residual type labels are consistent, then the correction record for that stage is marked as a stable record. If the parameter category changes significantly after recalculation, the system includes that stage along with its preceding and following adjacent stages in the verification set. The verification set does not directly change the on-site monitoring conclusions but is used by the subsequent model calibration component to adjust the reference relationships of stage parameter versions. This processing enables the deep foundation pit digital twin model to maintain multiple parameter versions during long-term construction and to explain the sources of parameter changes between different stages. This implementation enables model calibration to have long-term construction data management capabilities.

[0092] In a preferred embodiment, after completing the correction for the current stage, the system performs predictive simulation for the next excavation stage. The predictive simulation reads the equivalent soil stiffness, excavation unloading coefficient, and support constraint reduction coefficient after the correction for the current stage, and generates a simulation state vector for the next stage by combining this with the planned excavation depth, excavation zones, and support boundary states. If a high-uncertainty evolution region exists in the current stage, the predictive simulation for the next stage retains an uncertainty marker in the corresponding region and displays the residual source, sensitive parameters, and boundary proximity in the output results. The predictive simulation results do not replace the field monitoring data for the next stage; instead, they are used as the simulation state vector in residual calculation after the monitoring data for the next stage is received. This process ensures a continuous connection between model correction and the calculation for the next stage.

[0093] Furthermore, after the next stage of on-site monitoring data enters the system, the system calculates the residual between the predictive simulation state vector and the actual monitored state vector. If the predicted residual is within the allowable calculation range, the current stage parameter version can continue to be used as the initial parameters for the next stage; if the predicted residual is concentrated in a high-uncertainty evolution region, the system adjusts the parameter inversion weights and residual record density for that region; if the predicted residual is distributed in an uncalibrated region, the system re-decomposes the residual and updates the uncertainty calibration. In this way, the system can verify the adaptability of the previous stage model correction to the next stage and continuously update the digital twin model based on the new stage monitoring data. This implementation method enables the intelligent monitoring process of deep foundation pit excavation to have stage continuity and verifiability.

[0094] In a preferred embodiment, the system's data storage employs a hierarchical structure. The original layer stores unmodified field monitoring data, the excavation pit design outline, soil parameters, excavation stages, and support boundary conditions; the mapping layer stores coordinate normalization results, twin mesh binding results, and stage index results; the computational layer stores the excavation pit deformation state vector, simulation state vector, multidimensional residual matrix, residual decomposition results, and sensitivity perturbation results; and the calibration layer stores parameter inversion results, model parameter versions, parameter stability records, boundary proximity records, and high uncertainty evolution region markers. The layers are linked through spatial location identifiers and excavation stage identifiers, and the computational results are updated without overwriting the original data. This hierarchical structure ensures that each model calibration is traceable back to the corresponding original data and computational process.

[0095] In this embodiment, if the on-site monitoring data arrives late, the system writes the delayed data into the original layer and re-establishes the stage index based on its monitoring time and excavation stage. If the excavation stage corresponding to the delayed data has already completed parameter correction, the system will not directly overwrite the existing correction results, but will generate a recalculation task. The recalculation task reads the delayed data, the original stage parameter version, and the original stage residual matrix, recalculates the state vector, residual matrix, and parameter inversion results for that stage, and saves the new results as the recalculated version. If the recalculated version differs slightly from the original version, the parameter reference relationship for subsequent stages remains unchanged; if the recalculated version changes the high uncertainty evolution region or the proximity of parameter boundaries, subsequent stages can choose to reference the recalculated version for recalculation. This process allows delayed data to enter the digital twin correction chain while keeping the original stage records intact. This implementation allows the system to adapt to situations where the arrival order of data at the construction site is unstable.

