Pile body integrity intelligent detection method and system based on temperature field simulation data driving

Abstract

The invention discloses a pile body integrity intelligent detection method and system based on temperature field simulation data driving, and the method comprises the steps: constructing a mixed data set which is formed by mixing a temperature field simulation database and a pile body integrity field detection database, each data point comprises a working condition parameter of the cast-in-place pile, temperature field data and a pile body integrity category; the temperature field simulation database is constructed through a foundation pile hydration heat finite element analysis model; based on a machine learning regression model, training by utilizing the mixed data set to obtain a pile body integrity intelligent judgment model; and temperature field data and corresponding working condition parameters of the to-be-detected cast-in-place pile are obtained and input into the pile body integrity intelligent judgment model, and the pile body integrity category of the to-be-detected cast-in-place pile is output. According to the method, on the basis of hydration heat temperature field distribution, through a physical simulation and actual measurement data mixed training model, early-stage rapid detection of the formed pile can be achieved, geological interference and real defects are effectively distinguished by introducing multi-dimensional working condition parameters, and the detection reliability is improved.

Description

Technical Field

[0001] This invention belongs to the field of pile foundation testing technology, specifically, it relates to an intelligent detection method and system for pile integrity driven by temperature field simulation data. Background Technology

[0002] Cast-in-place piles are widely used in building construction, municipal engineering, and bridge projects due to their high bearing capacity and flexible construction. However, the construction process is complex and influenced by various factors such as geological conditions and construction techniques, making them prone to defects such as necking, mud inclusion, segregation, honeycombing, and loosening, which affect the pile's bearing capacity and service life. Therefore, quality control and integrity testing of cast-in-place piles are crucial.

[0003] Currently, the main methods for testing the integrity of cast-in-place piles include high-strain dynamics, low-strain dynamics, sonic logging, and core drilling. However, these methods have the following drawbacks: First, they have strict requirements on the pile age and concrete strength. Low-strain dynamics and sonic logging require the concrete strength to be no less than 70% of the design strength, while core drilling and high-strain dynamics require the concrete to be at least 28 days old, making it difficult to conduct testing in the early stages of pile formation and affecting the construction schedule. Second, data analysis still relies mainly on manual judgment combined with standards and engineering experience, which has a certain degree of subjectivity.

[0004] Utilizing the heat of hydration of concrete for pile integrity testing is a relatively new technology developed in recent years. The heat of hydration is generated during concrete setting, typically reaching its peak within 48 hours. In homogeneous, uniform-diameter piles without defects, the temperature at measuring points at the same horizontal position at different depths is essentially the same; however, at locations with defects such as fractures, cracks, necking, mud inclusions, or voids, the temperature at these measuring points is significantly lower than at locations without defects. Therefore, collecting temperature field data allows for the early identification of pile defects. However, existing heat of hydration testing technologies still have the following shortcomings: First, they only provide the testing method and cannot directly determine the type of pile integrity, failing to meet the standardized requirements of engineering acceptance; second, pile defects and complex geological conditions exhibit similar temperature curves, making it difficult to distinguish between genuine defects and geological interference; third, because both pile tip defects and normal pile tips show a decreasing trend on the temperature curve, it is difficult to accurately determine pile tip defects. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings and deficiencies of the existing technology and provide an intelligent detection method for pile integrity based on temperature field simulation data. This method can achieve rapid detection in the early stage of pile formation and effectively distinguish between geological interference and real defects, avoiding the subjectivity of human experience judgment.

[0006] The second objective of this invention is to provide an intelligent detection system for pile integrity driven by temperature field simulation data.

[0007] The objective of this invention is achieved through the following technical solution: an intelligent detection method for pile integrity driven by temperature field simulation data, comprising the following steps:

[0008] S1. Construct a hybrid dataset, which is composed of a temperature field simulation database and a pile integrity field detection database. Each data point in the hybrid dataset contains the working condition parameters of the cast-in-place pile, temperature field data, and the corresponding pile integrity category. The temperature field simulation database is constructed using a pile hydration heat finite element analysis model, which is based on the heat conduction mechanism and calibrated using pile integrity field detection data.

[0009] S2. Based on the machine learning regression model, a pile integrity intelligent judgment model is trained using the hybrid dataset;

[0010] S3. Obtain the working condition parameters corresponding to the pile under test and the temperature field data distributed along the depth direction of the pile.

[0011] S4. Input the working condition parameters and temperature field data from step S3 into the intelligent determination model for pile integrity. The intelligent determination model for pile integrity outputs the pile integrity category of the cast-in-place pile to be tested.

[0012] Preferably, the construction of the finite element analysis model for the hydration heat of the foundation pile includes the following steps:

[0013] S11. Establish a 1:1 geometric model including the foundation pile entity and the soil and rock entity;

[0014] S12. Set the geotechnical thermal data, hydration heat parameters, and defect parameters for the geometric model;

[0015] S13. Set boundary conditions and initial conditions based on Fourier's heat transfer law to obtain the initial finite element analysis model;

[0016] S14. Compare the temperature curve output by the initial finite element analysis model with the temperature data collected from the on-site pile integrity detection, and calibrate the initial finite element analysis model by iteratively adjusting the geotechnical thermal data, hydration heat parameters, defect parameters, and boundary conditions until the fitting requirements are met, thus obtaining the finite element analysis model of pile hydration heat:

[0017] ,

[0018] in, Let T represent the mapping function, S represent the temperature field data, P represent the soil and rock parameters, P represent the pile foundation parameters, and E represent the environmental parameters. This represents the boundary condition parameters.

[0019] Preferably, the construction of the temperature field simulation database specifically includes:

[0020] Step 3.1: Using the finite element analysis model of the hydration heat of the foundation pile, with the working condition parameters including the preset defect parameters as variables, the simulation calculations under different working condition parameters are performed in batches through an automated script to obtain simulation data composed of the working condition parameters [S,P,E] and the corresponding temperature field data T.

