Intelligent cone dynamic sounding detection system and method

Abstract

The invention provides an intelligent cone dynamic penetration detection system and method. The intelligent cone dynamic sounding detection system comprises automatic sounding equipment, an edge calculation module and a remote data analysis platform, the automatic sounding equipment comprises a sounding mechanism, a falling distance calibration module and a perpendicularity detection module; the drop distance calibration module is used for detecting the lifting height of the piercing hammer; the edge calculation module is in communication connection with the automatic sounding equipment and is used for performing real-time preprocessing and feature extraction on the sounding data; the remote data analysis platform is in communication connection with the edge calculation module and is used for operating an AI analysis model so as to carry out foundation bearing capacity prediction and / or soil parameter inversion on the processed penetration data; through cooperative work of the field automatic sounding equipment, the edge calculation module and the remote data analysis platform, real-time transmission, remote diagnosis, multi-source information fusion and visual early warning of detection data are realized, so that the working efficiency and the management level are greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of cone penetration testing technology, specifically relating to an intelligent cone dynamic penetration testing system and method. Background Technology

[0002] The dynamic cone penetration test (DCPT) is a widely used in-situ testing method. It involves driving a cone probe of a specific size into the soil layer using a standard-mass hammer. The number of blows required to penetrate to a certain depth determines the engineering properties of the soil and estimates the bearing capacity of the foundation. Traditional dynamic cone penetration testing equipment has the following technical problems:

[0003] Low level of automation: It relies heavily on manual operation, including pulling ropes to lift the hammer, manually recording the number of hammer blows and the penetration depth, which leads to strong data subjectivity, large errors and low efficiency;

[0004] Inaccurate control of key parameters: Key test parameters such as drop distance and verticality rely on manual visual inspection and experience for control, which makes it difficult to ensure the standardization and comparability of each test and affects the reliability of the data;

[0005] Data analysis is lagging and relies on experience: bearing capacity results cannot be obtained in real time on site, and engineers need to estimate them afterward based on empirical formulas. The analysis process is not transparent and it is difficult to invert richer soil mechanical parameters.

[0006] Outdated data management: Paper records are used, making it difficult to achieve data digitization, traceability, and remote collaboration.

[0007] Therefore, it is necessary to design an intelligent cone dynamic penetration test system to solve the current technical problems. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an intelligent cone dynamic penetration testing system and method that can achieve technical effects such as automated operation, intelligent calibration, real-time analysis, and remote management.

[0009] The technical solution of this invention is: an intelligent cone dynamic penetration testing system, comprising:

[0010] Automated probing equipment, edge computing modules, and remote data analysis platforms;

[0011] The automated probing device includes a probing mechanism, a drop distance calibration module, and a verticality detection module.

[0012] The probing mechanism has a core hammer that is driven upward by a lifting motor.

[0013] The drop distance calibration module is used to detect the lifting height of the hammer and keep the drop distance of the hammer at a preset drop distance value.

[0014] The verticality detection module is used to detect the tilt angle of the probe rod;

[0015] The edge computing module is communicatively connected to the automated probing device and is used for real-time preprocessing and feature extraction of the probing data;

[0016] The remote data analysis platform is communicatively connected to the edge computing module and is used to run AI analysis models to predict the foundation bearing capacity and / or invert soil parameters on the processed penetration data.

[0017] Furthermore, the drop distance calibration module includes a controller and a drop hammer ranging component;

[0018] The penetrating hammer ranging component is set at the upper end of the penetrating rod of the penetrating mechanism, corresponding to the top of the penetrating hammer, and is used to measure the current actual drop distance value in real time.

[0019] The controller is used to calculate the deviation between the actual drop distance and the preset drop distance, and when the deviation exceeds the preset range, it controls the lifting motor to adjust the height of the hammer.

[0020] Furthermore, the probing mechanism includes an electromagnetic adsorption component disposed above the hammer. The electromagnetic adsorption component controls the on / off state of the electromagnetic force by triggering a switch to achieve magnetic attraction or release of the hammer.