[0096] In a preferred embodiment, the system can use the digital twin model calibration results to generate a data verification report for the construction phase. The report content originates from internal system calculations, including the current stage's monitoring state vector, simulation state vector, residual matrix, parameter inversion results, residual type candidate labels, completed monitoring values ​​and their confidence levels, parameter stability records, and high-uncertainty evolution regions. The report does not generate judgment statements independent of the calculation chain, but instead outputs the state of each calculation location using structured fields. For data involved in parameter inversion, the report records its data source; for data not involved in parameter inversion, the report records the reasons for exclusion; for completed data, the report records the estimated values ​​of adjacent grids involved in completion, the digital twin predicted values, and the estimated values ​​of historical samples from the same stage. This report serves as a technical record of the model calibration process.

[0097] In this embodiment, the data verification report can also be used to compare parameter changes between different excavation stages. The system reads the stage correction records of consecutive stages and correlates the changes in soil equivalent stiffness, excavation unloading coefficient, and support constraint reduction coefficient with the changes in residual target values. If the residual target value decreases after a parameter change in a certain stage and the residual type label changes from spatial continuous residual to low residual state, the report records it as parameter correction closure; if the residual target value remains high after the parameter change and the parameter is close to the physical constraint boundary, the report records it as boundary-constrained state; if the parameter change has little impact on the residual matrix, the report records it as low-sensitivity state. The report content is all derived from system-calculated fields and does not generate additional conclusions that are not supported by data. This implementation method enables the deep foundation pit digital twin correction process to be retained in a structured manner.

Claims

1. An intelligent monitoring system for deep foundation pit excavation on highways, characterized in that, It includes digital twin modeling components, monitoring status construction components, simulation residual calculation components, parameter inversion components, and model calibration components; The digital twin modeling component is used to construct a digital twin model of a deep foundation pit based on the foundation pit design outline, stratum parameters, excavation stage, and support boundary conditions. The monitoring status construction component is used to convert on-site monitoring data into a foundation pit deformation status vector. The simulation residual calculation component is used to calculate the residual between the foundation pit deformation state vector and the simulation state vector output by the deep foundation pit digital twin model; The parameter inversion component is used to invert the parameters to be corrected in the model within a preset physical constraint range; The model correction component is used to update the calculation parameters of the deep foundation pit digital twin model according to the inversion results and output the corrected foundation pit deformation state distribution, forming an intelligent monitoring system for highway deep foundation pit excavation.

2. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 1, characterized in that, The monitoring status construction component includes an excavation stage index subcomponent, a spatial mapping subcomponent, and a status vector generation subcomponent. The excavation stage index sub-component is used to generate stage identifiers according to the excavation depth, excavation zones, and construction sequence. The spatial mapping sub-component is used to map the on-site monitoring data to the corresponding calculation location in the deep foundation pit digital twin model; The state vector generation sub-component is used to generate a foundation pit deformation state vector containing horizontal displacement, vertical settlement, support deformation increment, deformation rate, and stage code based on the stage identifier and the corresponding calculation position.

3. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 1, characterized in that, The parameter inversion component includes a parameter candidate space construction sub-component, a constraint optimization sub-component, and a parameter combination output sub-component. The parameter candidate space construction sub-component is used to set the soil equivalent stiffness, excavation unloading coefficient and support constraint reduction coefficient as parameters to be corrected, and to generate corresponding physical constraint intervals based on the stratum type, excavation stage and support boundary conditions. The constraint optimization sub-component is used to search for the combination of parameters to be corrected within the physical constraint interval, with the normalized value of the residual as the optimization objective. The parameter combination output sub-component is used to output the parameter inversion results that satisfy the constraints.

4. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 3, characterized in that, The simulation residual calculation component includes a residual matrix generation sub-component, a residual decomposition sub-component, and a residual type marking sub-component; The residual matrix generation sub-component is used to generate multi-dimensional residual matrices according to monitoring type, spatial location, and excavation stage. The residual decomposition sub-component is used to decompose the residual into spatially continuous residuals, stage cumulative residuals, and local abrupt residuals based on the multidimensional residual matrix. The residual type labeling sub-component is used to label the spatial continuous residual, the stage cumulative residual, and the local mutation residual as candidate types of model parameter deviation, construction disturbance deviation, or monitoring anomaly deviation, respectively.

5. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 2, characterized in that, The state vector generation subcomponent includes a temporal window configuration subcomponent and a multi-scale feature splicing subcomponent; The timing window configuration sub-component is used to configure short-time windows and cross-stage windows according to the continuous monitoring periods preceding the current excavation stage; The multi-scale feature splicing sub-component is used to extract deformation increment and deformation rate from the short-term window, extract deformation accumulation and stage change trend from the cross-stage window, and splice the deformation increment, deformation rate, deformation accumulation, stage change trend and stage code into a multi-scale foundation pit deformation state vector.

6. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 2, characterized in that, The spatial mapping subcomponent includes a coordinate normalization subcomponent, a computational grid binding subcomponent, and a monitoring and missing information completion subcomponent. The coordinate normalization sub-component is used to transform the coordinates of the on-site monitoring points to the unified coordinate system of the deep foundation pit digital twin model; The computational grid binding sub-component is used to bind the converted field monitoring point coordinates to adjacent twin computational grids; The monitoring gap completion sub-component is used to generate completion monitoring values ​​and corresponding confidence levels for state vector construction based on spatial adjacency and deformation trends at the same stage when there are monitoring gaps between adjacent twin computing grids.

7. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 3, characterized in that, The constraint optimization subcomponent includes a hierarchical inversion subcomponent and a coupled constraint subcomponent; The layered inversion sub-component is used to first perform grouped inversion of the equivalent stiffness of soil corresponding to different strata, then perform stage inversion of the excavation unloading coefficient corresponding to different excavation stages, and perform local inversion of the support constraint reduction coefficient corresponding to the support boundary. The coupling constraint sub-component is used to apply parameter continuity constraints, excavation stage progression constraints, and support boundary consistency constraints between the group inversion, the stage inversion, and the local inversion, generating layered coupled parameter inversion results.

8. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 3, characterized in that, The parameter combination output subcomponent includes a parameter stability evaluation subcomponent and a boundary proximity recording subcomponent. The parameter stability evaluation sub-component is used to evaluate the sequence consistency of parameter inversion results output from multiple consecutive excavation stages and generate parameter fluctuation characteristics. The boundary proximity recording sub-component is used to record the degree of proximity of the parameter inversion result to the upper or lower limit of the physical constraint interval, and to write the parameter fluctuation characteristics and the degree of proximity into the stage correction record of the deep foundation pit digital twin model.

9. The intelligent monitoring system for deep foundation pit excavation on highways according to claim 4, characterized in that, The residual type marking subcomponent includes a sensitivity perturbation subcomponent and an uncertainty region calibration subcomponent; The sensitive perturbation sub-component is used to apply single-parameter perturbations to the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient, respectively, to generate the residual sensitivity of each parameter to be corrected to the multidimensional residual matrix. The uncertainty region calibration sub-component is used to calibrate the calculation location where the residual sensitivity is lower than the preset sensitivity condition and the residual continuously increases as a high uncertainty evolution region, and write the high uncertainty evolution region into the digital twin calculation boundary of the subsequent excavation stage.

10. A method for intelligent monitoring of deep foundation pit excavation for highways, characterized in that, include: A digital twin model of the deep foundation pit is constructed based on the design outline, stratum parameters, excavation stage, and support boundary conditions. The on-site monitoring data is mapped to the deep foundation pit digital twin model according to the excavation stage and spatial location to generate the foundation pit deformation state vector; Obtain the simulation state vector output by the digital twin model of the deep foundation pit, and calculate the residual between the foundation pit deformation state vector and the simulation state vector; Within the preset physical constraints, the equivalent stiffness of the soil, the excavation unloading coefficient, and the support constraint reduction coefficient are inverted with the residual as the target. The calculation parameters of the deep foundation pit digital twin model are updated based on the inversion results, and the corrected foundation pit deformation state distribution is output, forming an intelligent monitoring method for deep foundation pit excavation of highways.

Images