Method and system for detecting porosity of a concrete pile foundation, medium, electronic device

Abstract

The present application provides a kind of porosity detection method and system of concrete pile foundation, medium, electronic equipment, it is related to pile foundation detection technical field.The method comprises: the space-time coordinates of each position of the pile body of target pile foundation are input into the physical information neural network model trained in advance, and the thermal diffusion coefficient spatial distribution field in the hydration process of cementing material inside target pile foundation is obtained by inversion;Based on the quantitative mapping relationship between the thermal diffusion coefficient and porosity established in advance, the thermal diffusion coefficient spatial distribution field is converted into the porosity spatial distribution field of target pile foundation.The present application solves the problems of low detection efficiency and accuracy of pile foundation internal porosity and inability to detect in a large range.

Description

Technical Field

[0001] This invention relates to the field of pile foundation testing technology, specifically to a method and system, medium, and electronic equipment for testing the porosity of concrete pile foundations. Background Technology

[0002] Since pile foundations are located below the ground surface and are mostly constructed using cast-in-place methods, environmental or human factors during construction can easily cause a large number of pores to appear inside the pile foundation, endangering structural safety. Therefore, it is necessary to detect and evaluate the pores inside the pile foundation.

[0003] Current methods for detecting porosity inside pile foundations rely on destructive sampling or indirect assessment using evaluation indicators, resulting in slow assessment processes, low accuracy, and an inability to detect porosity over a large area inside the pile foundation. Summary of the Invention

[0004] To overcome the problems of low efficiency and accuracy in detecting the internal porosity of pile foundations and the inability to detect it over a wide area, this invention provides a method, system, medium, and electronic equipment for detecting the porosity of concrete pile foundations.

[0005] On one hand, the present invention provides a method for detecting the porosity of concrete pile foundations, comprising: The spatiotemporal coordinates of each position of the target pile foundation are input into a pre-trained physical information neural network model to invert the spatial distribution field of the thermal diffusivity coefficient inside the target pile foundation during the hydration process of the cementitious material. Based on the pre-established quantitative mapping relationship between thermal diffusivity and porosity, the spatial distribution field of thermal diffusivity is converted into the spatial distribution field of porosity of the target pile foundation; The spatiotemporal coordinates include spatial coordinates and the time after pouring.

[0006] Optionally, the physical information neural network model includes a shared feature encoder and a thermal diffusivity prediction branch; the spatiotemporal coordinates of each position of the target pile foundation are input into the pre-trained physical information neural network model to invert and obtain the spatial distribution field of the thermal diffusivity inside the target pile foundation during the hydration process of the cementitious material, including: The spatiotemporal coordinates of each position of the target pile foundation are input into a pre-trained physical information neural network model. The spatiotemporal coordinates of each position are then subjected to high-dimensional nonlinear mapping encoding by the shared feature encoder to obtain latent space features. The latent space features are filtered by the feature filtering module of the thermal diffusivity prediction branch to remove feature components that are strongly correlated with the time dimension. The filtered features are decoded by an inversion decoder to obtain the corresponding predicted values ​​of thermal diffusivity. Based on the predicted values ​​of thermal diffusivity at each location, the spatial distribution field of thermal diffusivity inside the target pile foundation during the hydration process of the cementitious material is obtained.

[0007] Optionally, the physical information neural network model has two outputs: the predicted temperature value and the predicted thermal diffusivity value at each location of the pile body. It is obtained by training the model using spatial temperature field observation data and a loss function with embedded thermal conduction physical constraints. The spatial temperature field observation data refers to the three-dimensional spatial temperature distribution data of the pile foundation during the hydration process of the cementitious material; the thermal conduction physical constraint is used to ensure that the temperature field and thermal diffusivity coefficient field output by the model satisfy the thermal conduction control equation within the preset solution domain.

[0008] Optionally, the training process of the physical information neural network model includes: The spatial temperature field observation data is obtained by temperature sensing optical fibers spirally deployed along the test pile foundation, and the spatiotemporal coordinates of each observation point and the spatial temperature field observation data are used as a training sample set. A fusion loss function is constructed based on data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The physical information neural network model is trained based on the training sample set and the fusion loss function; The data fitting loss is used to ensure that the temperature prediction value output by the model is consistent with the observed data of the space temperature field; the prior knowledge loss is used to ensure that the output thermal diffusivity coefficient conforms to the actual prior conditions.

[0009] Optionally, the pitch of the spiral arrangement is 0.2 to 0.5 times the diameter of the target pile foundation, and the pitch is increased in the pile end area.

[0010] Optionally, the physical information neural network model includes a temperature prediction branch and a thermal diffusivity prediction branch; training the physical information neural network model based on the training sample set and the fusion loss function includes: The spatiotemporal coordinates of each observation point in the training sample set are input into the physical information neural network model, and the temperature prediction branch and the thermal diffusivity prediction branch are respectively passed to obtain the temperature prediction value and the thermal diffusivity prediction value of each observation point. Based on the predicted temperature values ​​at each observation point and the observed spatial temperature field data, the data fitting loss is calculated. Dense sampling is performed within the solution domain to obtain the spatiotemporal coordinates of multiple sampling points. The temperature prediction and thermal diffusivity prediction based on the spatiotemporal coordinates of each sampling point are used to calculate the residual of the heat conduction control equation within the solution domain, so as to obtain the physical constraint loss of heat conduction. Based on the gradient of the predicted thermal diffusivity coefficient of each sampling point relative to the spatial coordinates, the upper and lower limits of the thermal diffusivity coefficient, and the boundary heat exchange law of the pile foundation, the prior knowledge loss is calculated. The data fitting loss, heat conduction physical constraint loss, and prior knowledge loss are weighted and fused to obtain the total loss value of the training process. With the goal of minimizing the total loss, the network parameters of the physical information neural network model are updated using the gradient descent algorithm until the model converges.

[0011] Optionally, the physical constraint loss for heat conduction is:

[0012] in, For heat conduction physical constraint loss, This represents the total number of sampling points in dense sampling. For concrete density, The specific heat capacity of concrete, Let represent the derivative of the predicted temperature value at the j-th sampling point with respect to time. This indicates the calculation of divergence. This is the predicted value of the thermal diffusivity at the j-th sampling point. The predicted temperature value for the j-th sampling point. Let be the intensity of the concrete hydration heat source at the j-th sampling point.

[0013] Optionally, the prior knowledge loss is:

[0014]

[0015]

[0016]

[0017] in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, For boundary condition constraint loss, This represents the total number of sampling points in dense sampling. This represents the total number of sampling points at the pile-soil interface. Let the thermal diffusivity be the coefficient at the j-th sampling point. Spatial gradient at that location To correct the linear unit, This is the predicted value of the thermal diffusivity at the j-th sampling point. , These are the upper and lower limits of the thermal diffusivity, respectively. The m-th sampling point at the pile-soil interface Predicted value of thermal diffusivity at that location. This represents the derivative of the predicted temperature value with respect to the direction outside the pile-soil boundary. Let be the convective heat transfer coefficient at the pile-soil interface. For concrete density, The specific heat capacity of concrete, Boundary sampling points Predicted temperature value at time t. Boundary sampling points The ambient temperature at that location.

[0018] Optionally, the quantitative mapping relationship between the thermal diffusivity and porosity is established based on the effective medium theory, and its expression is:

[0019] in, Porosity The effective thermal diffusivity inside the target pile foundation, i.e., the predicted value of the thermal diffusivity. is the thermal diffusivity of the concrete solid; For shape factor; is the thermal diffusivity of the fluid in the pores.

[0020] Optionally, after obtaining the spatial distribution field of porosity of the target pile foundation, the method further includes: Based on the spatial distribution field of porosity of the target pile foundation, a three-dimensional image segmentation algorithm is used to identify high-risk connected regions with porosity greater than a preset threshold.

[0021] On the other hand, the present invention also provides a porosity detection system for concrete pile foundations, comprising: The inversion module is used to input the spatiotemporal coordinates of each position of the target pile foundation into a pre-trained physical information neural network model to invert and obtain the spatial distribution field of the thermal diffusivity coefficient inside the target pile foundation during the hydration process of the cementitious material. The conversion module is used to convert the spatial distribution field of the thermal diffusivity coefficient into the spatial distribution field of the porosity of the target pile foundation based on the pre-established quantitative mapping relationship between the thermal diffusivity coefficient and porosity. The spatiotemporal coordinates include spatial coordinates and the time after pouring.