[0021] Step 3.2: Based on the preset defect parameters, assign a corresponding pile integrity category C as a label to each set of simulation data to form a temperature field simulation database composed of data points [S,P,E,T,C].

[0022] Preferably, the construction of the on-site pile integrity testing database specifically includes:

[0023] Step 4.1: Introduce preset defect types into the test piles by artificial means to construct field test piles with corresponding pile integrity categories;

[0024] Step 4.2: After the concrete is poured, collect the temperature sensor readings along the longitudinal direction of the pile to obtain the temperature field data T, and simultaneously record the corresponding working parameters [S,P,E] to form the measured data;

[0025] Step 4.3: Use existing pile integrity testing methods to verify whether the actual pile integrity category of the field test pile matches the preset defect type;

[0026] Step 4.4: Assign a verified and consistent pile integrity category C as a label to each set of measured data, forming a pile integrity field detection database composed of data points [S,P,E,T,C].

[0027] Preferably, the training of the intelligent determination model for pile integrity includes:

[0028] S21. Randomly mix the temperature field simulation database with the pile integrity field detection database to obtain a mixed dataset. Use the working condition parameters [S,P,E] and temperature field data T as feature variables and the pile integrity category C as the target variable for model training.

[0029] S22. Determine the candidate machine learning regression model types and the corresponding hyperparameter search range;

[0030] S23. Construct a joint search space Ω, wherein each candidate configuration θ of the joint search space Ω is composed of a candidate machine learning regression model type and the corresponding hyperparameter combination;

[0031] S24. Using the hybrid dataset, perform K-fold cross-validation on each candidate configuration θ, and calculate the average coefficient of determination and average mean squared error for each candidate configuration θ:

[0032] ,

[0033] ,

[0034] Where K is the cross-validation fold number, Let be the coefficient of determination of the model on the validation set at the k-th fold validation. Let be the mean squared error of the model on the validation set at the k-th fold validation;

[0035] S25. With the constraints that the average coefficient of determination is greater than 0.95 and the average mean square error is less than 0.05, the candidate configuration with the largest average coefficient of determination is selected as the global optimal configuration ω.

[0036] S26. Based on the global optimal configuration ω, retrain using the entire mixed dataset to obtain the intelligent judgment model for pile integrity.

[0037] Preferably, in step S2, the machine learning regression model includes linear regression model, ridge regression model, lasso regression model, decision tree regression model, random forest regression model, support vector regression model, and neural network regression model.

[0038] Preferably, the working condition parameters include soil and rock parameters, pile parameters, and environmental parameters. The soil and rock parameters include the type and thickness of each soil and rock layer, as well as whether it is below the groundwater level. The pile parameters include pile diameter, pile length, sleeve length, cement type, water-cement ratio, defect location, defect length, and defect type. The environmental parameters include the initial ambient temperature.

[0039] The temperature field data includes the mapping relationship between the temperature at each measuring point and the corresponding depth during the peak of the heat of hydration.

[0040] The pile integrity category is 1, 2, 3, and 4, corresponding to Class I piles, Class II piles, Class III piles, and Class IV piles, respectively.

[0041] Preferably, the hydration heat parameter adopts an exponential simplified model:

[0042] ,

[0043] Where Q(t) represents the total heat of hydration, Q0 and m are model constant parameters related to cement type and water-cement ratio, and t is time.

[0044] Preferably, the defect parameters include defect location, defect length, defect type, and the thermal conductivity or heat of hydration parameter of the concrete at the defect location, which is assigned a value differentiated according to the defect type.

[0045] A pile integrity intelligent detection system driven by temperature field simulation data, applied to the above-mentioned methods, includes:

[0046] A hybrid dataset construction module is used to construct a hybrid dataset, which is composed of a temperature field simulation database and a pile integrity field detection database. Each data point in the hybrid dataset contains the working condition parameters of the cast-in-place pile, temperature field data, and the corresponding pile integrity category. The temperature field simulation database is constructed through a finite element analysis model of pile hydration heat, which is based on the heat conduction mechanism and calibrated using pile integrity field detection data.

[0047] The model training module is used to train an intelligent judgment model for pile integrity based on the machine learning regression model and the hybrid dataset.

[0048] The data acquisition module is used to acquire the working condition parameters of the pile under test and the temperature field data distributed along the depth direction of the pile.

[0049] The intelligent judgment module is used to input the working condition parameters and temperature field data from the data acquisition module into the intelligent judgment model of pile integrity, and the intelligent judgment model of pile integrity outputs the pile integrity category of the cast-in-place pile to be tested.

[0050] The present invention has the following advantages and effects compared with the prior art:

[0051] (1) This invention constructs a hybrid dataset consisting of a temperature field simulation database and a pile integrity field detection database. The temperature field simulation database is constructed using a finite element analysis model of pile hydration heat calibrated with field detection data of cast-in-place pile integrity. The hybrid dataset is then used to train a machine learning regression model to obtain an intelligent pile integrity judgment model. The collected temperature field data and working parameters are used as inputs, and the intelligent pile integrity judgment model directly outputs pile integrity categories I-IV. This method, through mixed training of physical simulation and measured data, endows the model with clear physical interpretability, avoids the subjectivity of manual experience judgment, and realizes rapid detection in the early stage of pile formation.

[0052] (2) Based on temperature field data, this invention introduces multi-dimensional working condition parameters (soil and soil parameters, pile parameters and environmental parameters) as input features of the intelligent judgment model of pile integrity, so that the intelligent judgment model of pile integrity can establish temperature distribution benchmarks under different working conditions, accurately identify abnormal features caused by defects, and solve the limitation of traditional methods that rely solely on temperature curves to cope with complex geological working conditions.