[0021] The lifting motor is connected to the electromagnetic adsorption component via a transmission component, and is used to control the raising or lowering of the electromagnetic adsorption component.

[0022] Furthermore, the tactile sensing mechanism includes a depth detection component and a tactile detection component;

[0023] The depth detection component is used to detect the penetration depth of the probe; the tactile detection component is used to detect the contact state with the soil layer.

[0024] Furthermore, the tactile detection component is a flexible thin-film pressure sensor that is attached in an array to the outer surface of the conical probe or to the probe rod.

[0025] Furthermore, the AI ​​analysis model is one or more of the following models:

[0026] A hybrid model based on 1D CNN-BiLSTM is used to process real-time hammer impact sequence data and perform dynamic prediction of foundation bearing capacity.

[0027] A smart inversion framework based on RBF neural network and NSGA-II multi-objective optimization algorithm is used to simultaneously invert multiple soil mechanical parameters from probe data.

[0028] Furthermore, the hybrid model of the 1D CNN-BiLSTM includes a one-dimensional convolutional neural network layer, a bidirectional long short-term memory network layer, and a fully connected output layer;

[0029] The one-dimensional convolutional neural network layer is used to perform convolution and pooling operations on the input temporal probe data to extract local impact features;

[0030] The bidirectional long short-term memory network layer is connected after the one-dimensional convolutional neural network layer and is used to capture the long-term dependence of the local impact features and the stratigraphic change trend during the penetration process from both forward and backward directions.

[0031] The fully connected output layer is connected after the bidirectional long short-term memory network layer and is used to integrate the long-term dependency relationship and the trend of stratum change to output the dynamic prediction value of foundation bearing capacity.

[0032] Furthermore, the intelligent inversion framework of the RBF neural network and NSGA-II multi-objective optimization algorithm includes an RBF neural network model and an NSGA-II multi-objective optimization algorithm module;

[0033] The RBF neural network model is pre-trained using historical penetration data and corresponding soil mechanics true values ​​to establish a nonlinear mapping relationship from penetration response characteristics to multiple soil mechanics parameters.

[0034] The NSGA-II multi-objective optimization algorithm module is used to input the real-time collected penetration data into the trained RBF neural network, and to perform multi-objective optimization with the goal of minimizing the comprehensive fitting error between the multiple soil mechanical parameters predicted by the RBF neural network and the measured penetration data, and simultaneously invert the optimal solution set of multiple soil mechanical parameters including cohesion, internal friction angle and density.

[0035] Furthermore, the remote data analysis platform includes:

[0036] A multi-source data fusion interface is used to access and fuse auxiliary data from the UAV visual inspection module and / or the soil in-situ sensor module;

[0037] The visualization and early warning module is used to display detection data and analysis results, and to issue early warning signals when anomalies are detected.

[0038] The intelligent cone dynamic penetration test method, employing the test system described in any of the preceding methods, includes the following steps:

[0039] The automated probing device is used to conduct automated probing tests on-site and collect time-series data in real time.

[0040] The edge computing module performs real-time preprocessing and feature extraction on time-series data.

[0041] The processed feature data is input into the trained AI analysis model to obtain the predicted value of foundation bearing capacity and / or soil parameter inversion results in real time.

[0042] The prediction results, inversion results, and raw data are transmitted to the remote data analysis platform via a communication network for display, storage, and early warning.

[0043] The beneficial effects of this invention are:

[0044] (1) In this invention, the drop distance calibration module realizes automatic drop distance calibration, and the verticality detection module realizes real-time verticality monitoring, which completely eliminates human operation error and ensures the standardization and high reproducibility of the test process;

[0045] (2) By introducing AI models, end-to-end intelligent mapping from data to decision-making is realized. It can not only predict bearing capacity with high accuracy, but also invert multiple key soil parameters, which has strong interpretability.