[0022] Optionally, the physical information neural network model includes a shared feature encoder and a thermal diffusivity prediction branch; the inversion module includes: The encoding submodule is used to input the spatiotemporal coordinates of each position of the target pile foundation into the pre-trained physical information neural network model. The spatiotemporal coordinates of each position are encoded by the shared feature encoder through high-dimensional nonlinear mapping to obtain latent space features. The feature filtering submodule is used to filter the latent space features through the feature filtering module of the thermal diffusivity prediction branch to remove feature components that are strongly correlated with the time dimension. The decoding submodule is used to decode the filtered features through the inversion decoder to obtain the corresponding predicted value of thermal diffusivity; based on the predicted value of thermal diffusivity at each location, the spatial distribution field of thermal diffusivity inside the target pile foundation during the hydration process of cementitious materials is obtained.

[0023] Optionally, the physical information neural network model has two outputs: the predicted temperature value and the predicted thermal diffusivity value at each location of the pile body. It is obtained by training the model using spatial temperature field observation data and a loss function with embedded thermal conduction physical constraints. The spatial temperature field observation data refers to the three-dimensional spatial temperature distribution data of the pile foundation during the hydration process of the cementitious material; the thermal conduction physical constraint is used to ensure that the temperature field and thermal diffusivity coefficient field output by the model satisfy the thermal conduction control equation within the preset solution domain.

[0024] Optionally, it also includes a training module, the training module comprising: The sample acquisition submodule is used to acquire the spatial temperature field observation data through temperature sensing optical fibers spirally deployed along the test pile foundation, and use the spatiotemporal coordinates of each observation point and the spatial temperature field observation data as a training sample set. The loss construction submodule is used to construct a fusion loss function based on data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The training submodule is used to train the physical information neural network model based on the training sample set and the fusion loss function; The data fitting loss is used to ensure that the temperature prediction value output by the model is consistent with the observed data of the space temperature field; the prior knowledge loss is used to ensure that the output thermal diffusivity coefficient conforms to the actual prior conditions.

[0025] Optionally, the pitch of the spiral arrangement is 0.2 to 0.5 times the diameter of the target pile foundation, and the pitch is increased in the pile end area.

[0026] Optionally, the physical information neural network model includes a temperature prediction branch and a thermal diffusivity prediction branch; the training submodule is specifically used for: The spatiotemporal coordinates of each observation point in the training sample set are input into the physical information neural network model, and the temperature prediction branch and the thermal diffusivity prediction branch are respectively passed to obtain the temperature prediction value and the thermal diffusivity prediction value of each observation point. Based on the predicted temperature values ​​at each observation point and the observed spatial temperature field data, the data fitting loss is calculated. Dense sampling is performed within the solution domain to obtain the spatiotemporal coordinates of multiple sampling points. The temperature prediction and thermal diffusivity prediction based on the spatiotemporal coordinates of each sampling point are used to calculate the residual of the heat conduction control equation within the solution domain, so as to obtain the physical constraint loss of heat conduction. Based on the gradient of the predicted thermal diffusivity coefficient of each sampling point relative to the spatial coordinates, the upper and lower limits of the thermal diffusivity coefficient, and the boundary heat exchange law of the pile foundation, the prior knowledge loss is calculated. The data fitting loss, heat conduction physical constraint loss, and prior knowledge loss are weighted and fused to obtain the total loss value of the training process. With the goal of minimizing the total loss, the network parameters of the physical information neural network model are updated using the gradient descent algorithm until the model converges.

[0027] Optionally, the physical constraint loss for heat conduction is:

[0028] in, For heat conduction physical constraint loss, This represents the total number of sampling points in dense sampling. For concrete density, The specific heat capacity of concrete, Let represent the derivative of the predicted temperature value at the j-th sampling point with respect to time. This indicates the calculation of divergence. This is the predicted value of the thermal diffusivity at the j-th sampling point. The predicted temperature value for the j-th sampling point. Let be the intensity of the concrete hydration heat source at the j-th sampling point.

[0029] Optionally, the prior knowledge loss is:

[0030]

[0031]

[0032]

[0033] in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, For boundary condition constraint loss, This represents the total number of sampling points in dense sampling. This represents the total number of sampling points at the pile-soil interface. Let the thermal diffusivity be the coefficient at the j-th sampling point. Spatial gradient at that location To correct the linear unit, This is the predicted value of the thermal diffusivity at the j-th sampling point. , These are the upper and lower limits of the thermal diffusivity, respectively. The m-th sampling point at the pile-soil interface Predicted value of thermal diffusivity at that location. This represents the derivative of the predicted temperature value with respect to the direction outside the pile-soil boundary. Let be the convective heat transfer coefficient at the pile-soil interface. For concrete density, The specific heat capacity of concrete, Boundary sampling points Predicted temperature value at time t. Boundary sampling points The ambient temperature at that location.

[0034] Optionally, the quantitative mapping relationship between the thermal diffusivity and porosity is established based on the effective medium theory, and its expression is:

[0035] in, Porosity The effective thermal diffusivity inside the target pile foundation, i.e., the predicted value of the thermal diffusivity. is the thermal diffusivity of the concrete solid; For shape factor; is the thermal diffusivity of the fluid in the pores.

[0036] Optionally, it also includes: The risk identification module is used to identify high-risk connected regions with porosity greater than a preset threshold based on the spatial distribution field of porosity of the target pile foundation using a three-dimensional image segmentation algorithm.

[0037] On the other hand, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the method described in any of the foregoing is implemented.

[0038] On the other hand, the present invention also provides a readable storage medium having an executable program stored thereon, wherein when the executable program is executed, it implements the method described in any one of the above.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a method and system for detecting the porosity of concrete pile foundations. By acquiring three-dimensional spatial temperature distribution data (i.e., spatial temperature field observation data) of the pile foundation during the hydration process of cementitious materials, and using this data as the basis for model training and inversion, there is no need to perform destructive drilling sampling on the pile body. At the same time, the spatiotemporal coordinates cover all positions of the entire pile body. Combined with a physical information neural network model with thermal diffusivity as the output, a complete three-dimensional thermal diffusivity spatial distribution field can be obtained through inversion, which can then be converted into a porosity spatial distribution field, realizing the detection of porosity of the entire cross section and three-dimensional space of the pile foundation.

[0040] This invention employs a pre-trained physical information neural network model. During actual testing, only the spatiotemporal coordinates of the target pile foundation need to be input into the model to directly output the thermal diffusivity field. This field is then automatically converted into a porosity field through a pre-calibrated mapping relationship. The entire process eliminates the need for iterative solutions to the inverse heat conduction problem, resulting in fast testing speeds suitable for batch applications in the field. Furthermore, the output porosity provides a quantitative numerical distribution, offering direct microstructural parameters for pile foundation quality grading and durability evaluation. Attached Figure Description

[0041] Figure 1 This is a schematic flowchart of a method for detecting the porosity of a concrete pile foundation according to the present invention. Figure 2 This is a schematic diagram of the layout of a spiral fiber optic sensing network for a pile foundation, as an example of the present invention. Figure 3 This is a schematic diagram of a pile foundation testing system according to an example of the present invention; Figure 4 This is a schematic diagram of a pile foundation porosity detection process according to an example of the present invention; Figure 5 This is a structural block diagram of an electronic device according to the present invention. Detailed Implementation

[0042] Existing technologies generally employ the following methods for pile foundation temperature measurement. For example, a distributed temperature-measuring optical cable and its application in the accurate evaluation of the integrity of cast-in-place piles include: deploying the temperature-measuring optical cable along the reinforcing cage of the cast-in-place pile; using the optical cable to perform refined temperature measurements of the cast-in-place pile at different depths before and after pouring; correcting the measured temperatures to obtain temperature data at different depths of the cast-in-place pile; converting the temperature data at the corresponding depth into an equivalent concrete cover thickness, and combining this with the radius of the reinforcing cage to obtain the equivalent radius of the cast-in-place pile at the current depth; using spline interpolation based on the equivalent radius to draw the cross-section corresponding to the cast-in-place pile at the current depth; constructing the 3D shape of the cast-in-place pile from the cross-sections at different depths for visualization; through refined temperature measurement, the spatial resolution of temperature measurement is effectively improved, not only reducing or even avoiding missed detections of defects and errors in severity assessment of cast-in-place piles, but also achieving visual transparency of the cast-in-place pile. The overall operation is simple, convenient, intuitive, and accurate, with broad application prospects. For example, a device and method for detecting the integrity of cast-in-place piles based on fiber Bragg grating temperature measurement is provided. The device includes a cast-in-place pile and a reinforcing cage. A capillary seamless steel tube fiber Bragg grating temperature sensor and an actively heated fiber Bragg grating temperature sensor are fixed at equal intervals along the axial direction of the cast-in-place pile on the reinforcing cage, and both are connected to a signal demodulator. The actively heated fiber Bragg grating temperature sensor is then connected to a pulse heating power supply. The signal demodulator is connected to an IoT gateway module, which is then connected to a data visualization processing module. The signal demodulator, IoT gateway module, and pulse heating power supply are all connected to a DC power supply module via interface power. This method can improve the intelligence and real-time nature of pile foundation detection, while also improving the accuracy of pile foundation detection data, simplifying the detection process, and reducing costs. For example, a cast-in-place pile foundation integrity detection system based on distributed optical fiber temperature sensing has the following structure: the sensing optical fiber is placed inside the cast-in-place pile, and a heating device, a distributed optical fiber temperature measuring instrument, and a cast-in-place pile integrity evaluation device are connected sequentially. The heating device heats the sensing optical fiber, which senses the pile foundation temperature. The distributed optical fiber temperature measuring instrument measures the temperature, and the cast-in-place pile integrity evaluation device converts the temperature value into an integrity value. This scheme can effectively and quickly measure the integrity of cast-in-place pile foundations; it can provide long-term, real-time, and online monitoring of the integrity of cast-in-place pile foundations; the detection is simple and can meet the requirements of various cast-in-place pile foundations; it is suitable for the integrity detection of various cast-in-place pile foundations.

