Pile foundation quality evaluation method based on multi-source data fusion

Abstract

The application discloses a pile foundation quality evaluation method based on multi-source data fusion, which comprises the following steps: collecting multi-source monitoring data in the pile foundation construction process, processing the multi-source monitoring data, and generating a standardized monitoring data set; using a time series analysis module to perform trend prediction on the pile body temperature distribution, the pile body axial strain gradient and the pile body strain distribution by using the standardized monitoring data set, and generating a pile foundation settlement trend prediction result and a pile foundation strain trend prediction result; using a spatial distribution analysis module to perform spatial correlation analysis on the drilling verticality deviation, the soil pressure distribution around the pile and the horizontal displacement amount by using the standardized monitoring data set, and generating a pile foundation verticality deviation distribution diagram, a soil pressure distribution diagram around the pile and a pile foundation horizontal displacement distribution diagram; and combining the pile foundation settlement trend prediction result, the pile foundation strain trend prediction result, the pile foundation verticality deviation distribution diagram, the soil pressure distribution diagram around the pile and the pile foundation horizontal displacement distribution diagram to generate a pile foundation quality comprehensive evaluation report.

Description

Technical Field

[0001] This invention relates to the field of pile foundation quality assessment technology, and more specifically, to a pile foundation quality assessment method based on multi-source data fusion. Background Technology

[0002] Current pile foundation quality assessments often rely on single types of monitoring data or focus only on static parameters after construction, failing to integrate dynamic data from the construction process for analysis. Conventional techniques frequently neglect key dynamic parameters such as borehole verticality deviation, pile temperature distribution, and axial strain gradient, judging quality solely based on static indicators like pile top settlement and pile strength, which fails to reflect the true state of the pile foundation during construction and use.

[0003] Existing technologies lack systematic integration of multi-source data. Significant differences in units and magnitudes among different data types lead to a lack of comparability and hinder effective synergy. Furthermore, conventional assessment methods fail to conduct dynamic trend prediction and spatial correlation analysis. This makes it impossible to anticipate potential risks arising from changes in pile-related parameters over time, or to grasp quality differences across different parts of the pile foundation. Consequently, it is difficult to comprehensively and accurately identify existing hazards during the construction phase, and thus cannot meet the high requirements for pile foundation quality assessment in complex projects. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a pile foundation quality assessment method based on multi-source data fusion to meet the high requirements of complex engineering projects for pile foundation quality assessment.

[0005] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0006] The pile foundation quality assessment method based on multi-source data fusion includes the following steps: Collect multi-source monitoring data during pile foundation construction, including: borehole verticality deviation, pile body temperature distribution, pile body axial strain gradient, pile perimeter earth pressure distribution, pile body strain distribution, and horizontal displacement. The multi-source monitoring data is processed to generate a standardized monitoring dataset; Construct a pile foundation quality assessment model that includes a time series analysis module and a spatial distribution analysis module; The standardized monitoring dataset is used to predict the trends of pile body temperature distribution, pile body axial strain gradient and pile body strain distribution using the time series analysis module, generating pile foundation settlement trend prediction results and pile foundation strain trend prediction results. The standardized monitoring dataset is used to perform spatial correlation analysis on borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement using the spatial distribution analysis module, generating pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map. Based on the predicted results of pile foundation settlement trend, pile foundation strain trend, pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map, a comprehensive pile foundation quality assessment report is generated.

[0007] Furthermore, the process of processing the multi-source monitoring data to generate a standardized monitoring dataset includes: The maximum and minimum values ​​of each type of monitoring data are extracted from the multi-source monitoring data. The minimum-maximum scaling method is used to normalize each monitoring data to obtain the normalized value. Set the weight coefficient for each type of monitoring data, multiply the normalized value of each monitoring data by the corresponding weight coefficient, and generate weighted monitoring data. The generated weighted monitoring data are aligned according to timestamps and spatial locations to form a standardized monitoring dataset.

[0008] Furthermore, the generation process of the pile foundation settlement trend prediction results and pile foundation strain trend prediction results includes: The standardized monitoring data corresponding to the pile body temperature distribution, pile body axial strain gradient and pile body strain distribution are centrally input into the time series analysis module to extract time series features. Based on the aforementioned time series characteristics, predict the changes in pile body temperature, axial strain gradient, and strain within future time windows; The changes in pile body temperature, axial strain gradient, and strain are compared with preset warning thresholds, and combined with the intrinsic relationship between these factors and pile settlement and strain, to generate pile settlement trend prediction results and pile strain trend prediction results.

[0009] Furthermore, the methods for generating the pile foundation verticality deviation distribution map, the pile perimeter earth pressure distribution map, and the pile foundation horizontal displacement distribution map include: The standardized borehole verticality deviation, standardized pile perimeter earth pressure distribution, and standardized horizontal displacement from the standardized monitoring dataset are mapped onto the three-dimensional model of the pile foundation to generate three spatial distribution matrices, including: borehole verticality deviation matrix, pile perimeter earth pressure distribution matrix, and horizontal displacement matrix. The spatial distribution analysis module sequentially captures the spatial correlation of three spatial distribution matrices in the pile foundation mesh cells of the three-dimensional pile foundation model, and outputs the anomaly probability of the pile foundation mesh cells. The anomaly probabilities of the pile foundation grid cells are numerically fused with three spatial distribution matrices to generate pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map.

[0010] Furthermore, the construction process of the three-dimensional model of the pile foundation is as follows: An initial three-dimensional model of the pile foundation is created based on the geometric parameters of the pile body on the design drawings, and the coordinates of the actual measuring points monitored by multi-source monitoring data during the pile foundation construction process are aligned to the coordinate system of the three-dimensional model of the pile foundation. The boundary representation method is used to describe the geometry of the pile in the three-dimensional model of the pile foundation, and the layered structure of the soil around the pile is embedded in the three-dimensional model of the pile foundation.

[0011] Furthermore, the process for generating the comprehensive assessment report on pile foundation quality is as follows: The pile foundation settlement trend prediction results and pile foundation strain trend prediction results are weighted and integrated to form a time-dimensional evaluation result. The verticality deviation distribution map, the earth pressure distribution map around the pile, and the horizontal displacement distribution map of the pile foundation are correlated and calculated in each pile foundation grid cell of the three-dimensional model of the pile foundation to form a spatial dimension evaluation result. The time-dimensional and spatial-dimensional assessment results are mapped onto a unified scoring scale, and a weighted comprehensive score for pile foundation quality is calculated. The quality grade of the pile foundation during the construction process is determined based on the comprehensive score of the pile foundation quality.

[0012] Furthermore, the process of forming the spatial dimension assessment result is as follows:

[0013] in, Represents the pile foundation mesh element in the 3D model of the pile foundation. Spatial joint index value at the location, The distribution diagram of pile foundation verticality deviation in the grid cell The normalized value of the borehole verticality deviation. express The weighting coefficients, This represents the distribution of earth pressure around the pile in the pile foundation grid unit. The normalized value of the earth pressure around the pile. express The weighting coefficients, The distribution diagram of horizontal displacement of pile foundation in pile foundation grid cells The normalized value of the horizontal displacement. express The weighting coefficients.