[0053] (3) This invention generates a large amount of data covering different geological conditions and defect types through automated multi-condition simulation, and trains a machine learning regression model with verified measured data to obtain an intelligent judgment model for pile integrity. The mixed dataset contains both the influence characteristics of defects on the temperature field and the interference characteristics of geological conditions and ambient temperature on the temperature field, enabling the intelligent judgment model for pile integrity to automatically distinguish between the two, effectively eliminate geological interference in actual detection, and accurately identify real defects.

[0054] (4) This invention constructs a joint search space, using model type and hyperparameters as unified variables for joint optimization. This allows multiple machine learning regression model types and their hyperparameter combinations to participate in the evaluation in parallel. Cross-validation is performed on all candidate configurations, with the average coefficient of determination greater than 0.95 and the average mean square error less than 0.05 as constraints. The candidate configuration with the largest average coefficient of determination is selected as the global optimal configuration, and then the intelligent pile integrity judgment model is trained based on the global optimal configuration. This selection criterion can screen out models with strong explanatory power for changes in integrity categories and small prediction bias, ensuring the reliability of engineering applications. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the intelligent detection method for pile integrity based on temperature field simulation data according to the present invention.

[0056] Figure 2 This is a schematic diagram of the training process for the intelligent determination model of pile integrity of the present invention.

[0057] Figure 3 This is a schematic diagram illustrating the construction process of the temperature field simulation database of this invention.

[0058] Figure 4 This is a schematic diagram illustrating the process of constructing the on-site detection database for pile integrity according to the present invention. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0060] Example 1

[0061] like Figure 1As shown in the figure, this embodiment details the overall process of the intelligent pile integrity detection method driven by temperature field simulation data according to the present invention. The method includes four stages: training data preparation, model training, on-site data acquisition, and intelligent judgment.

[0062] Step 1.1: Constructing a Hybrid Dataset

[0063] A hybrid dataset was constructed, consisting of a temperature field simulation database and a pile integrity field testing database. The temperature field simulation database was built using a finite element analysis model of pile hydration heat, based on heat conduction mechanisms and calibrated using field testing data of cast-in-place pile integrity. The field testing database collected measured data from field cast-in-place pile tests with artificially created defects, and after verification using existing pile integrity testing methods, a measured temperature field database with validated and consistent integrity category labels was established. The construction details of the two databases are described in detail in Examples 3 and 4, respectively.

[0064] Step 1.2: Training the intelligent judgment model

[0065] Based on a machine learning regression model, an intelligent model for determining pile integrity is trained using the hybrid dataset constructed in step 1.1. To improve the model's generalization ability and prediction accuracy, this embodiment adopts a joint search space optimization training strategy, the specific training process of which is described in detail in embodiment 2.

[0066] Step 1.3, On-site data collection

[0067] (1) Test preparation and sensor deployment: The detection device used in this embodiment includes a host, a temperature acquisition port, a temperature sensor and a signal transmission cable. The host is a computer or microprocessor with data acquisition, recording, processing and display functions; the temperature acquisition port is used to transmit the temperature data collected by the temperature sensor; the temperature sensor and the signal transmission cable are wear-resistant and corrosion-resistant to adapt to the underground humid and corrosive environment.

[0068] First, check the equipment's operational status to ensure the data acquisition system is functioning correctly, and calibrate all temperature sensors. In this embodiment, ensure their accuracy is better than or equal to 0.1℃ and their maximum operating temperature is not less than 80℃ to meet the measurement requirements of the high-temperature environment of hydration heat. Mark each temperature sensor with a unique serial number, connect it to the temperature acquisition port, and check the longitudinal continuity of the temperature data through a short-term test.

[0069] Then, the sensor cables are symmetrically and equidistantly tied to the longitudinal reinforcing bars of the rebar cage. In this embodiment, the sensor spacing is 500mm, which effectively captures the temperature distribution details along the depth of the pile. The distances from the temperature sensors at the maximum and minimum elevations to the pile top and pile end, respectively, are close to the thickness of the rebar cage protective layer, ensuring that the temperature sensors are embedded inside the pile concrete, accurately reflecting the core temperature and avoiding interference from ambient temperature.

[0070] (2) Recording of working parameters: For the cast-in-place pile to be tested, obtain the working parameters of the cast-in-place pile. Record the geological data near the cast-in-place pile (type and thickness of each soil and rock layer and groundwater level information) to form the soil and rock parameters S. Record the pile design parameters (pile diameter D, pile length L, sleeve length l, cement type s, water-cement ratio r) to form the pile parameters P. Obtain the initial ambient temperature T during concrete pouring through meteorological data or on-site measurement. e This constitutes the environmental parameter E. The above working condition parameters can be determined before concrete pouring and correspond to the subsequently collected temperature field data, together serving as the input features of the intelligent judgment model for pile integrity.

[0071] (3) Temperature data acquisition: After the concrete is poured, the data acquisition system is started immediately to continuously record the readings of all temperature sensors. Since the heat of hydration of concrete reaches its peak within 48 hours after pouring, this embodiment starts data acquisition immediately after pouring, without waiting for the concrete to reach 28 days of age or 70% of the design strength, thus achieving rapid detection in the early stage of pile formation. When the temperature of all sensors shows a downward trend (the heat of hydration begins to decrease after reaching its peak), the data acquisition is stopped. The temperature-time curve distributed along the depth direction of the pile is exported, and the temperature value of each measuring point at the peak of the heat of hydration and the corresponding depth are extracted to form the temperature field data T, which is represented as T={[T1,d1],[T2,d2],…,[T u ,d u ]}, where u is the measuring point number along the depth direction of the pile, T u Let d be the temperature value at the u-th measuring point at the peak of the heat of hydration. u Let be the depth corresponding to the u-th measuring point. Thus, the eigenvector X=[S,P,E,T] of the cast-in-place pile is obtained.