[0043] The above examples demonstrate the following problems with existing methods: (1) Insufficient monitoring data dimensions lead to blind spots in defect identification: Conventional straight-line fiber optic cables can only acquire one-dimensional axial temperature information and cannot sense the temperature distribution of the pile cross-section (circumferential direction). This results in insensitivity to defects such as eccentric voids and circumferential segregation, leading to a high risk of missed detection and making it impossible to achieve true three-dimensional quality diagnosis.

[0044] (2) The chain from data to evaluation is broken, and the process is inefficient and non-quantitative: There is a lack of an automatic and accurate physical model and algorithm bridge between temperature data and the final quality conclusions. It relies on manual interpretation, and the interpretation only stays at the level of temperature anomalies. This results in a slow evaluation process, strong subjectivity, and the inability to quantify temperature anomalies into indicators directly needed for engineering, such as "density" and "porosity". The evaluation results are crude and have weak support for engineering decision-making.

[0045] (3) The technical links are fragmented, and an intelligent closed loop has not been formed: Different technologies may be used in the temperature measurement, inversion, and evaluation stages. The system integration is low, the operation is complex, and the degree of automation and intelligence is insufficient, making it difficult to efficiently and in a standardized manner output quantitative porosity distribution maps that can be directly used for durability analysis.

[0046] To address the aforementioned issues, this invention proposes a novel method for detecting the porosity of pile foundations. Starting from the actual construction process of concrete pile foundations, it utilizes an integrated non-destructive sensing and intelligent analysis approach to efficiently and accurately acquire and analyze the internal three-dimensional temperature field. This allows for the direct inversion of the thermal diffusivity distribution, which is sensitive to pore structure, and ultimately, a quantitative assessment of the three-dimensional spatial variation of porosity. This overcomes the technical bottlenecks of existing methods, which rely on destructive sampling, have indirect evaluation indicators, and cannot achieve large-scale quantitative imaging of three-dimensional pore structures. Consequently, it provides precise microstructural parameters for assessing the quality and durability of pile foundation casting.

[0047] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0048] Example 1 This invention provides a method for detecting the porosity of concrete pile foundations, as illustrated in the schematic diagram below. Figure 1 As shown, the method includes: Step S110: Input the spatiotemporal coordinates of each position of the target pile body into the pre-trained physical information neural network model to invert and obtain the spatial distribution field of the thermal diffusivity coefficient inside the target pile during the hydration process of the cementitious material. Step S120: Based on the pre-established quantitative mapping relationship between thermal diffusivity and porosity, the spatial distribution field of thermal diffusivity is converted into the spatial distribution field of porosity of the target pile foundation; The spatiotemporal coordinates include spatial coordinates and the time after pouring.

[0049] In this exemplary embodiment, the present invention can target the stage after pile foundation pouring and vibration are completed, when external disturbances are minimal, such as the hydration process of the cementitious material after vibration, during which hydration heat is released within the concrete. The physical information neural network model is also trained for this stage. Model training can be performed on a single pile foundation or a pile foundation group composed of multiple micropiles, and the target pile foundation can be the same as or different from the pile foundation measured by spatial temperature field observation data. During the construction of concrete pile foundations, the hydration of cementitious materials releases heat, forming a three-dimensional temperature field that evolves over time. The spatial distribution of this temperature field is closely related to the thermal diffusivity coefficient inside the pile body, which is extremely sensitive to the pore structure. Based on this physical principle, this scheme first acquires three-dimensional spatial temperature distribution data of the target pile foundation during the hydration process of the cementitious material. This data can be collected through pre-deployed temperature-sensing optical fibers, covering the spatial coordinates of each location on the pile body and the corresponding post-pouring time (i.e., the time starting from the measurement start moment after pouring completion). The temperature-sensing optical fibers measure temperature using the distributed optical fiber temperature measurement principle, that is, based on optical time-domain reflectometry, by analyzing the intensity ratio of backscattered Raman light in the optical fiber, to achieve continuous and real-time temperature measurement along the optical fiber path. Physical information neural networks are a novel framework that embeds physical laws into the training process of deep learning models. By using the residuals of physical equations as part of the loss function, the network can both fit the data and obey physical laws after training. In the detection phase, the spatial coordinates of each location on the target pile foundation and the post-pouring time are used as inputs and fed into the pre-trained physical information neural network model. The output of this model simultaneously provides two physical quantities: one is the temperature value predicted based on the input spatiotemporal coordinates, and the other is the predicted value of the thermal diffusivity at that location. Once the spatiotemporal coordinates of the target pile foundation are input point by point, the model inverts the three-dimensional spatial distribution field of the thermal diffusivity coefficient within the entire pile foundation during the hydration process of the cementitious material. The relationship between the thermal diffusivity coefficient and material structure is such that the thermal diffusivity coefficient comprehensively reflects the material's thermal conductivity and heat storage capacity, and is sensitive to changes in the material's microstructure (such as porosity and pore morphology). It is an important thermophysical parameter characterizing the internal structure of porous media (such as concrete). Based on this, a quantitative mapping relationship between the thermal diffusivity coefficient and porosity is established through experimental calibration. Subsequently, using the calibrated quantitative mapping relationship between the thermal diffusivity coefficient and porosity, the thermal diffusivity coefficient at each spatial location is converted into the corresponding porosity value, thus obtaining the three-dimensional spatial distribution field of the porosity of the target pile foundation. This process does not require any destructive sampling of the pile body, nor does it rely on indirect indicators for manual inference. It directly achieves a non-destructive, quantitative, three-dimensional inversion from macroscopic thermal response to microscopic pore structure, overcoming the shortcomings of existing technologies such as slow evaluation processes, low accuracy, and inability to detect over a wide range.

[0050] In some implementations, the spatiotemporal coordinates of various locations within the target pile foundation are input into a pre-trained physical information neural network model to invert and obtain the spatial distribution field of the thermal diffusivity coefficient inside the target pile foundation during the hydration process of the cementitious material, including: The spatiotemporal coordinates of each position of the target pile foundation are input into a pre-trained physical information neural network model. The spatiotemporal coordinates of each position are then subjected to high-dimensional nonlinear mapping encoding by the shared feature encoder to obtain latent space features. The latent space features are filtered by the feature filtering module of the temperature prediction branch to remove feature components that are strongly correlated with the time dimension. The filtered features are then decoded by the inversion decoder to obtain the corresponding predicted value of the thermal diffusivity. Based on the predicted thermal diffusivity values ​​at various locations, the spatial distribution field of thermal diffusivity within the target pile foundation during the hydration process of the cementitious material is obtained.