[0046] (3) Through the collaborative work of on-site automated probing equipment, edge computing modules and remote data analysis platforms, real-time transmission of detection data, remote diagnosis, multi-source information fusion and visual early warning are realized, thereby greatly improving work efficiency and management level. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the intelligent cone dynamic penetration test system of the present invention.

[0048] Figure 2 This is a schematic diagram of the automated probing device in this invention.

[0049] Figure 3 This is a flowchart of the intelligent cone dynamic penetration test method in this invention. Detailed Implementation

[0050] Various exemplary embodiments of the invention will now be described in detail with reference to the accompanying drawings. The descriptions of the exemplary embodiments are merely illustrative and are in no way intended to limit the invention or its application or use. The invention can be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to make the invention thorough and complete, and to fully express the scope of the invention to those skilled in the art. It should be noted that, unless otherwise specifically stated, the relative arrangement of components and steps, the composition of materials, numerical expressions, and values ​​set forth in these embodiments should be interpreted as merely exemplary and not as limiting.

[0051] The terms "first," "second," and similar words used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different parts. Words such as "including" or "comprising" mean that the element preceding the word encompasses the element listed after it, without excluding the possibility of encompassing other elements. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0052] like Figure 1 and 2 As shown, an intelligent cone penetration testing system is disclosed, comprising: an automated penetration testing device 1, an edge computing module 2, and a remote data analysis platform 3; the automated penetration testing device 1 includes a penetration mechanism, a drop height calibration module, and a verticality detection module; the penetration mechanism has a mandrel 12 driven upward by a lifting motor 15; the drop height calibration module is used to detect the lifting height of the mandrel, so that the drop height of the mandrel is maintained at a preset drop height value; the verticality detection module is used to detect the inclination angle of the penetration rod; the edge computing module 2 is communicatively connected to the automated penetration testing device 1 and is used to perform real-time preprocessing and feature extraction of the penetration data; the remote data analysis platform 3 is communicatively connected to the edge computing module 2 and is used to run an AI analysis model to predict the foundation bearing capacity and / or invert soil parameters on the processed penetration data.

[0053] In the above embodiments, the drop distance calibration module enables automatic drop distance calibration, and the verticality detection module enables real-time verticality monitoring, completely eliminating human error and ensuring the standardization and high reproducibility of the test process. The introduction of AI models realizes end-to-end intelligent mapping from data to decision-making, which can not only predict bearing capacity with high accuracy, but also invert multiple key soil parameters, providing strong interpretability. Through the collaborative work of on-site automated penetration testing equipment, edge computing modules, and remote data analysis platforms, real-time transmission of test data, remote diagnosis, multi-source information fusion, and visual early warning are realized, thereby greatly improving work efficiency and management level.

[0054] In some embodiments, the drop distance calibration module includes a controller and a hammer measuring component; the hammer measuring component 13 is disposed on the upper end of the probe rod 11 of the probe mechanism corresponding to the top of the hammer, and is used to measure the current actual drop distance value in real time; the controller is used to calculate the deviation between the actual drop distance value and the preset drop distance value, and control the lifting motor to adjust the height of the hammer when the deviation exceeds the preset range.

[0055] Specifically, the hammer-through-the-center distance measuring component 13 is a distance measuring sensor used to detect the distance between the upper end of the probe rod 11 and the probe rod 11 in real time, which is recorded as the actual drop distance value. The controller calculates the deviation between the actual drop distance value and the preset drop distance value, and adjusts the height of the hammer-through-the-center 12 according to the deviation value so that the actual drop distance value and the preset drop distance value tend to 0. Then, the hammer-through-the-center 12 is released to fall freely and strike the probe rod 11, which can ensure that the lifting height and falling speed of the hammer-through-the-center 12 are consistent.