[0072] Step 1.4: Intelligent Determination of Pile Integrity

[0073] Input the feature vector X=[S,P,E,T] obtained in step 1.3 into the intelligent determination model of pile integrity trained in step 1.2.

[0074] (1) Model Inference: The intelligent determination model for pile integrity is constructed based on machine learning regression algorithm. Taking the random forest regression model as an example, when the feature vector X is input into the model, each decision tree in the forest performs inference independently: starting from the root node, it traverses layer by layer according to the decision rules inside the node, and finally falls into a unique leaf node. Each leaf node stores the average value of the pile integrity category of all samples that fall into the node in the training set.

[0075] (2) Integration output: Traverse all N decision trees (N is the number of decision trees in the forest) to obtain N predicted values. Calculate the arithmetic mean of all single-tree predicted values ​​as the output value of the random forest regression model, i.e., the stake integrity category.

[0076] (3) Category Mapping: Since the model output is a continuous numerical value, it needs to be rounded or thresholded to convert it into a discrete integrity category: 1 corresponds to Class I pile (pile body intact), 2 corresponds to Class II pile (minor defects, not affecting bearing capacity), 3 corresponds to Class III pile (obvious defects, affecting bearing capacity), and 4 corresponds to Class IV pile (serious defects, requiring rework). This output result directly matches the national standard Class I-IV pile integrity categories, avoiding the subjectivity of manual experience judgment.

[0077] Example 2

[0078] like Figure 2 As shown in the illustration, this embodiment details the construction and training process of the intelligent pile integrity assessment model. First, data fusion and feature engineering are performed. Then, by constructing a joint search space, the model type (machine learning regression model type) and hyperparameters are jointly optimized as unified variables. Finally, through a unified cross-validation process and selection criteria, the globally optimal model type and its hyperparameter combination are directly determined, ensuring that the final intelligent pile integrity assessment model has high accuracy and strong generalization ability. The specific steps are as follows:

[0079] Step 2.1: Data Fusion and Feature Definition

[0080] The temperature field simulation database constructed in Example 3 and the pile integrity field detection database constructed in Example 4 were randomly mixed to obtain a hybrid dataset. The purpose of random mixing is to avoid order bias in the data from the two sources during training and to improve the generalization ability of the model.

[0081] Each data point in the mixed dataset is represented as [S, P, E, T, C], where S represents soil and rock parameters, P represents pile parameters, E represents environmental parameters, T represents temperature field data, and C represents the pile integrity category (values ​​1, 2, 3, and 4, corresponding to Class I, Class II, Class III, and Class IV piles, respectively). In subsequent training, [S, P, E, T] are used as feature variables, and C is used as the target variable.

[0082] Step 2.2: Determine the candidate model type and the corresponding hyperparameter search range.

[0083] This embodiment selects a variety of machine learning regression models as candidate models, including linear regression model, ridge regression model, lasso regression model, decision tree regression model, random forest regression model, support vector regression model and neural network regression model.

[0084] Define the hyperparameter search range for the selected candidate model type. Taking Random Forest as an example, its hyperparameters include:

[0085] Number of decision trees n estimators The value range is [50, 100, 200, 500].

[0086] Maximum depth of decision tree depth The value range is [5, 10, 20, unlimited].

[0087] Minimum number of samples required for node splitting (min) samples_split The value range is [2, 5, 10].

[0088] The maximum number of features extracted during node splitting. features The value range is [square root sampling, logarithmic sampling];

[0089] For other models with hyperparameters, a similar method is used to define their hyperparameter search range. For models without hyperparameters or with fixed hyperparameters, the default parameters are used directly.

[0090] Step 2.3: Construct the joint search space Ω

[0091] We also consider the type of machine learning regression model as a dimension of hyperparameters to construct a unified joint search space Ω. Each element in the joint search space Ω is called a candidate configuration θ, which consists of a specific machine learning regression model type and its corresponding set of hyperparameters. For example: the joint search space Ω = {linear regression_no parameters, ridge regression_alpha∈{0.1,1.0}, …}.

[0092] This joint search space uses model type and hyperparameters as unified variables for joint optimization, enabling multiple machine learning regression model types and their hyperparameter combinations to participate in the evaluation in parallel, avoiding the problem of missing better models if the model is selected first and then the parameters are tuned.

[0093] Step 2.4: K-fold cross-validation and performance calculation

[0094] For each candidate configuration in the joint search space, a unified K-fold cross-validation is performed (K=5 in this embodiment) to ensure that the performance metrics of each candidate configuration are comparable. The specific process is as follows:

[0095] The mixed dataset obtained in step 2.1 is randomly divided into 5 non-overlapping subsets of similar size;

[0096] For each candidate configuration θ, perform 5 loops, each time using 4 subsets as the training set and the remaining 1 subset as the validation set;

[0097] Train the model on the training set based on the current candidate configuration θ, and make predictions on the validation set, calculating the decision coefficient of the k-th fold validation. and mean square error ;

[0098] After completing 5 iterations, calculate the average coefficient of determination for the candidate configuration θ. and mean square error :

[0099] ,

[0100] ,

[0101] In the above formula, the coefficient of determination R is used. 2 Ri and mean squared error (MSE) are used as performance evaluation metrics for the model. 2 The value of MSE represents the model's explanatory power for variations in pile integrity categories; a value closer to 1 indicates a better fit. MSE measures the average deviation between the model's predicted values ​​and the actual values; a smaller value indicates higher prediction accuracy. The formulas for both are as follows:

[0102] ,

[0103] ,

[0104] Among them, y i Let i be the true value of the pile integrity category for the i-th sample. Let i be the predicted value for the pile integrity category of the i-th sample. Let be the mean of the true values ​​of the pile integrity category for the i-th sample.