[0051] In this example implementation, the physical information neural network model includes a shared feature encoder, a temperature prediction branch, and a thermal diffusivity prediction branch, employing a multi-branch shared encoding architecture. Internally, a shared feature encoder is first set up. This encoder performs a high-dimensional nonlinear mapping between the input spatial coordinates and the time after pouring, transforming low-dimensional spatiotemporal coordinates (e.g., three-dimensional coordinates plus one-dimensional time) into high-dimensional feature vectors in the latent space. This mapping can be achieved through multiple fully connected layers combined with nonlinear activation functions, expanding the original coordinate information to a richer feature representation space, facilitating subsequent branch networks to capture complex physical laws. Subsequently, the latent space features are simultaneously fed into two parallel branches: the temperature prediction branch and the thermal diffusivity prediction branch. The temperature prediction branch outputs the predicted temperature value, while the thermal diffusivity prediction branch outputs the thermal diffusivity at various points in space. The predicted temperature value can assess temperature anomaly areas within the pile foundation. Notably, the thermal diffusivity prediction branch includes a feature filtering module (e.g., an attention mask). This module filters out components in the latent space features that are strongly correlated with the time dimension, ensuring that the thermal diffusivity theoretically only changes with spatial location and should not change with time. Through feature filtering, the model eliminates feature channels that fluctuate significantly over time. The filtered features are then decoded by an inversion decoder (several fully connected layers) to obtain the predicted thermal diffusivity coefficients for the corresponding spatial coordinates. The predicted values ​​at all locations together constitute the spatial distribution field of the thermal diffusivity coefficients within the target pile foundation. This branching structure, combined with feature filtering, effectively prevents the model from erroneously attributing time variations to changes in the thermal diffusivity coefficient, thus improving the physical consistency and numerical stability of the inversion.

[0052] In some implementations, the physical information neural network model has two outputs: the predicted temperature value and the predicted thermal diffusivity value at each location of the pile body. It is obtained by training the model using spatial temperature field observation data and a loss function with embedded thermal conduction physical constraints. The spatial temperature field observation data refers to the three-dimensional spatial temperature distribution data of the pile foundation during the hydration process of the cementitious material; the thermal conduction physical constraint is used to ensure that the temperature field and thermal diffusivity coefficient field output by the model satisfy the thermal conduction control equation within the preset solution domain.

[0053] In this example implementation, the physical information neural network model outputs both predicted temperature and thermal diffusivity at various locations on the pile. It samples the three-dimensional spatial temperature distribution data of the pile foundation during the cementitious material hydration process and uses this data as the label for the temperature branch. The error between the actual temperature value and the model's predicted temperature value is used as the data fitting loss. Simultaneously, during the training phase, the model forces its output temperature field and thermal diffusivity field to satisfy the heat conduction control equation within a preset solution domain. By embedding the physical constraints of heat conduction into the loss function, the inversion process is no longer a purely data-driven black-box fitting, but an intelligent solution constrained by real thermophysical laws. This invention, through a physical information neural network model with predicted temperature and thermal diffusivity as dual outputs and using a loss function embedded with physical constraints of heat conduction for model training, forces the model's output temperature field and thermal diffusivity field to satisfy the heat conduction control equation. This ensures the inversion process is constrained by real physical laws, significantly reducing sensitivity to observational noise and the ambiguity of the inverse problem. It can more accurately calculate the thermal diffusivity distribution, which is sensitive to pore structure, from the macroscopic temperature response, and then obtain a highly reliable porosity value through quantitative mapping.

[0054] In some implementations, the training process of the physical information neural network model includes: The spatial temperature field observation data is obtained by temperature sensing optical fibers spirally deployed along the test pile foundation, and the spatiotemporal coordinates of each observation point and the spatial temperature field observation data are used as a training sample set. A fusion loss function is constructed based on data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The physical information neural network model is trained based on the training sample set and the fusion loss function; The data fitting loss is used to ensure that the temperature prediction value output by the model is consistent with the observed data of the space temperature field; the prior knowledge loss is used to ensure that the output thermal diffusivity coefficient conforms to the actual prior conditions.

[0055] In this example implementation, a temperature-sensing optical fiber is deployed along a spiral path along the test pile foundation (or representative sample pile foundation). Based on the principle of optical time-domain reflectometry and Raman scattering, this fiber can continuously and in real-time collect temperature data at various points on the pile body throughout the entire hydration process of the cementitious material, forming a three-dimensional observation dataset containing spatial coordinates, timestamps, and temperature values. The spatiotemporal coordinates and temperature observation data of each observation point are used as the training sample set. During training, a fusion loss function with three constraints is constructed. First, the data fitting loss, used to ensure that the temperature prediction value output by the model is as consistent as possible with the measured temperature observation data, can be determined by calculating the mean square error between the predicted and observed values. Second, the thermal conduction physical constraint loss involves randomly and densely sampling a large number of configuration points within the solution domain. For each configuration point, the model outputs a temperature prediction value and a thermal diffusivity prediction value. Then, it is calculated whether the temperature prediction value and the thermal diffusivity prediction value satisfy the transient thermal conduction control equation, i.e., whether the rate of change of temperature with respect to time is equal to the sum of the divergence of the thermal diffusivity gradient and the temperature gradient plus the hydration heat source term divided by the heat capacity. The sum of squares of the equation residuals is the physical constraint loss, used to force the model to learn a solution that conforms to the law of thermal conductivity. Third, the prior knowledge loss can include several sub-items, such as the thermal diffusivity coefficient should vary smoothly in space (avoiding unreasonable and drastic fluctuations), the thermal diffusivity coefficient should be within a reasonable range for concrete materials (e.g., upper and lower limits determined experimentally), and the pile foundation surface should satisfy convective heat transfer boundary conditions. During training, the training sample set is input into the model, forward propagation obtains the predicted values, the fusion loss function value is calculated, and then the network parameters of the model are iteratively updated through backpropagation and gradient descent algorithms until the loss converges. After sufficient training, the model can not only fit the observed temperature data, but more importantly, it can deduce the distribution of the thermal diffusivity coefficient that conforms to physical laws from the temperature field.

[0056] For example, the spiral fiber optic sensor network deployment in this example is as follows: Figure 2 As shown, corrosion-resistant temperature sensing optical cables are wound longitudinally and helically around the outside of the micropile's reinforcing steel skeleton. This allows the measurement path of a single optical fiber to simultaneously cover both the axial and circumferential directions, forming a three-dimensional sensing network covering the entire surface of the pile. This enables the acquisition of dense temperature data across the entire pile's cylindrical surface using a limited number of sensing units. The fiber optic pitch is related to the pile diameter; for example, the pitch of the helical layout is 0.2 to 0.5 times the target pile diameter, with increased pitch density in the pile tip region. In some examples, the pitch can be dynamically adjusted according to the pile foundation bearing capacity design and spatial resolution requirements. Smaller, denser pitches can be used in critical areas such as the pile tip, with enhanced monitoring sections (e.g., denser stirrups or auxiliary helical supports).

[0057] In some examples, such as Figure 3As shown, by using fiber optic sensors wound around the outside of the micropile's reinforcing steel skeleton and employing distributed fiber optic sensing technology, a temperature demodulator continuously collects three-dimensional temperature data of the entire cementitious material hydration process with high spatiotemporal resolution throughout. Based on the linear distributed temperature data collected along the fiber optic path, a temperature field data processing unit uses a spatial interpolation algorithm to reconstruct the three-dimensional temperature field inside the pile body and its four-dimensional evolution over time, resulting in four-dimensional temperature field data. By comparing this four-dimensional temperature field data with theoretical models or adjacent intact areas through a defect analysis and quality evaluation unit, three-dimensional areas of temperature anomalies can be intuitively identified.

[0058] In some implementations, training the physical information neural network model based on the training sample set and the fusion loss function includes: The spatiotemporal coordinates of each observation point in the training sample set are input into the physical information neural network model, and the temperature prediction branch and the thermal diffusivity prediction branch are respectively passed to obtain the temperature prediction value and the thermal diffusivity prediction value of each observation point. Based on the predicted temperature values ​​at each observation point and the observed spatial temperature field data, the data fitting loss is calculated. Dense sampling is performed within the solution domain to obtain the spatiotemporal coordinates of multiple sampling points. The temperature prediction and thermal diffusivity prediction based on the spatiotemporal coordinates of each sampling point are used to calculate the residual of the heat conduction control equation within the solution domain, so as to obtain the physical constraint loss of heat conduction. Based on the gradient of the predicted thermal diffusivity coefficient of each sampling point relative to the spatial coordinates, the upper and lower limits of the thermal diffusivity coefficient, and the boundary heat exchange law of the pile foundation, the prior knowledge loss is calculated. The data fitting loss, heat conduction physical constraint loss, and prior knowledge loss are weighted and fused to obtain the total loss value of the training process. With the goal of minimizing the total loss, the network parameters of the physical information neural network model are updated using the gradient descent algorithm until the model converges.