[0014] Furthermore, the specific process of mapping the time-dimensional assessment results to the score scale is as follows: If the time dimension assessment result does not exceed the first time dimension warning threshold, 100 points are awarded; If the time dimension assessment result exceeds the first time dimension warning threshold but does not exceed the second time dimension warning threshold, based on a baseline of 100 points, points will be deducted according to the magnitude of the exceedance of the first time dimension warning threshold. The score range is as follows: ; If the time dimension assessment result exceeds the second time dimension warning threshold, with 80 points as the baseline, points will be deducted according to the magnitude of the excess. The score range is as follows. ; Among them, the warning threshold for the first time dimension is lower than the warning threshold for the second time dimension.

[0015] Furthermore, the specific process of mapping the spatial dimension assessment results to the score scale is as follows: If the anomaly probability of the pile foundation grid unit does not exceed 0.85 and the spatial dimension evaluation results do not exceed the preset benchmark value, 100 points are awarded. If the number of pile foundation mesh elements with an anomaly probability exceeding 0.85 does not exceed 10% of the total number of pile foundation mesh elements in the 3D pile foundation model, the maximum anomaly probability of the pile foundation mesh elements does not exceed 0.9, and the spatial dimension evaluation results do not exceed 120% of the preset benchmark value, a deduction of 100 points will be made, with the score range being [insert range here]. ; If the number of pile foundation mesh elements with an anomaly probability exceeding 0.85 accounts for more than 10% of the total number of pile foundation mesh elements in the 3D pile foundation model, the maximum anomaly probability of the pile foundation mesh elements exceeds 0.9, or the spatial dimension assessment result of some pile foundation mesh elements exceeds 120% of the preset benchmark value, a deduction of 80 points will be made, with the score range being [insert range here]. .

[0016] Furthermore, it also includes: Based on the borehole verticality deviation matrix, the pile peri-earth pressure distribution matrix, and the horizontal displacement matrix, respectively, the borehole verticality deviation exceedance coefficient, the pile peri-earth pressure anomaly coefficient, and the horizontal displacement concentration coefficient are generated; Set abnormal thresholds for borehole verticality deviation exceeding the standard coefficient, pile perimeter earth pressure abnormal coefficient, and horizontal displacement concentration coefficient respectively, and record the type and value of coefficients exceeding the abnormal thresholds in the comprehensive evaluation report of pile foundation quality.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention presents a pile foundation quality assessment method based on multi-source data fusion. It collects multi-source monitoring data, including borehole verticality deviation, pile temperature distribution, pile axial strain gradient, pile perimeter earth pressure distribution, pile strain distribution, and horizontal displacement. This covers key dynamic parameters during pile foundation construction and static parameters during use, overcoming the limitations of traditional assessment methods that rely on single data sources. Different types of monitoring data reflect pile foundation quality from multiple dimensions, such as construction operations, structural deformation, and stress state, enabling a more comprehensive capture of the quality characteristics of the pile foundation at different stages. This avoids assessment bias caused by incomplete data, resulting in pile foundation quality assessment results that more closely reflect the true quality state of the pile foundation.

[0018] This invention presents a pile foundation quality assessment method based on multi-source data fusion. By constructing a pile foundation quality assessment model that includes a time series analysis module and a spatial distribution analysis module, the time series analysis module predicts trends in pile temperature distribution, axial strain gradient, and strain distribution, enabling the understanding of key parameter changes over time and early identification of potential quality risks. The spatial distribution analysis module performs spatial correlation analysis on borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement, generating verticality deviation distribution maps, pile perimeter earth pressure distribution maps, and horizontal displacement distribution maps. These maps visually present the spatial quality differences in different parts of the pile foundation, helping to accurately locate weak areas and providing clear guidance for targeted rectification. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the working principle of the pile foundation quality assessment method based on multi-source data fusion of the present invention. Figure 2 A flowchart for the construction and verification of a pile foundation quality assessment model; Figure 3 A flowchart for spatial distribution analysis and report generation. Detailed Implementation

[0020] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] Please see Figure 1This invention provides a pile foundation quality assessment method based on multi-source data fusion, comprising: collecting multi-source monitoring data during pile foundation construction, including borehole verticality deviation, pile body temperature distribution, pile body axial strain gradient, pile periphery earth pressure distribution, pile body strain distribution, and horizontal displacement, covering key stages of pile foundation construction from drilling to pile formation. The collected multi-source monitoring data is normalized to eliminate dimensional differences and unify data scales, generating a standardized monitoring dataset, which serves as the basis for subsequent analysis. A pile foundation quality assessment model is constructed based on the standardized monitoring dataset. This model includes a time series analysis module and a spatial distribution analysis module. The time series analysis module focuses on predicting dynamic trends, while the spatial distribution analysis module handles geometric and location-related features. The time series analysis module is used to predict the trends of pile body temperature distribution, pile body axial strain gradient, and pile body strain distribution. By analyzing historical data sequences, future changes are inferred, generating pile foundation settlement trend prediction results and pile foundation strain trend prediction results. Simultaneously, the spatial distribution analysis module performs spatial correlation analysis on borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement, identifying abnormal areas and generating verticality deviation distribution maps, pile perimeter earth pressure distribution maps, and horizontal displacement distribution maps. Combining the results of time series and spatial analysis, a comprehensive pile foundation quality assessment report is generated. This report integrates trend predictions and distribution information, providing a basis for construction quality control.

[0022] Example 1 During the normalization process of multi-source monitoring data, the raw multi-source monitoring data is aggregated into a central database. This data originates from sensors and measuring devices installed at different locations on the construction site. For example, tilt sensors are used to measure borehole verticality deviation, fiber optic grating sensors are used to collect pile temperature distribution, fiber optic strain sensors are used to obtain the axial strain gradient of the pile, fiber optic distributed monitoring cables are used to record the distribution of soil pressure around the pile, fiber optic sensor networks monitor the pile strain distribution, and inclinometers provide information on horizontal displacement. The acquisition frequency of the raw multi-source monitoring data is dynamically adjusted according to the construction stage. All acquired multi-source monitoring data is accompanied by precise timestamps and three-dimensional spatial coordinates. This metadata is crucial for subsequent data alignment. The original multi-source monitoring data typically contains parameters with different dimensions and numerical ranges. For example, borehole verticality deviation may be expressed in angle units, with a numerical range of 0-1°; pile temperature distribution measured by fiber optic gratings may be expressed in degrees Celsius, with a numerical range of 15-35℃; pile axial strain gradient measured by fiber optic strain sensors may be expressed in microstrain units, with a numerical range of -200-200με; pile soil pressure distribution recorded by fiber optic distributed monitoring cables may be expressed in kilopascals, with a numerical range of 50-300kPa; pile strain distribution may be expressed in microstrain units, with a numerical range of -150-150με; and horizontal displacement may be expressed in millimeters, with a numerical range of 0-20mm.

[0023] The first step in normalization is to extract the global maximum and minimum values ​​from the complete sequence of each type of raw monitoring data. For example, when processing borehole verticality deviation data, all borehole verticality deviation values ​​recorded at all measuring points from the start of construction to the current time point are traversed to find the maximum and minimum values. This process requires ensuring the integrity of the data sequence; any missing values ​​need to be filled in using appropriate interpolation methods before extreme value calculation to avoid bias in extreme value calculation. The normalized value for each type of raw monitoring data is calculated using the min-max scaling method. This method linearly transforms the raw monitoring data to a closed interval between zero and one. Specifically, each raw monitoring data value is subtracted from the minimum value of its corresponding type sequence, and the difference is then divided by the difference between the maximum and minimum values ​​of that type sequence. This transformation maps all raw monitoring data to a uniform numerical range.