[0056] In some embodiments, the probe mechanism includes an electromagnetic adsorption component 14 disposed above the hammer 12. The electromagnetic adsorption component 14 controls the on / off state of the electromagnetic force by triggering the state change of the switch to achieve magnetic attraction or release of the hammer 12. The lifting motor 15 is connected to the electromagnetic adsorption component 14 through a transmission component and is used to control the rise or fall of the electromagnetic adsorption component.

[0057] Specifically, the lifting motor 15 uses a steel wire rope and a pulley set on the top of the support 17 to control the raising and lowering of the electromagnetic adsorption component 14. During construction, the electromagnetic adsorption component 14 adsorbs the hammer 12, and the lifting motor 15 controls the electromagnetic adsorption component 14 to rise. The electromagnetic adsorption component 14 magnetically attracts the hammer 12 and rises to the height corresponding to the preset drop distance value. When the electromagnetic adsorption component 14 is de-energized, it releases the hammer 12, and the hammer 12 falls freely to the top of the hammer pad 16. As an example, the electromagnetic adsorption component 14 is an electromagnet.

[0058] In some embodiments, the penetration mechanism includes a depth detection component and a tactile detection component; the depth detection component is used to detect the penetration depth of the penetration rod; the tactile detection component is used to detect the soil contact state; the system integrates the depth detection component and the tactile detection component, which can not only record the penetration curve, but also sense the soil contact state, providing richer input features for AI analysis.

[0059] Specifically, the tactile detection component is a flexible thin-film pressure sensor that is attached in an array to the outer surface of the conical probe or to the probe rod.

[0060] Specifically, the depth detection component 18 is a distance sensor installed on the top of the bracket 17 corresponding to the upper end of the probe rod 11. Before construction, the probe rod 11 is brought into contact with the ground, and the distance between the depth detection component 18 and the probe rod 11 at this time is obtained as the first distance value. During construction, the distance between the depth detection component 18 and the probe rod 11 is collected in real time as the second distance value. The difference between the second distance value and the first distance value can be used to obtain the depth value of the probe rod 11 entering the ground.

[0061] In some embodiments, the AI ​​analysis model is one or more of the following models:

[0062] A hybrid model based on 1D CNN-BiLSTM is used to process real-time hammer impact sequence data and perform dynamic prediction of foundation bearing capacity.

[0063] A smart inversion framework based on RBF neural network and NSGA-II multi-objective optimization algorithm is used to simultaneously invert multiple soil mechanical parameters from probe data.

[0064] Specifically, the 1D CNN-BiLSTM hybrid model includes a one-dimensional convolutional neural network layer, a bidirectional long short-term memory network layer, and a fully connected output layer;

[0065] One-dimensional convolutional neural network layers are used to perform convolution and pooling operations on the input temporal probe data to extract local impact features;

[0066] A bidirectional long short-term memory network layer is connected after a one-dimensional convolutional neural network layer to capture the long-term dependence of local impact features and the trend of stratigraphic change during the penetration process from both forward and backward directions.

[0067] The fully connected output layer is connected after the bidirectional long short-term memory network layer. It is used to integrate long-term dependencies and geological change trends to output dynamic prediction values ​​of foundation bearing capacity.

[0068] As a more concrete example of a 1D CNN-BiLSTM hybrid model, the 1D CNN-BiLSTM hybrid model includes the following steps:

[0069] Data preparation and input: The real-time collected time-series penetration data (such as hammer acceleration signals and penetration depth sequences) are divided into time steps to form the model input; the input data needs to be normalized and preprocessed before entering the model;

[0070] Feature extraction: The one-dimensional convolutional neural network layer uses multiple convolutional kernels of different widths to perform one-dimensional convolution operations on the input sequence along the time dimension to capture subtle features such as local impact waveforms and frequencies contained in each hammer impact signal; after convolution, nonlinearity is introduced through the ReLU activation function, and downsampling is performed through a pooling layer (preferably max pooling) to enhance the robustness of the features;

[0071] Sequence modeling: The local feature sequence extracted by CNN is input into a bidirectional long short-term memory network layer; the network scans the sequence from the forward and backward directions respectively through two independent LSTMs, thereby comprehensively learning the stratigraphic context information before and after the current hammer impact point during the penetration process, effectively capturing the long-term dependence and overall trend of soil layer changes.