[0059] In this example implementation, the physical information neural network model includes a structure with two parallel output branches: a temperature prediction branch and a thermal diffusivity prediction branch. The temperature prediction branch outputs the predicted temperature value, and the thermal diffusivity prediction branch outputs the predicted thermal diffusivity value. Spatial temperature field observation data of the test pile foundation (which can be a similar pile to the target pile foundation or the target pile foundation itself) are collected via a spiral optical fiber. Each observation point is accompanied by its spatial coordinates and timestamp. The spatiotemporal coordinates of these observation points are used as input, and the corresponding measured temperature observation data are used as supervision signals. During training, the spatiotemporal coordinates of the training sample set are first input into the model. After passing through a shared feature encoder, they are fed into the temperature prediction branch and the thermal diffusivity prediction branch respectively, obtaining the predicted temperature value and the predicted thermal diffusivity value for each observation point. Then, the data fitting loss is calculated: the predicted temperature value is compared with the measured temperature value, and the mean square error is used to ensure that the model can accurately reproduce the observed thermal response. Simultaneously, dense sampling is performed within the solution domain (including the internal space of the pile body and the time interval after pouring), generating a large number of sampling points / configuration points that are not limited to the location of the observation points. Each sampling point / configuration point has its own spatiotemporal coordinates. These sampling points / configuration points are input into the model, which outputs corresponding predicted temperature and thermal diffusivity values. Using these predicted values, the residuals of the heat conduction governing equation are calculated: the partial derivative of temperature with respect to time minus the combination of the thermal diffusivity gradient and the temperature gradient divergence, minus the square of the hydration heat source term divided by the heat capacity. The average of the residuals across all configuration points yields the physical constraint loss for heat conduction. This loss term embeds the mandatory requirements of the physical equations into the training process, ensuring the model adheres to the laws of heat conduction even in regions without measured data. Furthermore, prior knowledge loss is calculated based on the predicted thermal diffusivity values ​​at each configuration point, including: the squared L2 norm of the thermal diffusivity spatial gradient (encouraging smoothness), the penalty for thermal diffusivity exceeding physical upper and lower limits (implemented via the ReLU function), and the deviation of convective heat transfer conditions at the boundary (by substituting the predicted temperature and thermal diffusivity at the boundary points into the third type of boundary condition formula and calculating the squared difference between the two sides of the equation). The data fitting loss, the physical constraint loss for heat conduction, and the prior knowledge loss are weighted and summed according to preset weights to obtain the total loss. With the goal of minimizing the total loss, a gradient descent algorithm (such as the Adam optimizer) is used to iteratively update all network parameters of the model until the loss value no longer decreases significantly. The finally converged model can then be used for the inversion of the thermal diffusivity of the target pile foundation.

[0060] For example, a thermodynamic inverse problem-solving framework is constructed, taking the temperature field as input and using an optimization algorithm to continuously adjust the thermal diffusivity in the three-dimensional computational model, minimizing the error between the temperature field calculated by the model and the measured field. Ultimately, the distribution of the thermal diffusivity inside the cementitious material is inverted, and then mapped to the porosity quantitative assessment result.

[0061] The transient heat conduction during concrete hydration follows the following partial differential equation:

[0062] in: Let x be the temperature field, which is a function of spatial position x and time t. Density (kg / m³); Specific heat capacity (J / (kg·K)); Let be the thermal diffusivity (m² / s), representing the rate at which heat propagates through a material, and be a function of the spatial coordinate x. For non-uniform concrete, As the location changes, it directly reflects the differences in internal structure such as porosity and density; The heat released during concrete hydration (W / m³) is typically represented by an exponential decay model. ,in, For the final release of heat, This is the heat release rate parameter.

[0063] The boundary and initial conditions are as follows: Initial conditions: (Ambient temperature or mold temperature).

[0064] Boundary conditions: Third type of boundary conditions (Robin conditions): , The surface heat transfer coefficient, This represents the temperature gradient along the direction normal to the heat transfer boundary.

[0065] The final physical information neural network model is as follows: Input is spatial coordinates The output, along with the time t after pouring, is the predicted temperature field. and predicted thermal diffusivity distribution .

[0066]

[0067] in, For physical information neural network models, For network parameters, The temperature field predicted by the neural network, This is the thermal diffusivity field predicted by the neural network.

[0068] The loss function consists of three parts: data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The total loss function / fusion loss function is defined as follows:

[0069] in, For data fitting loss, These are the weighting coefficients for the data fitting loss. For heat conduction physical constraint loss, The weighting coefficient for the physical constraint loss of heat conduction. For prior knowledge loss, These are the weighting coefficients for the prior knowledge loss. The total loss value is calculated using the total loss function.

[0070] (1) Data fitting loss :

[0071] in, For data fitting loss, This represents the total number of observation points in the temperature observation data. Spatial location ,time The temperature prediction value of the model is below. Let i be the spatial location of the i-th sensor. Spatial location and time The network uses temperature observation data. Constraints are applied only to a limited, sparse number of sensor locations to ensure that the network's predicted temperature at the actual sensor locations matches the actual measured value.

[0072] (2) Physical constraint loss of heat conduction: The forced prediction of the temperature field and thermal diffusivity distribution satisfies the heat conduction equation, and random sampling is performed within the solution domain. By configuring a network with the coordinates of several points, the predicted temperature and thermal diffusivity are obtained. The details are as follows:

[0073] in, For heat conduction physical constraint loss, This represents the total number of sampling points in dense sampling. For concrete density, The specific heat capacity of concrete, Let represent the derivative of the predicted temperature value at the j-th sampling point with respect to time. This indicates the calculation of divergence. This is the predicted value of the thermal diffusivity at the j-th sampling point. The predicted temperature value for the j-th sampling point. Let be the intensity of the concrete hydration heat source at the j-th sampling point.

[0074] (3) Prior constraint loss term

[0075] Introducing constraints on the rationality of the thermal diffusivity distribution:

[0076] in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, The loss includes boundary condition constraint loss and smoothness constraint loss, used to avoid drastic changes in the parameter field that do not conform to physical laws. The range constraint loss is used to limit parameter values ​​to a reasonable physical range (the thermal diffusivity should conform to the range of thermal diffusivity in intact or defective concrete tested in experiments). The boundary condition constraint loss is used to force known boundary conditions (such as adiabatic, convection, soil ambient temperature, etc.) to be satisfied at the boundary configuration points obtained from pile-soil boundary sampling; it forces the network to satisfy the third type of thermal boundary condition (convective heat transfer condition) at the boundary, ensuring that the inversion results satisfy the heat exchange law overall. The losses are expressed as follows:

[0077]

[0078]

[0079] in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, For boundary condition constraint loss, This represents the total number of sampling points in dense sampling. This represents the total number of sampling points at the pile-soil interface. Let the thermal diffusivity be the coefficient at the j-th sampling point. Spatial gradient at that location To correct the linear unit, This is the predicted value of the thermal diffusivity at the j-th sampling point. , These are the upper and lower limits of the thermal diffusivity, respectively. The m-th sampling point at the pile-soil interface Predicted value of thermal diffusivity at that location. This represents the derivative of the predicted temperature value with respect to the direction outside the pile-soil boundary. The convective heat transfer coefficient (unit: W / (m²·K)) at the pile-soil interface describes the strength of heat exchange between the concrete surface and the surrounding soil, and can be estimated based on geological conditions. For concrete density, The specific heat capacity of concrete, Boundary sampling points Predicted temperature value at time t. Boundary sampling points The ambient temperature at the location, i.e., the soil temperature far from the pile or the known temperature boundary.

[0080] In some implementations, a thermal conductivity distribution is obtained. Subsequently, based on the effective medium theory, concrete is considered as a two-phase composite material consisting of a continuous solid matrix (cement, aggregate, etc.) and a randomly distributed porous phase (usually air or water), with an equivalent thermal diffusivity. With porosity The relationship satisfies the following general formula:

[0081] in: The effective thermal diffusivity of concrete is the effective thermal diffusivity inside the target pile foundation (i.e., the predicted value of thermal diffusivity obtained by physical information neural network inversion), with the unit: m² / s. The thermal diffusivity of concrete solid represents the thermal diffusivity of the pure solid skeleton after removing all pores. The unit is m² / s; it can be measured experimentally, and its value ranges from 1.2 to 1.7. This represents the thermal diffusivity of the fluid in the pores, specifically the thermal diffusivity of water taken from the concrete during the initial pouring stage. ( ; This represents porosity, with a value ranging from 0 to 1. This is a shape factor, related to the pore geometry and spatial distribution. For spherical pores, the theoretical value is n=2; for non-spherical pores, ... It can be determined through experimental calibration.

[0082] Therefore, the quantitative mapping relationship between thermal diffusivity and porosity is as follows:

[0083] The target pile obtained by inversion By substituting these values ​​into the pre-calibrated quantitative mapping relationship, the porosity value corresponding to each spatial location can be calculated. This forms a three-dimensional porosity distribution field.