[0024] To differentiate the relative importance of different monitoring data to pile foundation quality, each normalized data value is multiplied by a preset weighting coefficient. This weighting coefficient is pre-set based on the degree of influence of the monitoring data on the final quality and safety of the pile. Specifically, the weighting coefficient for the axial strain gradient of the pile is set to 0.3, the weighting coefficient for the distribution of soil pressure around the pile is set to 0.2, the weighting coefficient for the temperature distribution of the pile is set to 0.2, the weighting coefficient for the borehole verticality deviation is set to 0.1, the weighting coefficient for the strain distribution of the pile is set to 0.1, and the weighting coefficient for the horizontal displacement is set to 0.1. The set of weighting coefficients forms a weight vector, which is matched one-to-one with the normalized monitoring data to generate weighted monitoring data.

[0025] The weighting coefficients are not fixed; they can be dynamically adjusted. The importance of certain monitoring data may change at different construction stages: during the drilling stage, the weight of borehole verticality deviation is increased to 0.25, while the weighting coefficients of other monitoring data are proportionally decreased; during the concrete pouring stage, the weight of pile axial strain gradient is increased to 0.35, while the weighting coefficients of other monitoring data are proportionally decreased; during the concrete curing stage, the weight of pile temperature distribution is increased to 0.25, while the weighting coefficients of other monitoring data are proportionally decreased; and during the post-pile foundation stabilization stage, the weight of pile perimeter earth pressure distribution is increased to 0.3, while the weighting coefficients of other monitoring data are proportionally decreased. This dynamic adjustment ensures that the impact of core monitoring data at each stage is prominently reflected. It can be automatically optimized based on an expert knowledge base or by analyzing historical data through machine learning algorithms. The weighting coefficients themselves can also be iteratively updated as more data accumulates.

[0026] Perform data alignment operations on the weighted monitoring data, including timestamp alignment and spatial location alignment: Timestamp alignment requires that all data points of weighted monitoring data have a consistent reference on the time axis. Since the acquisition frequency and start time of different sensors may be different, the weighted monitoring data needs to be resampled according to the time sequence. Linear interpolation is usually used to unify the weighted monitoring data of different frequencies to a common time point to form time series data with equal intervals. Spatial alignment is more complex, requiring data from different measuring points and coordinate systems to be unified into the three-dimensional model of the pile foundation. Each data point is associated with a specific location in the three-dimensional model of the pile foundation, such as pile depth and radial distance from the pile center, through its spatial coordinates, such as east coordinates, north coordinates, and elevation. For continuous spatial data such as pile strain distribution, it may be necessary to use spatial interpolation techniques to generate a continuous distribution field and then sample it onto a unified grid node.

[0027] The accuracy of data alignment directly affects the quality of subsequent analysis. Therefore, high-precision interpolation algorithms and coordinate transformation models are required. For example, for GPS measurement data, it may be necessary to apply accurate geodetic transformation parameters to convert WGS-84 coordinates into the local coordinate system used in project construction.

[0028] After the above data alignment process, the weighted monitoring data is formed into a standardized monitoring dataset. This standardized dataset is structurally a five-dimensional tensor [time, spatial X, spatial Y, spatial Z, weighted monitoring data]. The time dimension corresponds to an equally spaced time series, for example, at 5-minute intervals. The spatial X, Y, and Z dimensions correspond to the grid of the pile foundation's 3D model, with X ranging from -5 to 5 meters, Y ranging from -5 to 5 meters, and Z ranging from 0 to 30 meters. The weighted monitoring data is then mapped to the pile foundation's 3D model grid. Based on the spatial coordinates of each monitoring data point, the weighted monitoring data of the six categories are precisely associated with the corresponding grid nodes of the pile foundation's 3D model. Specifically, the values ​​of borehole verticality deviation and horizontal displacement are matched to the depth level corresponding to the Z-axis of the grid and the horizontal coordinate nodes corresponding to the X and Y axes, based on the pile depth and radial position at the time of measurement. The values ​​of pile temperature distribution, pile axial strain gradient, and pile strain distribution are mapped to the corresponding three-dimensional coordinate nodes of the pile in the grid according to the spatial path of the fiber optic cable. The values ​​of pile perimeter soil pressure distribution are mapped to the nodes corresponding to the layer interval of the Z-axis and the horizontal coverage of the X and Y axes of the grid according to the layer position and horizontal monitoring range of the perimeter soil. This achieves a comprehensive correspondence between the monitoring data and the pile foundation three-dimensional model grid, providing structured spatial data support for feature extraction in the subsequent spatial distribution analysis module.

[0029] Each element of the standardized monitoring dataset is a dimensionless weighted value, representing the standardized observation of specific monitoring data at a specific time and location. The entire standardized monitoring dataset is stored in a self-describing format, such as NetCDF or HDF5, containing complete metadata information, such as monitoring data name, unit, weight coefficients, alignment method, etc., for traceability and verification. Data quality checkpoints are set at each stage of the normalization process, such as checking the rationality of extreme value calculations, whether the sum of weight coefficients is 1, and whether there are a large number of missing values ​​after alignment. Any data anomalies will trigger an alarm, prompting engineering technicians to intervene for manual inspection or correction.

[0030] Example 2 See Figure 2 The process of constructing a pile foundation quality assessment model begins with the partitioning of a standardized monitoring dataset. This dataset contains weighted monitoring data aligned by time series and spatial location. The standardized monitoring dataset is randomly divided into a training set and a validation set. The training set is used for parameter learning of the pile foundation quality assessment model, while the validation set is used to evaluate the model's generalization ability. To avoid over-adaptation of the pile foundation quality assessment model to the training data, the partitioning ensures that the training set covers data from all stages of construction, while the validation set selects continuous construction period data to simulate real-world application scenarios. The training set contains monitoring records for most of the construction process, ensuring that the pile foundation quality assessment model can learn sufficiently rich patterns. The validation set typically selects continuous time periods or spatial areas to simulate the model's performance on unseen data. The partitioning ratio needs to consider the total size of the dataset. A 7:3 ratio can be used when the data volume is sufficient; if the data volume is limited, the ratio may need to be adjusted or cross-validation techniques may need to be employed.

[0031] A Long Short-Term Memory (LSTM) network was used to model the time-series data in the training set to generate a time-series analysis module. The time-series data specifically refers to the time-series data related to pile temperature distribution, pile axial strain gradient, and pile strain distribution in the standardized monitoring dataset. The LTM network structure was chosen due to its advantages in handling long-term dependencies in time-series data. The input layer of the LTM network receives pile temperature distribution, pile axial strain gradient, and pile strain distribution data from the training set within a time window. The window length is set to 2 hours, and the time step interval is 10 minutes. This data is input into the input layer sequentially in multi-dimensional vector form, with the vector dimension consistent with the number of monitoring points for the three types of monitoring data. The LTM network internally contains memory cells and three gating mechanisms: an input gate, a forget gate, and an output gate. The forget gate determines which information is discarded from the previous state, the input gate controls the inflow of new information, and the output gate regulates the output of the current state. Through this mechanism, the LTM network can selectively remember or forget long-term historical information. The modeling process involves iterative optimization of the weights of the Long Short-Term Memory (LSTM) network: gradients are calculated using backpropagation via a time-based algorithm, and the parameters of the LTM network are updated using an adaptive moment estimation algorithm. The initial learning rate is set to 0.001, decreasing by 10% every 100 iterations. The loss function is the mean squared error, aiming to minimize the difference between the predicted pile body temperature change, axial strain gradient change, strain change, and the true values. The training process continues until the loss function value fluctuates less than 1e-5 for 50 consecutive iterations or the number of iterations reaches 2000, thus completing the training of the LTM network.