[0072] Output: The output state of the last time step of the bidirectional long short-term memory network layer (or the summary information of the outputs of all time steps) is passed to the fully connected output layer. This layer performs nonlinear combination and mapping of the learned high-dimensional features, and finally outputs a continuous value as the dynamic prediction value of the foundation bearing capacity corresponding to the current depth;

[0073] Model training: A large amount of historical penetration data and corresponding static load test measured bearing capacity values ​​are used as the training set. The mean squared error is used as the loss function, and the Adam optimization algorithm is used to perform end-to-end supervised training on all parameters in the model (including convolutional kernel weights, LSTM unit weights, and fully connected layer weights).

[0074] Specifically, the intelligent inversion framework of RBF neural network and NSGA-II multi-objective optimization algorithm includes RBF neural network model and NSGA-II multi-objective optimization algorithm module;

[0075] The RBF neural network model is pre-trained using historical penetration data and corresponding soil mechanics ground truth values ​​to establish a nonlinear mapping relationship from penetration response characteristics to multiple soil mechanics parameters.

[0076] The NSGA-II multi-objective optimization algorithm module is used to input the real-time collected penetration data into the trained RBF neural network, and to perform multi-objective optimization with the goal of minimizing the comprehensive fitting error between the multiple soil mechanical parameters predicted by the RBF neural network and the measured penetration data. Simultaneously, it inversely derives the optimal solution set of multiple soil mechanical parameters, including cohesion, internal friction angle and density.

[0077] As a more specific example of the intelligent inversion framework of RBF neural network and NSGA-II multi-objective optimization algorithm, the intelligent inversion framework of RBF neural network and NSGA-II multi-objective optimization algorithm includes building and training RBF neural network model and solving NSGA-II multi-objective optimization inversion.

[0078] The construction and training of the RBF neural network model includes the following steps:

[0079] Input and output: The input feature vector is the characteristics of historical penetration data (such as dynamic penetration resistance, number of blows, and penetration depth); the target output is the true values ​​of soil mechanical parameters obtained from laboratory tests at the corresponding location (such as cohesion, internal friction angle, and density).

[0080] Network training: Using historical datasets, learning algorithms such as gradient descent or least squares are employed to determine the center, width, and output layer weights of the RBF neural network, thereby establishing a high-precision nonlinear mapping relationship from probe features to soil parameters. The trained network then becomes a fast surrogate model.

[0081] The NSGA-II multi-objective optimization inversion solution includes the following steps:

[0082] Initialization: For the soil parameters to be inverted (such as cohesion c, internal friction angle φ, density ρ), a reasonable range of values ​​is set based on prior knowledge, and an initial population is randomly generated within this range;

[0083] Objective function definition: Input a set of real-time collected penetration data into the aforementioned trained RBF neural network to obtain a set of predicted soil parameters; based on these predicted values ​​and combined with the dynamic penetration formula, calculate the theoretical penetration resistance corresponding to the set of parameters; the optimization objective is set to minimize the fitting error between the theoretical response and the measured penetration data; if there are multiple measuring points or indices, multiple objective functions to be minimized are formed.

[0084] Optimization process: The NSGA-II algorithm iteratively evolves the population through genetic operations such as selection, crossover, and mutation. In each generation, it selects a Pareto optimal set that performs well on multiple fitting error targets and has good dispersion based on fast non-dominated sorting and crowding calculation.

[0085] Output: After the optimization process converges, the solution with the smallest comprehensive fitting error is selected from the Pareto optimal solution set of the final generation. This solution is the optimal estimate of multiple soil mechanical parameters (cohesion, internal friction angle, density) obtained by synchronous inversion.