[0105] Step 2.5, Joint Optimal Selection Criteria

[0106] Based on the joint selection criterion, the globally optimal configuration ω is selected from the joint search space Ω. The joint selection criterion is as follows: with the constraints that the average coefficient of determination is greater than 0.95 and the average mean square error is less than 0.05, the candidate configuration with the largest average coefficient of determination is selected as the globally optimal configuration ω. This criterion is expressed as:

[0107] ,

[0108] Let the candidate configuration that satisfies the above criteria be the global optimal configuration ω. ω specifically includes the optimal model type and its corresponding optimal hyperparameter combination, i.e., ω = [optimal model type, optimal hyperparameter combination].

[0109] This configuration simultaneously satisfies:

[0110] That is, the prediction accuracy meets the engineering requirements;

[0111] and The highest among all candidate configurations that meet the accuracy requirements indicates the best model interpretability.

[0112] in, This ensures that the model can explain more than 95% of the variation in the temperature field data. This ensures that the average deviation between the predicted and actual values ​​is within an acceptable range for engineering applications. These two conditions together ensure that the selected model has high reliability and practicality. If none of the candidate configurations are satisfied... If the conditions are not met, the range of values ​​in the joint search space needs to be adjusted or the mixed dataset needs to be expanded.

[0113] Step 2.6: Final Model Training

[0114] Based on the global optimal configuration ω, all training and validation sets used in cross-validation (i.e., all mixed datasets obtained in step 2.1) are merged. Using the model type and optimal hyperparameters determined by the global optimal configuration, the model is trained using conventional training methods in this field (such as the least squares method of linear regression, the backpropagation algorithm of neural networks, etc.) to obtain the final intelligent judgment model for pile integrity.

[0115] This training refers to combining the scattered training and validation sets from the original cross-validation into a full mixed dataset after determining the optimal model type and optimal hyperparameters, in order to maximize data utilization and improve the model's generalization ability.

[0116] The specific training process of the random forest regression model as an example:

[0117] If the optimal model in the globally optimal configuration ω is a random forest regression model, its final training process is as follows:

[0118] (1) Bootstrap sampling: Using bootstrap sampling with replacement, M subsets are generated from the entire mixed dataset:

[0119] ,

[0120] Among them, D m Let X represent the m-th subset of the dataset. iLet y be the feature vector of the i-th sample. i Here, N represents the true values ​​for the corresponding pile integrity category, M represents the total number of samples, and M represents the number of decision trees. Each subset of the dataset is used to train one decision tree. Sampling with replacement ensures that the training data for each tree is slightly different, increasing the diversity of the model and thus improving its generalization ability.

[0121] (2) Node splitting: At each node of each decision tree, a preset number of features (2 in this embodiment) are randomly selected from the feature vector X=[S,P,E,T]. The optimal splitting variable k and splitting point s are determined based on minimizing the squared error. For any splitting variable k and splitting point s, the solution is:

[0122] ,

[0123] in, and Two sample regions are defined based on the segmentation variable k and the segmentation point s. and The mean values ​​of pile integrity categories for all samples within regions R1 and R2, respectively. i Let be the true value of the pile integrity category for the i-th sample. The optimal (k,s) is recursively searched until a stopping condition is met (e.g., the number of node samples is less than a threshold, the maximum depth is reached, etc.), generating a complete decision tree. This splitting criterion minimizes the sum of squared errors of samples in the two regions after the split, thereby improving the prediction accuracy of the decision tree.

[0124] (3) Integration output: The arithmetic mean of the predictions from all decision trees is taken to obtain the final average value of the pile integrity category. :

[0125] ,

[0126] in, This represents the predicted value of the m-th decision tree for the input feature vector X. The average value for each category of pile integrity is also represented. Since the values ​​are continuous, they need to be rounded or thresholded to convert them into completeness categories (Classes I-IV) before output. This averaging process effectively offsets the prediction bias of a single decision tree, improving the stability and accuracy of the model.

[0127] Specifically, this embodiment selects the globally optimal configuration with the constraints of an average coefficient of determination greater than 0.95 and an average mean square error less than 0.05, and retrains it using the full mixed dataset to obtain an intelligent judgment model with strong explanatory power for changes in pile integrity categories and small prediction bias. This training strategy ensures the reliability and generalization ability of the model in practical engineering applications.

[0128] Example 3

[0129] like Figure 3 As shown, this embodiment details the construction process of the temperature field simulation database, providing simulation data for Embodiment 2. First, a finite element analysis model of the pile hydration heat is constructed through geometric modeling, assigning physical parameters, setting boundary conditions, and iterative calibration and verification using field inspection data of the pile body integrity. Then, automated scripts are used to perform batch simulations under multiple working conditions, generating a large amount of labeled data covering various geological conditions, pile type parameters, and defect types, providing rich sample data for subsequent machine learning model training. The specific steps are as follows:

[0130] Step 3.1: Establish a geometric model

[0131] A 1:1 geometric model was established, comprising the pile body and the soil and rock entities. Based on the actual geological data from the on-site inspection of the pile body integrity, the soil and rock entities were divided into layers. Specifically, based on the types, thicknesses, and groundwater levels of each soil and rock layer provided in the geological survey report, clear dividing lines were set in the geometric model, including: soil layer boundaries (such as the interface between gravelly soil and sandy soil layers) and groundwater levels (to distinguish between saturated and unsaturated soil). Using the test results obtained through conventional testing methods such as core drilling and low-strain methods, the pile body at defective locations was geometrically segmented, providing a basis for assigning different material properties subsequently.