[0032] A convolutional neural network is used to model the spatial distribution data in the training set to generate a spatial distribution analysis module. The spatial distribution data specifically refers to the spatial data related to borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement in the standardized monitoring dataset. In the preprocessing stage, Kriging interpolation is used to resample these data onto a regular three-dimensional grid. The size of the regular three-dimensional grid is set according to the actual engineering requirements, corresponding to the radial, circumferential, and depth directions of the pile body to ensure consistent spatial resolution of the data. Among them, the pile perimeter earth pressure distribution is aligned with the grid depth dimension according to the pile perimeter earth layer location. The convolutional neural network employs a structure of "3 convolutional layers + 2 pooling layers + 1 fully connected layer," capturing spatial data correlations through local connectivity and weight sharing. The convolutional layers use 3×3×3 three-dimensional convolutional kernels to slide across a spatial grid, extracting features at three different scales: axial gradient changes in borehole verticality deviation, radial differences in pile-surround earth pressure distribution, and circumferential concentrated features of horizontal displacement. The pooling layer uses max pooling to reduce the dimensionality of the feature map, increasing the translation invariance and computational efficiency of the pile foundation quality assessment model. The fully connected layer maps the reduced-dimensional feature map to anomaly probability values; when the probability value exceeds a set threshold, it is identified as an anomaly region, and the 3D coordinates of the anomaly region are output. The modeling process also uses labeled data, optimizing network parameters by minimizing reconstruction or classification errors.

[0033] The modeling results based on Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) were cross-validated using a validation set to optimize model parameters. For LSM-based modeling, a temporal cross-validation method was employed, using time-series data from the validation set corresponding to pile temperature distribution, pile axial strain gradient, and pile strain distribution as input. The prediction errors for these three types of variables were calculated, and the number of hidden layer neurons and gating parameters of the LSM network were adjusted based on the error feedback. For CNN-based modeling, spatial data from the validation set corresponding to borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement were input to evaluate the accuracy of anomaly area identification, thereby optimizing parameters such as convolution kernel size and pooling step size of the CNN. Cross-validation assesses the performance stability of the two modeling results on unseen data, avoids overfitting, and ensures that model parameters reach their optimal state.

[0034] Using the optimized modeling results based on a Long Short-Term Memory (LSTM) network, trend predictions are made for pile temperature distribution, pile axial strain gradient, and pile strain distribution. The specific steps are as follows: First, standardized monitoring data corresponding to the pile temperature distribution, pile axial strain gradient, and pile strain distribution are organized into continuous time-series data in timestamp order, ensuring no missing data and temporal continuity. Then, this time-series data is input into the optimized LSM network. Through its internal memory cells and gating mechanism, the input time-series data is processed step-by-step to capture the fluctuation patterns, amplitudes, and long-term dependencies of the data over time, thereby extracting time-series features that characterize the temporal evolution of the three types of data. Based on the extracted time-series features and the mapping relationships learned during training, the changes in pile temperature, axial strain gradient, and strain within a set time window are predicted. The time window is determined according to the quality monitoring cycle of pile foundation construction to ensure that the prediction results can provide an effective reference for construction quality control.

[0035] The predicted changes in pile body temperature, axial strain gradient, and strain are compared with preset warning thresholds to generate pile foundation settlement trend prediction results and pile foundation strain trend prediction results. The warning thresholds are determined based on relevant pile foundation construction specifications and quality control experience from similar projects, with reasonable ranges set for each of the three types of changes. By comparing each predicted change with its corresponding warning threshold, if a change falls within the threshold range, the pile foundation state corresponding to that indicator is considered stable, and it is marked as a stable trend in the pile foundation settlement trend prediction results or pile foundation strain trend prediction results. If a change exceeds the threshold range, considering the inherent correlation between that indicator and pile foundation settlement and strain, it is marked as an abnormal trend in the corresponding prediction results, and the type of indicator corresponding to the abnormality is identified. Finally, all comparison results are integrated to form complete pile foundation settlement trend prediction results and pile foundation strain trend prediction results.

[0036] Ultimately, the trained time series analysis module based on long short-term memory network and the spatial distribution analysis module based on convolutional neural network together constitute the core analysis engine of the pile foundation quality assessment model. The two can run independently to process information in time and space dimensions respectively, or they can interact at a higher level of fusion layer to provide multi-dimensional decision-making basis for generating a comprehensive assessment report.

[0037] Example 3 See Figure 3When performing spatial correlation analysis on borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement, the spatial distribution analysis module needs to map the discrete measurement point data of the three types of monitoring data into a unified pile foundation 3D model. This 3D model is constructed based on design drawings and actual measurement point coordinates, using boundary representation to express the pile geometry. Simultaneously, a layered soil structure around the pile is embedded in the 3D model to match the layered data of pile perimeter earth pressure distribution from fiber optic distributed monitoring. Specifically, the steps for constructing the pile foundation 3D model are as follows: Based on the geometric parameters of the pile body in the design drawings, an initial cylindrical pile foundation 3D model is created in the computer system. Combined with the actual coordinates of the measuring points monitored by multi-source monitoring data during construction, the measuring point data is aligned to the coordinate system of the pile foundation 3D model through a seven-parameter coordinate transformation to ensure that the point position error is ≤0.01m. The seven-parameter coordinate transformation achieves complete parameter correction between different 3D coordinate systems, including translation, rotation and scaling, through three translation parameters, three rotation parameters and one scaling parameter, which can meet the requirement of high-precision alignment between the actual measuring point coordinates and the coordinate system of the pile foundation 3D model.

[0038] The boundary representation method is adopted to represent the pile as the boundary surface of the cylindrical side and the upper and lower circular end faces. The surface is defined by 1000 edges, 500 vertices and 2 face elements. At the same time, the soil layering structure around the pile is added within 1m around the pile in the 3D model of the pile foundation to accurately express the spatial relationship between the pile and the soil around the pile, laying the foundation for the subsequent mapping of three types of monitoring data and the generation of spatial distribution matrix. The mapping process involves spatial coordinate transformation and data interpolation, converting the local construction coordinates (X, Y, Z) of the three types of monitoring data measurement points to the coordinate system of the pile foundation 3D model. A regular 3D mesh with dimensions of 21×21×61 is constructed. For non-mesh point data, the Kriging interpolation algorithm is used to calculate the mesh node values: standardized borehole verticality deviation is interpolated along the Z-axis depth direction; standardized pile perimeter earth pressure distribution is interpolated according to the corresponding Z-axis layer and the X and Y-axis horizontal directions of the pile perimeter earth layer; and standardized horizontal displacement is interpolated in the full 3D space. Finally, three structured spatial distribution matrices are generated, including: borehole verticality deviation matrix, pile perimeter earth pressure distribution matrix, and horizontal displacement matrix, all with dimensions of 21×21×61. The 21×21×61 dimension of the regular 3D mesh is based on the uniform division of the X, Y, and Z axis ranges to ensure a balance between spatial resolution and computational efficiency. The 21 mesh points in the X and Y directions correspond to a 0.5-meter spacing, and the 61 mesh points in the Z direction also correspond to a 0.5-meter spacing.