[0084] In some implementations, after obtaining the spatial distribution field of porosity of the target pile foundation, the method further includes: Based on the spatial distribution field of porosity of the target pile foundation, a three-dimensional image segmentation algorithm is used to identify high-risk connected regions with porosity greater than a preset threshold.

[0085] In this example implementation, after obtaining the three-dimensional spatial distribution field of porosity of the target pile foundation, a three-dimensional image segmentation algorithm can be introduced to automatically identify "high-risk connected regions" where the porosity exceeds a preset threshold. A porosity threshold can be set according to engineering design requirements. This threshold is typically set at 15% or determined experimentally based on specific engineering requirements. When the porosity at a certain spatial location... When the location is identified as a high-risk area, a three-dimensional connected component labeling algorithm (such as a 26-neighborhood-based 3D flood filling or disjoint-set data structure) is used to scan the entire porosity field, grouping adjacent locations with porosities exceeding the threshold into a single connected region. Each connected region corresponds to a continuous pore-rich zone within the pile foundation, potentially representing a defect caused by insufficient compaction, segregation, voids, or other reasons. For each identified high-risk connected region, its geometric characteristics are further calculated, including: region volume (i.e., the number of voxels multiplied by the actual volume represented by each voxel), region centroid location (three-dimensional coordinates), maximum porosity value, and average porosity value within the region. Based on these quantitative indicators, the defect can be classified into risk levels (e.g., low risk, medium risk, high risk), and its impact on the pile foundation's bearing capacity and durability can be estimated. For example, for each risk region, its volume, location, maximum, and average porosity are calculated. It can further classify risk levels based on porosity range (e.g., low risk, medium risk, high risk) and automatically generate a comprehensive test report containing a 3D porosity color cloud map, risk area annotation map, detailed statistical tables, and classification assessment conclusions (e.g., using color depth to indicate porosity levels). It automatically outputs a comprehensive report including a 3D porosity cloud map, defect statistics table, and classification conclusions. This report allows engineering technicians to intuitively understand the spatial distribution characteristics of porosity within the pile foundation, accurately locate defects requiring treatment, and provide precise spatial basis for subsequent grouting reinforcement or other strengthening measures.

[0086] This invention presents a complete technical solution: three-dimensional temperature measurement using spiral optical fibers, direct inversion of the thermal diffusivity field using PINN (Physical Information Neural Network) model, quantitative conversion of thermal diffusivity to porosity, and three-dimensional visualization of porosity. First, a temperature field rich in spatial information is acquired through a spiral distributed optical fiber sensor network. Then, a Physical Information Neural Network, specifically designed for thermal diffusivity inversion, directly and efficiently solves for the three-dimensional distribution of the thermal diffusivity from the temperature field. Next, based on porous media thermal theory, the thermal diffusivity field is converted into a porosity field using a calibrated quantitative relationship model of thermal diffusivity to porosity. Finally, spatial statistics and risk analysis are performed on the porosity field to generate intuitive assessment results.

[0087] For example, such as Figure 4 As shown, the porosity detection method for concrete pile foundations in this example includes the following steps: First, micropiles were constructed and fiber optic cables were laid, followed by the pouring of cementitious materials and the initiation of monitoring, entering the real-time monitoring phase of four-dimensional temperature field data. Simultaneously, a Physical Information Neural Network (PINN) was constructed. The collected temperature field data was preprocessed to complete the three-dimensional temperature field reconstruction.

[0088] The reconstructed temperature field data is input into a physical information neural network. The network structure includes an input layer (spatial coordinates x, time t), a hidden layer (multi-layer fully connected network), and a dual output layer (output 1 is the predicted temperature field, and output 2 is the predicted thermal diffusivity distribution). A loss function is calculated based on the network outputs. This loss function includes data loss (data fitting loss), physical constraint loss (thermal conduction physical constraint loss), and prior constraint loss (prior knowledge loss), and the total loss is obtained by weighting these three factors. Backpropagation and parameter updates are performed based on the total loss to optimize the network weights. The above training process is repeated to determine convergence. If convergence fails, iteration continues; if convergence occurs, the model is applied to the target pile foundation to output the corresponding thermal diffusivity distribution.

[0089] Next, the defect mapping and quantification stage begins. First, the porosity index is calculated, followed by defect region identification and severity classification, generating a 3D defect distribution map and assessment report. Based on the assessment results, a quality evaluation is conducted: if quality requirements are not met, an early warning is activated and engineering measures are recommended; if quality requirements are met, the quality is deemed acceptable and the data is archived. Finally, the quality assessment is completed.

[0090] This invention, through optimized sensor deployment, directly acquires high-quality temperature field data reflecting the three-dimensional spatial thermal state of the pile, providing a sufficient three-dimensional data foundation for porosity inversion. It achieves three-dimensional, quantitative, and non-destructive imaging of the internal pore structure of the pile foundation. From three-dimensional temperature measurement to three-dimensional thermal diffusivity coefficient field inversion, and then to three-dimensional porosity field calculation, the entire process maintains three-dimensional spatial attributes, ultimately outputting an intuitive three-dimensional porosity distribution map. This achieves non-destructive spatial visualization of the porosity within the concrete structure, providing a macroscopic means for microstructural analysis. A direct, accurate, and efficient inversion channel from the temperature field to the porosity field is established. A dedicated intelligent inversion model is designed, directly using the thermal diffusivity coefficient as the inversion target, and establishing a quantitative mapping model with a strong physical correlation between the thermal diffusivity coefficient and porosity. A direct and reliable physical channel from macroscopic thermal response to microscopic pore parameters is established. By using the thermal diffusivity coefficient, which is more sensitive to pore structure, as the direct inversion target, and applying the α (thermal diffusivity coefficient)-φ (porosity) physical model, errors and multiple solutions caused by indirect inference based on thermal conductivity are avoided. The test results have clear physical meaning, high accuracy, and strong reliability. An automated and quantitative three-dimensional porosity detection and evaluation system has been constructed, forming a complete closed loop from data acquisition and intelligent processing to result output. It can output a three-dimensional distribution cloud map of the porosity inside the pile body and a quantitative statistical report with a single click. This system forms a highly integrated and automated dedicated detection closed loop, combining efficiency and practicality. It seamlessly integrates directional optimized spiral temperature measurement, a dedicated PINN inversion algorithm, and a calibration physical model into a single system. From on-site data acquisition to the generation of porosity evaluation reports, the entire process is automated and intelligent, greatly improving detection efficiency, lowering the technical threshold, and making it possible to conduct rapid, large-scale porosity screening and preliminary durability evaluation of pile foundations, demonstrating significant engineering practical value.

[0091] Example 2 Based on the same inventive concept, the present invention also provides a porosity detection system for concrete pile foundations, comprising: The inversion module is used to input the spatiotemporal coordinates of each position of the target pile foundation into a pre-trained physical information neural network model to invert and obtain the spatial distribution field of the thermal diffusivity coefficient inside the target pile foundation during the hydration process of the cementitious material. The conversion module is used to convert the spatial distribution field of the thermal diffusivity coefficient into the spatial distribution field of the porosity of the target pile foundation based on the pre-established quantitative mapping relationship between the thermal diffusivity coefficient and porosity. The spatiotemporal coordinates include spatial coordinates and the time after pouring.

[0092] In one possible implementation, the physical information neural network model includes a shared feature encoder and a thermal diffusivity prediction branch; the inversion module includes: The encoding submodule is used to input the spatiotemporal coordinates of each position of the target pile foundation into the pre-trained physical information neural network model. The spatiotemporal coordinates of each position are encoded by the shared feature encoder through high-dimensional nonlinear mapping to obtain latent space features. The feature filtering submodule is used to filter the latent space features through the feature filtering module of the thermal diffusivity prediction branch to remove feature components that are strongly correlated with the time dimension. The decoding submodule is used to decode the filtered features through the inversion decoder to obtain the corresponding predicted value of thermal diffusivity; based on the predicted value of thermal diffusivity at each location, the spatial distribution field of thermal diffusivity inside the target pile foundation during the hydration process of cementitious materials is obtained.

[0093] In one possible implementation, the physical information neural network model has two outputs: the predicted temperature value and the predicted thermal diffusivity value at each location of the pile body. It is obtained by training the model using spatial temperature field observation data and a loss function with embedded thermal conduction physical constraints. The spatial temperature field observation data refers to the three-dimensional spatial temperature distribution data of the pile foundation during the hydration process of the cementitious material; the thermal conduction physical constraint is used to ensure that the temperature field and thermal diffusivity coefficient field output by the model satisfy the thermal conduction control equation within the preset solution domain.