[0132] Step 3.2: Set physical parameters for the geometric model

[0133] (1) Geotechnical thermal data: Based on the geotechnical stratification and groundwater level information, the density, specific heat, and heat transfer coefficient of each geotechnical layer are set. If no measured data is available, the equivalent heat transfer coefficient can be calculated using empirical formulas. :

[0134] ,

[0135] Where n is the porosity of the soil and rock. Let be the heat transfer coefficient of water. The heat transfer coefficient is the density of the soil particle skeleton. The reference ranges for density, specific heat, and heat transfer coefficient are 2250–2450 kg / m³, 850–1150 J / (kg·℃), and 1.5–2.5 W / (m·℃), respectively.

[0136] (2) Normal concrete hydration heat parameters: Hydration heat parameters are set based on an exponential simplified model to describe the change of total concrete hydration heat over time:

[0137] ,

[0138] Where Q(t) represents the total heat of hydration, Q0 and m are model constant parameters related to cement type s and water-cement ratio r, and t is time; the reference range for Q0 is 80-200 kJ / kg, and m can be taken as 0.05. The rate of heat of hydration generation can be obtained by differentiating with respect to time t. :

[0139] ,

[0140] In practical applications, the relationship between cement type s, water-cement ratio r, and Q0 and m can be established through literature data or hydration heat experiments, such as... , .

[0141] (3) Material parameters of the defect area: Different defect types have different influence mechanisms on the temperature field. Therefore, based on the actual defect situation (in this embodiment, it refers to the defect type obtained based on the detection results), differentiated values ​​are assigned to the defect. For example, for mud inclusion and segregation defects, half of the thermal conductivity and hydration heat parameter of normal concrete is used; for honeycomb and loose defects, the thermal conductivity is set to be close to 0; for necking defects, the values ​​of the geometric model and the solid are adjusted by using the actual necking size.

[0142] Step 3.3: Set boundary conditions and initial conditions

[0143] Based on the initial ambient temperature T e Set the thermal convection parameters between the foundation pile and the upper surface of the soil and rock, and set the side surface and the ground surface to a fixed temperature, i.e., the initial ambient temperature T. e The heat transfer process is governed by Fourier's law of heat transfer:

[0144] ,

[0145] Among them, C v T is the heat capacity (the product of density and specific heat). ’ The transient temperature field represents the temperature value at any location and time within the pile body and the soil / rock, where λ is the heat transfer coefficient. For the rate of heat of hydration, the divergence term It describes the propagation of heat in the pile soil. The temperature gradient represents the direction of heat flow from the high-temperature zone to the low-temperature zone. This equation describes the balance between heat conduction and the generation of internal heat sources in the pile-soil system. After completing the above settings, an initial finite element analysis model based on actual field test data is obtained.

[0146] Step 3.4, Model Calibration and Validation

[0147] The longitudinal path temperature curve output by the initial finite element analysis model was compared with the actual temperature-depth curve collected from the field inspection of the pile body integrity in Example 4. Since the parameters of the initial finite element analysis model were set based on theoretical values ​​or empirical formulas, there were deviations from the actual working conditions. It was necessary to iteratively adjust the geotechnical thermal data, hydration heat parameters, defect parameters, and boundary conditions within a physically reasonable range to calibrate the initial finite element analysis model. During the adjustment process, the adjustment range of each parameter is determined according to its physical meaning: geotechnical thermal data are fine-tuned within the reference range of the standard (e.g., density 2250-2450 kg / m³, specific heat 850-1150 J / (kg·℃), heat transfer coefficient 1.5-2.5 W / (m·℃)); the hydration heat parameter Q0 is adjusted within the range of 80-200 kJ / kg, and m is fine-tuned around 0.05; the defect parameters are assigned different values ​​according to the defect type (e.g., mud inclusions and segregation are assigned half of the normal value, and the thermal conductivity of honeycomb and loose materials is set close to 0); the boundary conditions are set according to the measured ambient temperature, and the convection coefficient is fine-tuned within a reasonable range.

[0148] Through repeated iterative comparisons and adjustments, until the degree of fit between the simulation curve and the measured curve meets the engineering requirements (e.g., the temperature error at key points is less than the threshold, and the curve shape is consistent), the finite element analysis model of the hydration heat of the foundation pile is obtained:

[0149] ,

[0150] in, Let T represent the mapping function, S represent the temperature field data, P represent the soil and rock parameters, P represent the pile foundation parameters, and E represent the environmental parameters. This represents the boundary condition parameters.

[0151] To verify the applicability of the model under various working conditions, representative field tests of different types were conducted (such as concrete with different mix proportions and cement types, different pile diameters, different rock-embedded working conditions, and different soil and rock geological conditions). The above comparison and adjustment process was repeated until the simulation curves and measured curves under all working conditions met the fitting requirements, thus completing the verification of the finite element analysis model. This calibration and verification process ensures the reliability and universality of the model.

[0152] Step 3.5, Simulation and Data Acquisition

[0153] A Python automated script was written to generate various thermal conditions for cast-in-place pile formation by randomly varying working condition parameters, including preset defect parameters, within a preset domain. The pile hydration heat finite element analysis model obtained in step 3.4 was used for batch simulation calculations to obtain simulation data consisting of working condition parameters [S, P, E] and corresponding temperature field data T, where:

[0154] (1) Geotechnical parameters S: including the type of different soil and rock layers j (gravelly soil, sandy soil, silty soil, cohesive soil, bedrock, etc.), thickness h, and whether it is below the groundwater level (0 indicates above the groundwater level, 1 indicates below), expressed as S=[[j1,h1,0],[j2, h2,0],[j3, h3,1],...];

[0155] (2) Pile parameters P: including pile diameter D, pile length L, sleeve length l, cement type s, water-cement ratio r, and preset defect parameters (defect location d). f Defect length l f Defect category f), where defect category f includes necking, inclusion, segregation, honeycomb, looseness, etc., represented as P=[D,L,l,s,r,d] f ,l f ,f];

[0156] (3) Environmental parameter E: different initial ambient temperatures T e This can be represented as E=[T e ].