[0039] A three-dimensional convolutional neural network was used to extract features from three structured spatial distribution matrices. Sliding calculations were performed in a three-dimensional grid space to capture spatial correlation patterns in three types of monitoring data: the gradient change of borehole verticality deviation along the Z-axis, the difference in soil pressure distribution around the pile in the horizontal direction, and the spatial coupling relationship between horizontal displacement and borehole verticality deviation. For example, convolutional operations identified an anomalous pattern where the borehole verticality deviation at Z=10-15m increased from 0.2° to 0.6°, corresponding to a horizontal displacement of 8mm in the negative X-axis direction. The network adopts a structure of "3 convolutional layers + 2 pooling layers + 1 fully connected layer". The three convolutional layers use 32, 64, and 128 convolutional kernels, respectively. The pooling layer uses 2×2×2 max pooling with a stride of 2 to reduce the dimensionality of the feature maps. The fully connected layer maps the dimensionality-reduced features to 1024-dimensional vectors and outputs the anomaly probability value of the pile foundation grid cells through a sigmoid activation function. This anomaly probability value is a comprehensive anomaly judgment result of three types of features: borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement. In the feature extraction stage of the convolutional neural network, feature maps in the spatial distribution matrix are extracted for the three types of features: borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement through convolutional and pooling layer operations. After the feature map of each feature dimension is compressed by the pooling layer, key feature values ​​are retained. These values ​​are quantified by an independent processing unit before the fully connected layer to generate their respective anomaly quantification indices, namely, the borehole verticality deviation exceedance coefficient, the pile perimeter earth pressure anomaly coefficient, and the horizontal displacement concentration coefficient. If the borehole verticality deviation exceeds the standard coefficient by ≥0.7, it is determined that the dimension exceeds the standard. If the pile perimeter earth pressure anomaly coefficient exceeds ≥0.75, it is determined that the dimension exceeds the standard. If the horizontal displacement concentration coefficient exceeds ≥0.8, it is determined that the dimension exceeds the standard. When the anomaly probability of a certain pile foundation grid unit exceeds ≥0.85, it is determined to be a spatial anomaly area. Combining the individual anomaly quantification indicators of the three types of features, the specific exceeding dimension is identified. If only a single coefficient meets the standard, the corresponding single parameter exceeding the standard type is output. If two or three coefficients meet the standard, the "multi-parameter coupling exceeding the standard" type is output. Finally, the anomaly area is output as the three-dimensional coordinates (X, Y, Z) of the model coordinate system and the corresponding parameter exceeding the standard type.

[0040] After identifying spatial anomaly areas and determining the types of parameter exceedances, this output information is directly applied to the generation of the comprehensive pile foundation quality assessment report. The three-dimensional coordinates and exceedance types of the anomaly areas serve as key inputs for precise marking on verticality deviation distribution maps, pile perimeter earth pressure distribution maps, and horizontal displacement distribution maps, highlighting spatial weaknesses in the pile foundation. Simultaneously, in the phased quality assessment reports, anomaly area information is correlated with monitoring data from the corresponding construction phase, such as borehole verticality deviation and pile body temperature distribution, to help pinpoint the root causes of quality deviations. For example, when the report shows an anomaly of "multi-parameter coupling exceedance" at a certain depth, engineers can check the compliance of drilling or grouting processes by referring to the construction records for that coordinate area. This allows for targeted adjustments to equipment parameters or process sequences when generating construction optimization suggestions, improving the accuracy of subsequent construction quality control.

[0041] Based on the abnormal regions identified by the convolutional neural network, a verticality deviation distribution map, a pile-surround earth pressure distribution map, and a horizontal displacement distribution map are generated. The generation process is as follows: the abnormal probability value is fused with the value of the spatial distribution matrix to obtain a spatial distribution matrix with abnormal annotations, which represents the verticality deviation distribution map, the pile-surround earth pressure distribution map, and the horizontal displacement distribution map.

[0042] The verticality deviation distribution map, the soil pressure distribution map around the pile, and the horizontal displacement distribution map of the pile foundation are correlated and calculated in each pile foundation grid cell of the 3D pile foundation model to obtain the spatial joint index value within the same pile foundation grid cell. The formula is as follows:

[0043] in, Represents the pile foundation mesh element in the 3D model of the pile foundation. Spatial joint index value at the location, The distribution diagram of pile verticality deviation in the pile foundation grid cell The normalized value of the borehole verticality deviation. express The weighting coefficients, This represents the distribution of earth pressure around the pile in the pile foundation grid unit. The normalized value of the earth pressure around the pile. express The weighting coefficients, The distribution diagram of horizontal displacement of pile foundation in pile foundation grid cells The normalized value of the horizontal displacement. express The weighting coefficients.

[0044] The anomaly probability value of the pile foundation grid unit is a comprehensive judgment result output by the spatial distribution analysis module after feature extraction, used to identify whether the pile foundation grid unit is an abnormal area. The spatial joint index value of the pile foundation grid unit is calculated by weighting the pile verticality deviation distribution map, pile surrounding earth pressure distribution map, and pile horizontal displacement distribution map of the same pile foundation grid unit, which can further quantify the risk level of abnormal areas. By calculating the spatial joint index value of all pile foundation grid units, a brand-new spatial joint index distribution can be generated. Based on this distribution, high-risk areas in space are identified by setting thresholds, thus forming a spatial dimension assessment result.

[0045] In one technical solution of the present invention, visualization rendering technology can be used. The verticality deviation distribution map uses red, yellow and green gradients to represent numerical differences, the pile perimeter earth pressure distribution map uses dark red, light red, yellow and green gradients, and the horizontal displacement distribution map uses red, yellow and green gradients. All distribution maps are marked with the boundaries of abnormal areas and their corresponding coordinates, and the numerical range and color correspondence are clearly stated in the legend. The three types of distribution maps can be presented in multiple formats. The verticality deviation distribution map includes a longitudinal profile of the pile (ZX axis, ZY axis) and a 3D contour map. The profile uses colored curves to mark the depth deviation values, and the 3D contour map uses transparent color blocks to overlay the pile model, highlighting areas exceeding the deviation limit of ≥0.5°. The pile perimeter soil pressure distribution map includes a layered plan view of the pile perimeter soil and a 3D isosurface map. The plan view uses colored grids to mark the horizontal pressure distribution, and the isosurface map uses color blocks of different transparency to represent the pressure isosurfaces. The horizontal displacement distribution map uses a vector arrow diagram overlaid with a 3D pile model, with separate top views output for the pile top and the middle part of the pile, clearly showing the spatial distribution differences of horizontal displacement. All distribution maps are accompanied by a 1:200 scale, color-value correspondence legend, coordinate system explanation, and text labels for abnormal areas.

[0046] The predicted results of pile foundation settlement trend and pile foundation strain trend are weighted and fused to form a time-dimensional assessment result. Specifically, firstly, the predicted results of pile foundation settlement trend and pile foundation strain trend are precisely aligned with timestamps to ensure that the predicted data at each time node in the two types of results correspond one-to-one, eliminating fusion deviations caused by inconsistent time bases. Secondly, based on the inherent correlation between the two types of results and pile foundation stability, the changing characteristics of settlement trend and strain trend at each time node are analyzed collaboratively. Through dynamic weighting, the potential bias of a single result is eliminated, ultimately forming a time-dimensional assessment result that can comprehensively characterize the state change law of the pile foundation in the time dimension, thereby quantifying the stability of the pile foundation in the time dimension. At the same time, the individual prediction curves of each parameter are retained to ensure the traceability of the fusion result.