[0094] In one possible implementation, a training module is also included, the training module comprising: The sample acquisition submodule is used to acquire the spatial temperature field observation data through temperature sensing optical fibers spirally deployed along the test pile foundation, and use the spatiotemporal coordinates of each observation point and the spatial temperature field observation data as a training sample set. The loss construction submodule is used to construct a fusion loss function based on data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The training submodule is used to train the physical information neural network model based on the training sample set and the fusion loss function; The data fitting loss is used to ensure that the temperature prediction value output by the model is consistent with the observed data of the space temperature field; the prior knowledge loss is used to ensure that the output thermal diffusivity coefficient conforms to the actual prior conditions.

[0095] In one possible implementation, the pitch of the spiral arrangement is 0.2 to 0.5 times the diameter of the target pile foundation, and the pitch is increased in the pile end area.

[0096] In one possible implementation, the physical information neural network model includes a temperature prediction branch and a thermal diffusivity prediction branch; the training submodule is specifically used for: The spatiotemporal coordinates of each observation point in the training sample set are input into the physical information neural network model, and the temperature prediction branch and the thermal diffusivity prediction branch are respectively passed to obtain the temperature prediction value and the thermal diffusivity prediction value of each observation point. Based on the predicted temperature values ​​at each observation point and the observed spatial temperature field data, the data fitting loss is calculated. Dense sampling is performed within the solution domain to obtain the spatiotemporal coordinates of multiple sampling points. The temperature prediction and thermal diffusivity prediction based on the spatiotemporal coordinates of each sampling point are used to calculate the residual of the heat conduction control equation within the solution domain, so as to obtain the physical constraint loss of heat conduction. Based on the gradient of the predicted thermal diffusivity coefficient of each sampling point relative to the spatial coordinates, the upper and lower limits of the thermal diffusivity coefficient, and the boundary heat exchange law of the pile foundation, the prior knowledge loss is calculated. The data fitting loss, heat conduction physical constraint loss, and prior knowledge loss are weighted and fused to obtain the total loss value of the training process. With the goal of minimizing the total loss, the network parameters of the physical information neural network model are updated using the gradient descent algorithm until the model converges.

[0097] In one possible implementation, the physical constraint loss for thermal conduction is:

[0098] in, For heat conduction physical constraint loss, This represents the total number of sampling points in dense sampling. For concrete density, The specific heat capacity of concrete, Let represent the derivative of the predicted temperature value at the j-th sampling point with respect to time. This indicates the calculation of divergence. This is the predicted value of the thermal diffusivity at the j-th sampling point. The predicted temperature value for the j-th sampling point. Let be the intensity of the concrete hydration heat source at the j-th sampling point.

[0099] In one possible implementation, the prior knowledge loss is:

[0100]

[0101]

[0102]

[0103] in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, For boundary condition constraint loss, This represents the total number of sampling points in dense sampling. This represents the total number of sampling points at the pile-soil interface. Let the thermal diffusivity be the coefficient at the j-th sampling point. Spatial gradient at that location To correct the linear unit, This is the predicted value of the thermal diffusivity at the j-th sampling point. , These are the upper and lower limits of the thermal diffusivity, respectively. The m-th sampling point at the pile-soil interface Predicted value of thermal diffusivity at that location. This represents the derivative of the predicted temperature value with respect to the direction outside the pile-soil boundary. Let be the convective heat transfer coefficient at the pile-soil interface. For concrete density, The specific heat capacity of concrete, Boundary sampling points Predicted temperature value at time t. Boundary sampling points The ambient temperature at that location.

[0104] In one possible implementation, the quantitative mapping relationship between the thermal diffusivity and porosity is established based on the effective medium theory, and its expression is:

[0105] in, Porosity The effective thermal diffusivity inside the target pile foundation, i.e., the predicted value of the thermal diffusivity. The thermal diffusivity of the concrete solid is measured experimentally. The shape factor was determined through experimental calibration. is the thermal diffusivity of the fluid in the pores.

[0106] In one possible implementation, it also includes: The risk identification module is used to identify high-risk connected regions with porosity greater than a preset threshold based on the spatial distribution field of porosity of the target pile foundation using a three-dimensional image segmentation algorithm.

[0107] Example 3 like Figure 5As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0108] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of the porosity detection method for concrete pile foundation in the above embodiments.

[0109] Example 4 Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). An electronic device readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both the built-in storage medium of the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the method for detecting the porosity of a concrete pile foundation described in the above embodiments.

[0110] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0111] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the claims pending approval.

Claims

1. A method for detecting the porosity of concrete pile foundations, characterized in that, include: The spatiotemporal coordinates of each position of the target pile foundation are input into a pre-trained physical information neural network model to invert the spatial distribution field of the thermal diffusivity coefficient inside the target pile foundation during the hydration process of the cementitious material. Based on the pre-established quantitative mapping relationship between thermal diffusivity and porosity, the spatial distribution field of thermal diffusivity is converted into the spatial distribution field of porosity of the target pile foundation; The spatiotemporal coordinates include spatial coordinates and the time after pouring.

2. The method according to claim 1, characterized in that, The physical information neural network model includes a shared feature encoder and a thermal diffusivity prediction branch. The spatiotemporal coordinates of each location within the target pile foundation are input into the pre-trained physical information neural network model to invert the spatial distribution field of the thermal diffusivity coefficient within the target pile foundation during the cementitious material hydration process, including: The spatiotemporal coordinates of each position of the target pile foundation are input into a pre-trained physical information neural network model. The spatiotemporal coordinates of each position are then subjected to high-dimensional nonlinear mapping encoding by the shared feature encoder to obtain latent space features. The latent space features are filtered by the feature filtering module of the thermal diffusivity prediction branch to remove feature components that are strongly correlated with the time dimension. The filtered features are decoded by an inversion decoder to obtain the corresponding predicted values ​​of thermal diffusivity. Based on the predicted values ​​of thermal diffusivity at each location, the spatial distribution field of thermal diffusivity inside the target pile foundation during the hydration process of the cementitious material is obtained.

3. The method according to claim 1, characterized in that, The physical information neural network model has two outputs: the predicted temperature value and the predicted thermal diffusivity value at each location of the pile body. It is obtained by training the model using spatial temperature field observation data and a loss function with embedded thermal conduction physical constraints. The spatial temperature field observation data refers to the three-dimensional spatial temperature distribution data of the pile foundation during the hydration process of the cementitious material; the thermal conduction physical constraint is used to ensure that the temperature field and thermal diffusivity coefficient field output by the model satisfy the thermal conduction control equation within the preset solution domain.

4. The method according to claim 3, characterized in that, The training process of the physical information neural network model includes: The spatial temperature field observation data is obtained by temperature sensing optical fibers spirally deployed along the test pile foundation, and the spatiotemporal coordinates of each observation point and the spatial temperature field observation data are used as a training sample set. A fusion loss function is constructed based on data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The physical information neural network model is trained based on the training sample set and the fusion loss function; The data fitting loss is used to ensure that the temperature prediction value output by the model is consistent with the observed data of the space temperature field; the prior knowledge loss is used to ensure that the output thermal diffusivity coefficient conforms to the actual prior conditions.

5. The method according to claim 4, characterized in that, The pitch of the spiral arrangement is 0.2 to 0.5 times the diameter of the target pile foundation, and the pitch is increased in the pile end area.

6. The method according to claim 4, characterized in that, The physical information neural network model includes a temperature prediction branch and a thermal diffusivity prediction branch; Training the physical information neural network model based on the training sample set and the fusion loss function includes: The spatiotemporal coordinates of each observation point in the training sample set are input into the physical information neural network model, and the temperature prediction branch and the thermal diffusivity prediction branch are respectively passed to obtain the temperature prediction value and the thermal diffusivity prediction value of each observation point. Based on the predicted temperature values ​​at each observation point and the observed spatial temperature field data, the data fitting loss is calculated. Dense sampling is performed within the solution domain to obtain the spatiotemporal coordinates of multiple sampling points. The temperature prediction and thermal diffusivity prediction based on the spatiotemporal coordinates of each sampling point are used to calculate the residual of the heat conduction control equation within the solution domain, so as to obtain the physical constraint loss of heat conduction. Based on the gradient of the predicted thermal diffusivity coefficient of each sampling point relative to the spatial coordinates, the upper and lower limits of the thermal diffusivity coefficient, and the boundary heat exchange law of the pile foundation, the prior knowledge loss is calculated. The data fitting loss, heat conduction physical constraint loss, and prior knowledge loss are weighted and fused to obtain the total loss value of the training process. With the goal of minimizing the total loss, the network parameters of the physical information neural network model are updated using the gradient descent algorithm until the model converges.