[0086] In some embodiments, the remote data analysis platform includes: a multi-source data fusion interface for accessing and fusing auxiliary data from the UAV visual inspection module and / or the soil in-situ sensor module; and a visualization and early warning module for displaying inspection data and analysis results, and issuing early warning signals when an anomaly is detected.

[0087] As an example, the multi-source data fusion interface provides one or more standardized communication protocols, such as HTTP API and MQTT, and data format specifications such as JSON and GeoJSON. Two-dimensional orthophotos or three-dimensional reality models generated from image data processed from the UAV visual inspection module, along with time-series data from the soil in-situ sensor module, such as resistivity, moisture content, and temperature, are uploaded to the platform through this interface. The platform creates a unified data context for each inspection project, aligning and associating the aforementioned auxiliary data with the spatial coordinates and timestamps obtained from the GPS positioning of the probing equipment, forming a multi-dimensional, spatiotemporally consistent fusion dataset, providing a data foundation for comprehensive analysis.

[0088] As an example, the visualization function is based on web technologies such as HTML5 and WebGL to develop a graphical interactive interface. The interface uses a two-dimensional map or a three-dimensional scene as the base map and overlays to display the geographical distribution of the penetration test sites, the penetration curves that change with depth (such as the number of hammer blows-depth curve, the dynamic penetration resistance-depth curve), the foundation bearing capacity prediction curve output by the AI ​​model and the soil parameter profile obtained by inversion, UAV aerial images, the location of soil sensor points and their monitoring data.

[0089] The system has pre-set warning thresholds based on rules or statistical models, such as predicted bearing capacity values ​​being lower than design values, sudden changes in hammer blow counts, and abnormal soil parameters. The platform monitors input data and analysis results in real time, and automatically triggers a warning once data anomalies or results exceeding limits are detected. Warning signals are issued through the messaging service integrated into the platform, by highlighting interface elements and sending SMS messages or application push notifications to designated responsible persons.

[0090] In some embodiments, such as Figure 3 As shown, an intelligent cone dynamic penetration test method is disclosed, employing any of the above-mentioned detection systems, including the following steps:

[0091] S1, Automated probing tests are conducted on-site using automated probing equipment, and time-series data is collected in real time; S2, The time-series data is preprocessed and features are extracted in real time using an edge computing module;

[0092] S3, input the processed feature data into the trained AI analysis model to obtain the predicted value of foundation bearing capacity and / or soil parameter inversion results in real time;

[0093] S4 transmits the prediction results, inversion results, and raw data to a remote data analysis platform via a communication network for display, storage, and early warning.

[0094] The various embodiments of the present invention have now been described in detail. To avoid obscuring the concept of the invention, some details known in the art have not been described. Those skilled in the art will fully understand how to implement the technical solutions disclosed herein based on the above description.

[0095] The embodiments described above only illustrate some implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. An intelligent cone dynamic penetration testing system, characterized in that, include: Automated probing equipment, edge computing modules, and remote data analysis platforms; The automated probing device includes a probing mechanism, a drop distance calibration module, and a verticality detection module. The probing mechanism has a core hammer that is driven upward by a lifting motor. The drop distance calibration module is used to detect the lifting height of the hammer and keep the drop distance of the hammer at a preset drop distance value. The verticality detection module is used to detect the tilt angle of the probe rod; The edge computing module is communicatively connected to the automated probing device and is used for real-time preprocessing and feature extraction of the probing data; The remote data analysis platform is communicatively connected to the edge computing module and is used to run AI analysis models to predict the foundation bearing capacity and / or invert soil parameters on the processed penetration data.

2. The intelligent cone dynamic penetration testing system according to claim 1, characterized in that: The drop distance calibration module includes a controller and a drop hammer measuring component; The penetrating hammer ranging component is set at the upper end of the penetrating rod of the penetrating mechanism, corresponding to the top of the penetrating hammer, and is used to measure the current actual drop distance value in real time. The controller is used to calculate the deviation between the actual drop distance and the preset drop distance, and when the deviation exceeds the preset range, it controls the lifting motor to adjust the height of the hammer.