[0047] A comprehensive assessment report on pile foundation quality is generated based on the spatial and temporal assessment results. The specific process is as follows: When analyzing the cross-influence of the temporal and spatial dimensions, abnormal areas in the spatial assessment results are first located. Then, the time series data corresponding to these abnormal areas are retrieved, and the pile foundation settlement trend prediction results and pile foundation strain trend prediction results are correlated for collaborative analysis. The focus is on verifying whether the settlement changes in spatially abnormal areas, such as areas with excessive verticality deviation or concentrated horizontal displacement, are more significant than in other areas, or whether the strain trend shows synchronous abrupt changes at spatially abnormal locations. At the same time, it is verified whether the stages of abnormal temporal trends, such as the stage of sudden increase in settlement rate, correspond to concentrated anomalies in parameters such as pile perimeter soil pressure distribution and verticality deviation in specific spatial areas. Through this spatiotemporal correlation, the key coupling factors affecting pile foundation quality are identified. Therefore, the spatial and temporal assessment results are mapped onto a unified scoring scale to calculate a comprehensive pile foundation quality score. This comprehensive score directly corresponds to the overall quality status of the pile foundation. The comprehensive pile foundation quality score, along with detailed temporal and spatial analysis, constitutes the core content of the comprehensive pile foundation quality assessment report. The report is output in a structured document format, including data summaries, analysis results, visualizations, and conclusions for engineering decision-making. When calculating the comprehensive pile foundation quality score, the core basis for the spatial dimension assessment results is the risk distribution generated by the spatial joint index values. The spatial joint index, through weighted fusion of the pile foundation verticality deviation distribution map, the pile perimeter earth pressure distribution map, and the pile foundation horizontal displacement distribution map, forms a quantified spatial risk map. This spatial risk map directly serves as the basis for the spatial dimension assessment results. During the assessment, parameter values ​​of abnormal areas are extracted from this spatial risk map to calculate the spatial dimension score, ensuring that the score objectively reflects the spatial quality differences and risk concentration of the pile foundation.

[0048] Example 4 Calculating the comprehensive score for pile foundation quality is the core step in generating the assessment report. This process first maps the time-dimensional and spatial-dimensional assessment results to a unified scoring scale based on the outputs of the time-series analysis and spatial distribution analysis modules. The time-dimensional assessment results are derived from the trend predictions of pile body temperature distribution, axial strain gradient, and pile body strain distribution. They integrate the changing patterns and risk warning information of key parameters over multiple future monitoring periods. The mapping must be performed according to preset time-dimensional scoring rules, using the time-dimensional assessment results as the core quantitative indicator. If the time-dimensional assessment result does not exceed the first time-dimensional warning threshold, 100 points are awarded; if the time-dimensional assessment result exceeds the first time-dimensional warning threshold but does not exceed the second time-dimensional warning threshold, points are deducted based on the magnitude of the exceedance of the first time-dimensional warning threshold, with the scoring range being [insert range here]. If the time dimension assessment result exceeds the warning threshold of the second time dimension, with 80 points as the baseline, points will be deducted according to the magnitude of the excess. The score range is as follows: The first time dimension warning threshold is lower than the second time dimension warning threshold. A standardized score of 0-100 is calculated based on the above deduction rules, and this score is used as the single time dimension score to complete the quantitative transformation of the comprehensive characteristics of the time dimension assessment results. The spatial dimension assessment results are based on the spatial correlation analysis of borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement, including core information such as abnormal area distribution and spatial joint index values. When mapping, the spatial dimension scoring standard is referenced. If the anomaly probability of the pile foundation grid unit does not exceed 0.85, and the spatial dimension assessment results do not exceed the preset benchmark value, 100 points are awarded. If the number of pile foundation grid units with an anomaly probability exceeding 0.85 accounts for no more than 10% of the total number of pile foundation grid units in the 3D pile foundation model, the maximum anomaly probability of the pile foundation grid unit does not exceed 0.9, and the spatial dimension assessment results do not exceed 120% of the preset benchmark value, it is judged as having a small anomaly area, low severity, and slightly exceeding the spatial dimension assessment standard, and a score is deducted from the baseline of 100 points. The score range is... If the number of pile foundation mesh elements with an anomaly probability exceeding 0.85 accounts for more than 10% of the total number of pile foundation mesh elements in the 3D pile foundation model, the maximum anomaly probability of the pile foundation mesh elements exceeds 0.9, or the spatial dimension assessment result of some pile foundation mesh elements exceeds 120% of the preset benchmark value, it is judged as having a large anomaly area, high severity, or significantly exceeding the standard for spatial dimension assessment results. A deduction of 80 points will be made, with the score range being... After completing the unified score scale mapping, a weighted sum of the time and spatial dimension scores is generated using weight coefficients dynamically adjusted based on the construction stage, resulting in a comprehensive pile foundation quality score. The weight coefficients are strictly set according to the core quality control objectives of each construction stage. During the drilling and concrete pouring stages, the focus is on controlling spatial geometric accuracy and stress distribution, with a spatial dimension score weight of 0.6 and a time dimension score weight of 0.4. During concrete curing and post-pile foundation stabilization stages, the focus is on the long-term trend and stability of parameters, with a time dimension score weight of 0.6 and a spatial dimension score weight of 0.4. This weighting method achieves a deep integration of the time and spatial dimension evaluation results, ultimately yielding a comprehensive score that fully reflects the pile foundation quality status. The calculation formula is as follows:

[0049] in: It is a comprehensive score for the quality of pile foundations. It is a time-based scoring weight. It is a time-dimension score mapped to a unified scale. It is the spatial dimension scoring weight. It is a spatial dimension score mapped to a unified scale.

[0050] Linking the pile foundation quality grade with monitoring data from each construction phase is crucial for generating phased reports. The construction process can be divided into multiple phases, each with its corresponding set of monitoring data. The linking operation is achieved through timestamp matching, establishing a correspondence between all monitoring data collected within a specific construction phase and the pile foundation quality grade calculated at the end of that phase, analyzing the correlation between construction parameters and the final quality grade. Generating a phased quality assessment report requires summarizing key data from each phase, the calculated comprehensive pile foundation quality score for that phase, the quality grade, and important spatial distribution and time trend diagrams. The report is presented in a clear format, facilitating engineers' ability to trace the origin of quality issues. See Table 1, which shows a phased assessment summary for a project.

[0051] Table 1: Summary of Phased Quality Assessment of Pile Foundation Construction

[0052] Table 1 does not contain actual project data; instead, it uses example values ​​to illustrate the report's structure and content. In practical applications, the table would include specific project names, pile numbers, precise monitoring data, and calculation results. Table 1 shows fluctuations in quality grades across different construction stages. For example, the S02 rebar cage installation stage achieved an "excellent" grade, while the S03 concrete pouring stage saw a decrease in score. This correlation suggests a need to carefully review the construction records and monitoring details of the concrete pouring stage. The core value of the phased quality assessment report lies in bridging the gap between dynamic data from the construction process and the final static quality assessment. This allows quality management to shift from post-construction inspection to process control. By analyzing changes in scores at each stage, project managers can accurately pinpoint construction steps requiring improvement, thereby optimizing subsequent construction plans or implementing targeted reinforcement measures for completed sections. The report can be output as an electronic document or integrated into a visual dashboard within the project management system, supporting interactive queries and drill-down analysis. For example, clicking on a stage row can display detailed analysis charts and raw data for that stage. The classification of pile foundation quality grades is based on a comprehensive pile foundation quality score, establishing unified and quantifiable grading rules to ensure consistency and objectivity in the rating standards. The specific classification criteria are as follows: When the comprehensive quality score of the pile foundation is ≥90 points, it is judged as "excellent", indicating that the evaluation results in both the time and space dimensions are at the optimal level, the geometric accuracy, stress state and long-term stability of the pile body fully meet the design requirements, and there are no potential quality risks; when the comprehensive score is between 75 and 89 points, it is judged as "qualified", indicating that the core quality indicators of the pile foundation meet the design and specification requirements, with only a small number of non-critical parameter fluctuations or minor local anomalies, and such situations do not have a substantial impact on the overall load-bearing safety and service life of the pile foundation, and it can be put into use without special rectification; when the comprehensive score is <75 points, it is judged as "unqualified", meaning that there are key problems affecting the structural stability of the pile foundation, which may manifest as core parameters exceeding the warning threshold or the spatial anomaly area exceeding the standard, and it can no longer meet the load-bearing safety requirements within the design service life. It is necessary to identify the root cause of the problem through special testing, and take targeted measures such as reinforcement and rework to deal with it before re-evaluating the quality.