7. The method according to claim 6, characterized in that, The physical constraint loss for heat conduction is: in, For heat conduction physical constraint loss, This represents the total number of sampling points in dense sampling. For concrete density, The specific heat capacity of concrete, Let represent the derivative of the predicted temperature value at the j-th sampling point with respect to time. This indicates the calculation of divergence. This is the predicted value of the thermal diffusivity at the j-th sampling point. The predicted temperature value for the j-th sampling point. Let be the intensity of the concrete hydration heat source at the j-th sampling point.

8. The method according to claim 6, characterized in that, The prior knowledge loss is: in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, For boundary condition constraint loss, This represents the total number of sampling points in dense sampling. This represents the total number of sampling points at the pile-soil interface. Let the thermal diffusivity be the coefficient at the j-th sampling point. Spatial gradient at that location To correct the linear unit, This is the predicted value of the thermal diffusivity at the j-th sampling point. , These are the upper and lower limits of the thermal diffusivity, respectively. The m-th sampling point at the pile-soil interface Predicted value of thermal diffusivity at that location. This represents the derivative of the predicted temperature value with respect to the direction outside the pile-soil boundary. Let be the convective heat transfer coefficient at the pile-soil interface. For concrete density, The specific heat capacity of concrete, Boundary sampling points Predicted temperature value at time t. Boundary sampling points The ambient temperature at that location.

9. The method according to claim 1, characterized in that, The quantitative mapping relationship between the thermal diffusivity and porosity is established based on the effective medium theory, and its expression is: in, Porosity The effective thermal diffusivity inside the target pile foundation, i.e., the predicted value of the thermal diffusivity. is the thermal diffusivity of the concrete solid; For shape factor; is the thermal diffusivity of the fluid in the pores.

10. The method according to claim 1, characterized in that, After obtaining the spatial distribution field of porosity of the target pile foundation, the following is also included: Based on the spatial distribution field of porosity of the target pile foundation, a three-dimensional image segmentation algorithm is used to identify high-risk connected regions with porosity greater than a preset threshold.

11. A porosity detection system for concrete pile foundations, characterized in that, include: The inversion module is used to input the spatiotemporal coordinates of each position of the target pile foundation into a pre-trained physical information neural network model to invert and obtain the spatial distribution field of the thermal diffusivity coefficient inside the target pile foundation during the hydration process of the cementitious material. The conversion module is used to convert the spatial distribution field of the thermal diffusivity coefficient into the spatial distribution field of the porosity of the target pile foundation based on the pre-established quantitative mapping relationship between the thermal diffusivity coefficient and porosity. The spatiotemporal coordinates include spatial coordinates and the time after pouring.

12. The system according to claim 11, characterized in that, The physical information neural network model includes a shared feature encoder and a thermal diffusivity prediction branch; The inversion module includes: The encoding submodule is used to input the spatiotemporal coordinates of each position of the target pile foundation into the pre-trained physical information neural network model. The spatiotemporal coordinates of each position are encoded by the shared feature encoder through high-dimensional nonlinear mapping to obtain latent space features. The feature filtering submodule is used to filter the latent space features through the feature filtering module of the thermal diffusivity prediction branch to remove feature components that are strongly correlated with the time dimension. The decoding submodule is used to decode the filtered features through the inversion decoder to obtain the corresponding predicted value of thermal diffusivity; based on the predicted value of thermal diffusivity at each location, the spatial distribution field of thermal diffusivity inside the target pile foundation during the hydration process of cementitious materials is obtained.

13. The system according to claim 11, characterized in that, The physical information neural network model has two outputs: the predicted temperature value and the predicted thermal diffusivity value at each location of the pile body. It is obtained by training the model using spatial temperature field observation data and a loss function with embedded thermal conduction physical constraints. The spatial temperature field observation data refers to the three-dimensional spatial temperature distribution data of the pile foundation during the hydration process of the cementitious material; the thermal conduction physical constraint is used to ensure that the temperature field and thermal diffusivity coefficient field output by the model satisfy the thermal conduction control equation within the preset solution domain.

14. The system according to claim 13, characterized in that, It also includes a training module, which includes: The sample acquisition submodule is used to acquire the spatial temperature field observation data through temperature sensing optical fibers spirally deployed along the test pile foundation, and use the spatiotemporal coordinates of each observation point and the spatial temperature field observation data as a training sample set. The loss construction submodule is used to construct a fusion loss function based on data fitting loss, heat conduction physical constraint loss, and prior knowledge loss. The training submodule is used to train the physical information neural network model based on the training sample set and the fusion loss function; The data fitting loss is used to ensure that the temperature prediction value output by the model is consistent with the observed data of the space temperature field; the prior knowledge loss is used to ensure that the output thermal diffusivity coefficient conforms to the actual prior conditions.

15. The system according to claim 14, characterized in that, The pitch of the spiral arrangement is 0.2 to 0.5 times the diameter of the target pile foundation, and the pitch is increased in the pile end area.

16. The system according to claim 14, characterized in that, The physical information neural network model includes a temperature prediction branch and a thermal diffusivity prediction branch; the training submodule is specifically used for: The spatiotemporal coordinates of each observation point in the training sample set are input into the physical information neural network model, and the temperature prediction branch and the thermal diffusivity prediction branch are respectively passed to obtain the temperature prediction value and the thermal diffusivity prediction value of each observation point. Based on the predicted temperature values ​​at each observation point and the observed spatial temperature field data, the data fitting loss is calculated. Dense sampling is performed within the solution domain to obtain the spatiotemporal coordinates of multiple sampling points. The temperature prediction and thermal diffusivity prediction based on the spatiotemporal coordinates of each sampling point are used to calculate the residual of the heat conduction control equation within the solution domain, so as to obtain the physical constraint loss of heat conduction. Based on the gradient of the predicted thermal diffusivity coefficient of each sampling point relative to the spatial coordinates, the upper and lower limits of the thermal diffusivity coefficient, and the boundary heat exchange law of the pile foundation, the prior knowledge loss is calculated. The data fitting loss, heat conduction physical constraint loss, and prior knowledge loss are weighted and fused to obtain the total loss value of the training process. With the goal of minimizing the total loss, the network parameters of the physical information neural network model are updated using the gradient descent algorithm until the model converges.

17. The system according to claim 16, characterized in that, The physical constraint loss for heat conduction is: in, For heat conduction physical constraint loss, This represents the total number of sampling points in dense sampling. For concrete density, The specific heat capacity of concrete, Let represent the derivative of the predicted temperature value at the j-th sampling point with respect to time. This indicates the calculation of divergence. This is the predicted value of the thermal diffusivity at the j-th sampling point. The predicted temperature value for the j-th sampling point. Let be the intensity of the concrete hydration heat source at the j-th sampling point.

18. The system according to claim 16, characterized in that, The prior knowledge loss is: in, For prior knowledge loss, For smoothness constraint loss, For range-constrained losses, For boundary condition constraint loss, This represents the total number of sampling points in dense sampling. This represents the total number of sampling points at the pile-soil interface. Let the thermal diffusivity be the coefficient at the j-th sampling point. Spatial gradient at that location To correct the linear unit, This is the predicted value of the thermal diffusivity at the j-th sampling point. , These are the upper and lower limits of the thermal diffusivity, respectively. The m-th sampling point at the pile-soil interface Predicted value of thermal diffusivity at that location. This represents the derivative of the predicted temperature value with respect to the direction outside the pile-soil boundary. Let be the convective heat transfer coefficient at the pile-soil interface. For concrete density, The specific heat capacity of concrete, Boundary sampling points Predicted temperature value at time t. Boundary sampling points The ambient temperature at that location.

19. The system according to claim 11, characterized in that, The quantitative mapping relationship between the thermal diffusivity and porosity is established based on the effective medium theory, and its expression is: in, Porosity The effective thermal diffusivity inside the target pile foundation, i.e., the predicted value of the thermal diffusivity. is the thermal diffusivity of the concrete solid; For shape factor; is the thermal diffusivity of the fluid in the pores.

20. The system according to claim 11, characterized in that, Also includes: The risk identification module is used to identify high-risk connected regions with porosity greater than a preset threshold based on the spatial distribution field of porosity of the target pile foundation using a three-dimensional image segmentation algorithm.

21. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the method as described in any one of claims 1 to 10 is implemented.

22. A readable storage medium, characterized in that, It contains an executable program, which, when executed, implements the method as described in any one of claims 1 to 10.

Images