3. The intelligent cone dynamic penetration testing system according to claim 1, characterized in that: The probing mechanism includes an electromagnetic adsorption component disposed above the hammer. The electromagnetic adsorption component controls the on / off state of the electromagnetic force by triggering the state change of the switch to achieve magnetic attraction or release of the hammer. The lifting motor is connected to the electromagnetic adsorption component via a transmission component, and is used to control the raising or lowering of the electromagnetic adsorption component.

4. The intelligent cone dynamic penetration testing system according to claim 1, characterized in that: The tactile detection mechanism includes a depth detection component and a tactile detection component; The depth detection component is used to detect the penetration depth of the probe; the tactile detection component is used to detect the contact state with the soil layer.

5. The intelligent cone dynamic penetration testing system according to claim 4, characterized in that: The tactile detection component is a flexible thin-film pressure sensor that is attached in an array to the outer surface of the conical probe or to the probe rod.

6. The intelligent cone dynamic penetration testing system according to claim 1, characterized in that, The AI ​​analysis model is one or more of the following models: A hybrid model based on 1D CNN-BiLSTM is used to process real-time hammer impact sequence data and perform dynamic prediction of foundation bearing capacity. A smart inversion framework based on RBF neural network and NSGA-II multi-objective optimization algorithm is used to simultaneously invert multiple soil mechanical parameters from probe data.

7. The intelligent cone dynamic penetration test system according to claim 6, characterized in that: The hybrid model of 1DCNN-BiLSTM includes a one-dimensional convolutional neural network layer, a bidirectional long short-term memory network layer, and a fully connected output layer; The one-dimensional convolutional neural network layer is used to perform convolution and pooling operations on the input temporal probe data to extract local impact features; The bidirectional long short-term memory network layer is connected after the one-dimensional convolutional neural network layer and is used to capture the long-term dependence of the local impact features and the stratigraphic change trend during the penetration process from both forward and backward directions. The fully connected output layer is connected after the bidirectional long short-term memory network layer and is used to integrate the long-term dependency relationship and the trend of stratum change to output the dynamic prediction value of foundation bearing capacity.

8. The intelligent cone dynamic penetration test system according to claim 6, characterized in that: The intelligent inversion framework of the RBF neural network and NSGA-II multi-objective optimization algorithm includes an RBF neural network model and an NSGA-II multi-objective optimization algorithm module; The RBF neural network model is pre-trained using historical penetration data and corresponding soil mechanics true values ​​to establish a nonlinear mapping relationship from penetration response characteristics to multiple soil mechanics parameters. The NSGA-II multi-objective optimization algorithm module is used to input the real-time collected penetration data into the trained RBF neural network, and to perform multi-objective optimization with the goal of minimizing the comprehensive fitting error between the multiple soil mechanical parameters predicted by the RBF neural network and the measured penetration data, and simultaneously invert the optimal solution set of multiple soil mechanical parameters including cohesion, internal friction angle and density.

9. The intelligent cone dynamic penetration test system according to claim 1, characterized in that, The remote data analysis platform includes: A multi-source data fusion interface is used to access and fuse auxiliary data from the UAV visual inspection module and / or the soil in-situ sensor module; The visualization and early warning module is used to display detection data and analysis results, and to issue early warning signals when anomalies are detected.

10. A method for intelligent cone dynamic penetration testing, employing the testing system described in any one of claims 1 to 9, characterized in that, Includes the following steps: The automated probing device is used to conduct automated probing tests on-site and collect time-series data in real time. The edge computing module performs real-time preprocessing and feature extraction on time-series data. The processed feature data is input into the trained AI analysis model to obtain the predicted value of foundation bearing capacity and / or soil parameter inversion results in real time. The prediction results, inversion results, and raw data are transmitted to the remote data analysis platform via a communication network for display, storage, and early warning.

Images