[0053] Example 5 When generating a phased quality assessment report, monitoring data streams for a specified construction phase are extracted from the database. These data are indexed by timestamps and measurement point numbers, forming a subset of data for that phase. The extracted key parameters strictly correspond to multi-source monitoring data during pile foundation construction, specifically including: borehole verticality deviation, pile temperature distribution, pile axial strain gradient, pile perimeter earth pressure distribution, pile strain distribution, and horizontal displacement. Multi-source monitoring data are core monitoring indicators that accurately reflect the core state of pile foundation construction quality, covering pile geometric accuracy, structural stress and deformation, environmental and stress response, and displacement stability, directly related to pile foundation bearing safety and construction compliance. All monitoring data are accompanied by precise timestamps and spatial coordinates, which relate to the thickness of the reinforcement protective layer and the uniformity of pile stress. Multi-source monitoring data exists in both time series and spatial distribution forms, and is compared and analyzed with the pile foundation quality grade calculated at the end of that phase. The analysis aims to establish a quantitative correlation between construction operations and pile quality. The comparison operation is not a simple numerical comparison, but uses statistical correlation and trend consistency analysis. Through this comparison, quality deviations in the construction process are identified. Quality deviations not only refer to parameters exceeding the allowable values ​​in the specifications, but also include situations where parameters are within the calibration range, but their change patterns or spatial distribution characteristics indicate potential risks. For example, a sudden increase of 5°C in the pile body temperature distribution within 1 hour, even if it does not exceed the upper limit of the curing temperature of 35°C, may reflect abnormal accumulation of hydration heat in the concrete, indicating the risk of internal cracks; a sudden drop of 40% in the soil pressure distribution around the pile at a certain depth, even if it does not exceed the pressure limit, may indicate that the foundation soil at that depth has been disturbed, and the drilling process needs to be checked carefully to see if it caused a hole collapse; a sudden change of 3με / m in the axial strain gradient of the pile body during the curing stage may reflect uneven stress on the reinforcing cage, and further verification with horizontal displacement data is required.

[0054] The core of this step, generating construction optimization suggestions based on identified quality deviations, is analyzing the correlation between these deviations and various construction parameters, including equipment, material, and process parameters. Equipment parameters include drilling rig model, drill bit speed, and drilling pressure; material parameters involve concrete mix proportions, slump, and rebar specifications; and process parameters refer to grouting speed, guide pipe embedment depth, and number of hole cleaning cycles. The analysis employs data mining techniques, such as decision trees or association rule learning, to identify frequent combinations of construction parameters that lead to specific quality deviations from historical data, thereby determining the main factors affecting pile foundation quality. For example, the analysis might show that when a certain type of drilling rig is used and the drilling speed is too fast in a specific soil layer, the probability of increased borehole verticality deviation increases significantly. If the soil pressure around the pile drops sharply by 40% at a certain depth, association rule learning reveals that this deviation is often related to the combination of construction parameters: 'drilling pressure < 0.3 MPa and number of cleaning cycles < 2'. Insufficient drilling pressure and incomplete cleaning are identified as the main influencing factors. If the axial strain gradient of the pile body shows a sudden change of 3 με / m, decision tree analysis indicates that it is mostly caused by 'steel cage hoisting tilt angle > 2° + positioning pad spacing > 1.5m'. The steel cage installation parameters are identified as the core adjustment target. If the horizontal displacement exceeds 10 mm, data mining results show that it is related to 'horizontal deviation during drilling > 0.5° and soil compaction degree around the pile < 90%'. The positioning parameters of the drilling equipment and the foundation pretreatment parameters need to be adjusted simultaneously. When using decision trees to mine correlations, a training dataset for the decision tree was constructed using monitoring data from the construction phase and corresponding pile quality status data. The target classification label was set as "whether the axial strain gradient of the pile exhibits a sudden change of 3με / m," and construction parameters such as the hoisting tilt angle of the rebar cage and the spacing of the positioning pads were set as feature attributes. Information gain ratio was used as the feature selection criterion, and features that significantly influenced the target label were sequentially screened. First, the rebar cage hoisting tilt angle was used as the root node for branching, and then the spacing of the positioning pads was used as the child node for further subdivision, constructing a multi-branch decision tree model. The model was pruned and optimized using the validation set to remove redundant branches and improve generalization ability. By calculating the model's accuracy in identifying "axial strain gradient abrupt changes" samples and the proportion of feature importance, it was determined that "rebar cage hoisting tilt angle > 2°" and "spacing of positioning pads > 1.5m" were the key splitting features on the decision path. Furthermore, the probability of triggering an axial strain gradient abrupt change when these two features acted together was significantly higher than that of a single feature. Finally, this combination of construction parameters was determined to be the core correlation factor leading to the sudden change in the axial strain gradient of the pile.

[0055] Based on the identified key factors, construction parameters are adjusted to generate an optimized construction plan. This adjustment is targeted rather than general. If the main factor is a problem in the concrete pouring process, the optimized plan might specify: using a higher pouring rate in the lower and middle parts of the pile to reduce the impact of mud cake on the borehole wall, reducing the pouring rate near the pile top to ensure pile head compaction, and specifying that the tremie pipe embedment depth be controlled within the range of 2 to 6 meters, with the concrete surface height measured after each certain volume of concrete is poured. If the main factor is insufficient installation accuracy of the reinforcing cage, the plan might require increasing the number and strength of positioning pads, or adopting a new hoisting technique to ensure vertical lowering.

[0056] The optimized construction plan was compared with historical construction data to verify its expected effects. Historical data came from previous completed projects with similar pile foundations or geological conditions. The comparison was conducted under controlled assumptions. Predictive models were built to simulate potential changes in key parameters and quality scores after applying the new plan. For example, old construction parameters recorded in historical data were replaced with parameters set by the new plan, and the pile foundation quality assessment model was rerun to predict whether the overall score would improve and the quality level would be enhanced. The verification process also included an assessment of the plan's feasibility and economy, checking whether the equipment, labor, or material costs required by the new plan were within the project's acceptable range. The final construction optimization recommendations were a feasibility report containing specific parameter adjustments, expected effect analysis, and implementation points, providing clear operational guidance for subsequent construction.