[0157] For each thermal condition, extract the temperature value T at each measuring point at the peak of the heat of hydration. u and corresponding depth d u This forms the temperature field data T, represented as T={[T1,d1],[T2,d2],…,[T u ,d u ]}.

[0158] Step 3.6: Tag assignment and database formation

[0159] Based on the preset defect parameters in each set of simulation data, the corresponding pile integrity category C is directly assigned as a label, where C takes values ​​of 1, 2, 3, and 4, corresponding to Class I, Class II, Class III, and Class IV piles, respectively. Each set of simulation data constitutes a data point [S, P, E, T, C], forming a temperature field simulation database. This temperature field database includes not only the influence of different defect types on the temperature field but also the interference of complex geological and environmental conditions on the temperature field, providing rich training samples for subsequent machine learning regression models.

[0160] Example 4

[0161] like Figure 4As shown, this embodiment details the construction process of the pile integrity field inspection database. This measured database provides measured data for Embodiment 2 and also provides a verification basis for the finite element analysis model in Embodiment 3. This embodiment mainly collects measured data through on-site cast-in-place pile tests with artificially created defects. After verification using conventional testing methods, a measured temperature field database (i.e., the pile integrity field inspection database) with verified and consistent integrity category labels is established, providing real sample data for subsequent machine learning regression model training. The specific steps are as follows:

[0162] Step 4.1, Construction of test piles

[0163] Pre-defined defect types are artificially introduced into test piles to construct field test piles with corresponding pile integrity categories. Defect types include necking, mud inclusion, segregation, honeycombing, and loosening, and their location, length, and severity can be set as needed. For example:

[0164] For necking defects, necking molds can be tied at a specific depth in the rebar cage;

[0165] For mud inclusion defects, sandbags filled with different soil types can be tied inside and outside the steel cage. The size of the sandbags determines the length of the defect, and the type of sand inside the sandbags can simulate different mud inclusion properties.

[0166] For segregation and honeycomb defects, concrete or filler with a specific mix ratio can be placed at a specific depth.

[0167] The defect settings should cover all situations corresponding to piles of types I to IV to ensure the representativeness of the database.

[0168] Step 4.2, Data Acquisition

[0169] The test preparation, sensor deployment, operating parameter recording, and temperature data acquisition were carried out according to the method in step 1.3 of Example 1. Temperature data were continuously collected after concrete pouring. Data collection was stopped when all sensor temperatures showed a decreasing trend (usually within 48 hours after pouring), and the temperature value T at each measuring point at the peak of hydration heat was obtained. u With corresponding depth d u This forms the temperature field data T={[T1,d1],[T2,d2],…,[T u ,d u ]}, and simultaneously record the corresponding operating parameters [S,P,E] to form the measured data [S,P,E,T].

[0170] Step 4.3, Method Validation

[0171] After the test piles have reached the required testing conditions, existing pile integrity testing methods (such as the core drilling method and low-strain method specified in the standards) are used to verify whether the actual pile integrity category matches the preset defect type. By comparing the test results with the preset defects, the correspondence between the preset category and the actual category is confirmed, ensuring the reliability of the data labels.

[0172] Step 4.4, Label Assignment

[0173] Based on the verification results of step 4.3, each set of measured data is assigned a verified and consistent pile integrity category C as a label, where C takes values ​​of 1, 2, 3, and 4, corresponding to Class I, Class II, Class III, and Class IV piles, respectively. This ultimately forms a field detection database for pile integrity consisting of data points [S, P, E, T, C]. The format of each data point is consistent with the simulation data points in Example 3, facilitating subsequent mixed use.

[0174] Example 5

[0175] The intelligent pile integrity detection system driven by temperature field simulation data, applied to the method described in Embodiment 1 above, includes:

[0176] A hybrid dataset construction module is used to construct a hybrid dataset, which is composed of a temperature field simulation database and a pile integrity field detection database. Each data point in the hybrid dataset contains the working condition parameters of the cast-in-place pile, temperature field data, and the corresponding pile integrity category. The temperature field simulation database is constructed through a finite element analysis model of pile hydration heat, which is based on the heat conduction mechanism and calibrated using pile integrity field detection data.

[0177] The model training module is used to train an intelligent judgment model for pile integrity based on the machine learning regression model and the hybrid dataset.

[0178] The data acquisition module is used to acquire the working condition parameters of the pile under test and the temperature field data distributed along the depth direction of the pile.

[0179] The intelligent judgment module is used to input the working condition parameters and temperature field data from the data acquisition module into the intelligent judgment model of pile integrity, and the intelligent judgment model of pile integrity outputs the pile integrity category of the cast-in-place pile to be tested.

[0180] The above embodiments are preferred embodiments of the present invention and are not intended to limit the present invention. Any changes or other equivalent substitutions made without departing from the technical solution of the present invention are included within the protection scope of the present invention.

Claims

1. A method for intelligent detection of pile integrity based on temperature field simulation data, characterized in that, Includes the following steps: S1. Construct a hybrid dataset, which is composed of a temperature field simulation database and a pile integrity field detection database. Each data point in the hybrid dataset contains the working condition parameters of the cast-in-place pile, temperature field data, and the corresponding pile integrity category. The temperature field simulation database is constructed through the finite element analysis model of pile hydration heat, which is based on the heat conduction mechanism and calibrated by field test data of pile integrity. S2. Based on the machine learning regression model, a pile integrity intelligent judgment model is trained using the hybrid dataset; S3. Obtain the working condition parameters corresponding to the pile under test and the temperature field data distributed along the depth direction of the pile. S4. Input the working condition parameters and temperature field data from step S3 into the intelligent determination model for pile integrity. The intelligent determination model for pile integrity outputs the pile integrity category of the cast-in-place pile to be tested.

2. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 1, characterized in that, The construction of the finite element analysis model for the hydration heat of the foundation pile includes the following steps: S11. Establish a 1:1 geometric model including the foundation pile entity and the soil and rock entity; S12. Set the geotechnical thermal data, hydration heat parameters, and defect parameters for the geometric model; S13. Set boundary conditions and initial conditions based on Fourier's heat transfer law to obtain the initial finite element analysis model; S14. Compare the temperature curve output by the initial finite element analysis model with the temperature data collected from the on-site pile integrity detection, and calibrate the initial finite element analysis model by iteratively adjusting the geotechnical thermal data, hydration heat parameters, defect parameters, and boundary conditions until the fitting requirements are met, thus obtaining the finite element analysis model of pile hydration heat: ， in, Let T represent the mapping function, S represent the temperature field data, P represent the soil and rock parameters, P represent the pile foundation parameters, and E represent the environmental parameters. This represents the boundary condition parameters.

3. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 1, characterized in that, The construction of the temperature field simulation database specifically includes: Step 3.1: Using the finite element analysis model of the hydration heat of the foundation pile, with the working condition parameters including the preset defect parameters as variables, the simulation calculations under different working condition parameters are performed in batches through an automated script to obtain simulation data composed of the working condition parameters [S,P,E] and the corresponding temperature field data T. Step 3.2: Based on the preset defect parameters, assign a corresponding pile integrity category C as a label to each set of simulation data to form a temperature field simulation database composed of data points [S,P,E,T,C].

4. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 1, characterized in that, The construction of the on-site pile integrity testing database specifically includes: Step 4.1: Introduce preset defect types into the test piles by artificial means to construct field test piles with corresponding pile integrity categories; Step 4.2: After the concrete is poured, collect the temperature sensor readings along the longitudinal direction of the pile to obtain the temperature field data T, and simultaneously record the corresponding working parameters [S,P,E] to form the measured data; Step 4.3: Use existing pile integrity testing methods to verify whether the actual pile integrity category of the field test pile matches the preset defect type; Step 4.4: Assign a verified and consistent pile integrity category C as a label to each set of measured data, forming a pile integrity field detection database composed of data points [S,P,E,T,C].

5. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 1, characterized in that, The training of the intelligent model for determining pile integrity includes: S21. Randomly mix the temperature field simulation database with the pile integrity field detection database to obtain a mixed dataset. Use the working condition parameters [S,P,E] and temperature field data T as feature variables and the pile integrity category C as the target variable for model training. S22. Determine the candidate machine learning regression model types and the corresponding hyperparameter search range; S23. Construct a joint search space Ω, wherein each candidate configuration θ of the joint search space Ω is composed of a candidate machine learning regression model type and the corresponding hyperparameter combination; S24. Using the hybrid dataset, perform K-fold cross-validation on each candidate configuration θ, and calculate the average coefficient of determination and average mean squared error for each candidate configuration θ: ， ， Where K is the cross-validation fold number, Let be the coefficient of determination of the model on the validation set at the k-th fold validation. Let be the mean squared error of the model on the validation set at the k-th fold validation; S25. With the constraints that the average coefficient of determination is greater than 0.95 and the average mean square error is less than 0.05, the candidate configuration with the largest average coefficient of determination is selected as the global optimal configuration ω. S26. Based on the global optimal configuration ω, retrain using the entire mixed dataset to obtain the intelligent judgment model for pile integrity.

6. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 1, characterized in that, In step S2, the machine learning regression model includes linear regression model, ridge regression model, lasso regression model, decision tree regression model, random forest regression model, support vector regression model and neural network regression model.

7. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 1, characterized in that, The operating parameters include soil and rock parameters, pile parameters, and environmental parameters. The soil and rock parameters include the type and thickness of each soil and rock layer, as well as whether it is below the groundwater level. The pile parameters include pile diameter, pile length, sleeve length, cement type, water-cement ratio, defect location, defect length, and defect type. The environmental parameters include the initial ambient temperature. The temperature field data includes the mapping relationship between the temperature at each measuring point and the corresponding depth during the peak of the heat of hydration. The pile integrity category is 1, 2, 3, and 4, corresponding to Class I piles, Class II piles, Class III piles, and Class IV piles, respectively.

8. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 2, characterized in that, The hydration heat parameters are modeled using an exponentially simplified model: ， Where Q(t) represents the total heat of hydration, Q0 and m are model constant parameters related to cement type and water-cement ratio, and t is time.

9. The intelligent pile integrity detection method based on temperature field simulation data as described in claim 2, characterized in that, The defect parameters include defect location, defect length, defect type, and the thermal conductivity or heat of hydration parameter of the concrete at the defect location, which is assigned a value based on the defect type.

10. A pile integrity intelligent detection system driven by temperature field simulation data, applied to the method described in any one of claims 1-9, characterized in that, include: A hybrid dataset construction module is used to construct a hybrid dataset, which is composed of a temperature field simulation database and a pile integrity field detection database. Each data point in the hybrid dataset contains the working condition parameters of the cast-in-place pile, temperature field data, and the corresponding pile integrity category. The temperature field simulation database is constructed through the finite element analysis model of pile hydration heat, which is based on the heat conduction mechanism and calibrated by field test data of pile integrity. The model training module is used to train an intelligent judgment model for pile integrity based on the machine learning regression model and the hybrid dataset. The data acquisition module is used to acquire the working condition parameters of the pile under test and the temperature field data distributed along the depth direction of the pile. The intelligent judgment module is used to input the working condition parameters and temperature field data from the data acquisition module into the intelligent judgment model of pile integrity, and the intelligent judgment model of pile integrity outputs the pile integrity category of the cast-in-place pile to be tested.

Images