[0057] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A pile foundation quality assessment method based on multi-source data fusion, characterized in that, Includes the following steps: Collect multi-source monitoring data during pile foundation construction, including: borehole verticality deviation, pile body temperature distribution, pile body axial strain gradient, pile perimeter earth pressure distribution, pile body strain distribution, and horizontal displacement. The multi-source monitoring data is processed to generate a standardized monitoring dataset; Construct a pile foundation quality assessment model that includes a time series analysis module and a spatial distribution analysis module; The standardized monitoring dataset is used to predict the trends of pile body temperature distribution, pile body axial strain gradient and pile body strain distribution using the time series analysis module, generating pile foundation settlement trend prediction results and pile foundation strain trend prediction results. The standardized monitoring dataset is used to perform spatial correlation analysis on borehole verticality deviation, pile perimeter earth pressure distribution, and horizontal displacement using the spatial distribution analysis module, generating pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map. Based on the predicted results of pile foundation settlement trend, pile foundation strain trend, pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map, a comprehensive pile foundation quality assessment report is generated.

2. The pile foundation quality assessment method based on multi-source data fusion according to claim 1, characterized in that, The process of processing the multi-source monitoring data to generate a standardized monitoring dataset includes: The maximum and minimum values ​​of each type of monitoring data are extracted from the multi-source monitoring data. The minimum-maximum scaling method is used to normalize each monitoring data to obtain the normalized value. Set the weight coefficient for each type of monitoring data, multiply the normalized value of each monitoring data by the corresponding weight coefficient, and generate weighted monitoring data. The generated weighted monitoring data are aligned according to timestamps and spatial locations to form a standardized monitoring dataset.

3. The pile foundation quality assessment method based on multi-source data fusion according to claim 1, characterized in that, The generation process of the pile foundation settlement trend prediction results and pile foundation strain trend prediction results includes: The standardized monitoring data corresponding to the pile body temperature distribution, pile body axial strain gradient and pile body strain distribution are centrally input into the time series analysis module to extract time series features. Based on the aforementioned time series characteristics, predict the changes in pile body temperature, axial strain gradient, and strain within future time windows; The changes in pile body temperature, axial strain gradient, and strain are compared with preset warning thresholds, and combined with the intrinsic relationship between these factors and pile settlement and strain, to generate pile settlement trend prediction results and pile strain trend prediction results.

4. The pile foundation quality assessment method based on multi-source data fusion according to claim 1, characterized in that, The methods for generating the pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map include: The standardized borehole verticality deviation, standardized pile perimeter earth pressure distribution, and standardized horizontal displacement from the standardized monitoring dataset are mapped onto the three-dimensional model of the pile foundation to generate three spatial distribution matrices, including: borehole verticality deviation matrix, pile perimeter earth pressure distribution matrix, and horizontal displacement matrix. The spatial distribution analysis module sequentially captures the spatial correlation of three spatial distribution matrices in the pile foundation mesh cells of the three-dimensional pile foundation model, and outputs the anomaly probability of the pile foundation mesh cells. The anomaly probabilities of the pile foundation grid cells are numerically fused with three spatial distribution matrices to generate pile foundation verticality deviation distribution map, pile perimeter earth pressure distribution map, and pile foundation horizontal displacement distribution map.

5. The pile foundation quality assessment method based on multi-source data fusion according to claim 4, characterized in that, The construction process of the three-dimensional model of the pile foundation is as follows: An initial three-dimensional model of the pile foundation is created based on the geometric parameters of the pile body on the design drawings, and the coordinates of the actual measuring points monitored by multi-source monitoring data during the pile foundation construction process are aligned to the coordinate system of the three-dimensional model of the pile foundation. The boundary representation method is used to describe the geometry of the pile in the three-dimensional model of the pile foundation, and the layered structure of the soil around the pile is embedded in the three-dimensional model of the pile foundation.

6. The pile foundation quality assessment method based on multi-source data fusion according to claim 4, characterized in that, The process for generating the comprehensive assessment report on pile foundation quality is as follows: The pile foundation settlement trend prediction results and pile foundation strain trend prediction results are weighted and integrated to form a time-dimensional evaluation result. The verticality deviation distribution map, the earth pressure distribution map around the pile, and the horizontal displacement distribution map of the pile foundation are correlated and calculated in each pile foundation grid cell of the three-dimensional model of the pile foundation to form a spatial dimension evaluation result. The time-dimensional and spatial-dimensional assessment results are mapped onto a unified scoring scale, and a weighted comprehensive score for pile foundation quality is calculated. The quality grade of the pile foundation during the construction process is determined based on the comprehensive score of the pile foundation quality.

7. The pile foundation quality assessment method based on multi-source data fusion according to claim 6, characterized in that, The process of forming the spatial dimension assessment results is as follows: in, Represents the pile foundation mesh element in the 3D model of the pile foundation. Spatial joint index value at the location, The distribution diagram of pile foundation verticality deviation in the grid cells The normalized value of the borehole verticality deviation. express The weighting coefficients, This represents the distribution of earth pressure around the pile in the pile foundation grid unit. The normalized value of the earth pressure around the pile. express The weighting coefficients, The distribution diagram of horizontal displacement of pile foundation in pile foundation grid cells The normalized value of the horizontal displacement. express The weighting coefficients.

8. The pile foundation quality assessment method based on multi-source data fusion according to claim 6, characterized in that, The specific process of mapping the time dimension assessment results to the score scale is as follows: If the time dimension assessment result does not exceed the first time dimension warning threshold, 100 points are awarded; If the time dimension assessment result exceeds the first time dimension warning threshold but does not exceed the second time dimension warning threshold, based on a baseline of 100 points, points will be deducted according to the magnitude of the exceedance of the first time dimension warning threshold. The score range is as follows: ; If the time dimension assessment result exceeds the second time dimension warning threshold, with 80 points as the baseline, points will be deducted according to the magnitude of the excess. The score range is as follows. ; Among them, the warning threshold for the first time dimension is lower than the warning threshold for the second time dimension.

9. The pile foundation quality assessment method based on multi-source data fusion according to claim 6, characterized in that, The specific process of mapping spatial dimension assessment results to a score scale is as follows: If the anomaly probability of the pile foundation grid unit does not exceed 0.85 and the spatial dimension evaluation results do not exceed the preset benchmark value, 100 points are awarded. If the number of pile foundation mesh elements with an anomaly probability exceeding 0.85 does not exceed 10% of the total number of pile foundation mesh elements in the 3D pile foundation model, the maximum anomaly probability of the pile foundation mesh elements does not exceed 0.9, and the spatial dimension evaluation results do not exceed 120% of the preset benchmark value, a deduction of 100 points will be made, with the score range being [insert range here]. ; If the number of pile foundation mesh elements with an anomaly probability exceeding 0.85 accounts for more than 10% of the total number of pile foundation mesh elements in the 3D pile foundation model, the maximum anomaly probability of the pile foundation mesh elements exceeds 0.9, or the spatial dimension assessment result of some pile foundation mesh elements exceeds 120% of the preset benchmark value, a deduction of 80 points will be made, with the score range being [insert range here]. .

10. The pile foundation quality assessment method based on multi-source data fusion according to claim 6, characterized in that, Also includes: Based on the borehole verticality deviation matrix, the pile peri-earth pressure distribution matrix, and the horizontal displacement matrix, respectively, the borehole verticality deviation exceedance coefficient, the pile peri-earth pressure anomaly coefficient, and the horizontal displacement concentration coefficient are generated; Set abnormal thresholds for borehole verticality deviation exceeding the standard coefficient, pile perimeter earth pressure abnormal coefficient, and horizontal displacement concentration coefficient respectively, and record the type and value of coefficients exceeding the abnormal thresholds in the comprehensive evaluation report of pile foundation quality.

Images