Intelligent system and construction method of UHPC under-reamed uplift anchor rod penetrating into rock stratum

Abstract

The application discloses a kind of UHPC hole expansion uplift anchor rod intelligent systems and construction methods in-depth rock formation;The system includes data acquisition module, intelligent decision module, intelligent execution module and quality feedback module;Data acquisition module obtains drilling parameter, borehole wall image and acoustic data in real time;Intelligent decision module analyzes geological information through pre-trained neural network model, and dynamically outputs recommended parameters including hole geometry, drilling process, UHPC material performance and grouting process;Intelligent execution module automatically completes drilling, hole expansion, UHPC preparation and grouting according to recommended parameters;Quality feedback module monitors anchor rod displacement and grouting body quality, forms closed-loop optimization;Construction method includes trial hole construction, geological information acquisition, parameter intelligent decision, process intelligent execution and effect feedback optimization;The application realizes self-adaptation and intelligentization of construction process, significantly improves anchor rod construction quality, efficiency and reliability in complex rock formation.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering technology, and in particular to a UHPC borehole-expanding anti-pull-out intelligent system and construction method for deep-penetrating rock strata. Background Technology

[0002] As a critical pull-out resisting and support component in geotechnical engineering, the construction quality of anchor bolts directly affects the long-term safety and stability of engineering structures. Traditional anchor bolt construction methods mainly include drilling, hole protection, anchor bolt installation, and grouting, relying heavily on the experience of construction personnel for parameter selection and process control. However, under complex and variable geological conditions, such as uneven rock weathering, well-developed joints and fissures, the presence of weak interlayers, or abundant groundwater, this experience-driven construction model reveals significant limitations.

[0003] Traditional methods struggle to dynamically adjust to the actual geological conditions revealed during drilling. Drilling with conventional parameters in fractured strata can easily lead to borehole instability and collapse, while using standard-mix grout in water-permeable layers may cause grout loss and reduced anchoring force. Furthermore, drilling, wall protection, and grouting are often performed independently, lacking collaborative optimization based on real-time geological information, making the transitions between processes weak points in quality. Construction quality is highly dependent on individual experience, resulting in significant quality fluctuations between different projects and even different sections of the same project, with low reproducibility. The cost and risk of trial and error are high when facing entirely new or rare geological conditions. Simultaneously, information such as drilling parameters and rock cuttings samples obtained during construction is usually only used for qualitative judgment, rarely quantified and fed back in real-time to guide subsequent processes; post-construction testing data is mostly used for result acceptance and is not effectively used to optimize the construction parameter system, preventing the systematic accumulation and iteration of knowledge and experience.

[0004] In recent years, ultra-high performance concrete (UHPC) has been introduced into anchor bolt engineering due to its excellent mechanical properties and durability. In particular, the increased borehole enlargement process can significantly increase the contact area between the anchor body and the rock mass, thereby improving pull-out bearing capacity. However, the performance of UHPC is extremely sensitive to mix proportions and construction techniques, requiring higher on-site adaptability under complex geological conditions, which further increases the difficulty of construction control. On the other hand, the development of intelligent sensing, the Internet of Things (IoT), and artificial intelligence (AI) technologies has made intelligent construction processes possible. While existing technologies have attempted to integrate unit technologies such as drilling measurement while drilling (DWD) or automated batching, most are limited to the automation of local processes or simple feedback. A closed-loop intelligent system capable of spanning the entire process of geological perception, intelligent decision-making, precise execution, and effect evaluation has not yet been formed, failing to achieve the core goal of adaptive adjustment and continuous learning optimization based on geological changes.

[0005] Therefore, in response to the prominent challenges faced by the construction of ultra-high performance concrete (UHPC) borehole enlargement anti-pull-out anchors in complex rock strata, such as high geological uncertainty, difficulty in coordinating multiple processes, and excessive reliance on experience for construction quality, there is an urgent need to invent a comprehensive intelligent construction system and method that can perceive geological information in real time, intelligently generate construction parameters, accurately execute customized processes, and have online optimization and iteration capabilities. This would promote the transformation of anchor construction from experience-driven to data and model-driven, ensuring the high-quality, efficient, and reliable implementation of anchor projects under complex geological conditions. Summary of the Invention

[0006] This invention aims to provide an intelligent system and construction method for deep-penetrating UHPC borehole enlargement anti-pull-out anchor bolts capable of adapting to complex geological conditions. This solution achieves self-adaptation and intelligence throughout the entire construction process by constructing a complete closed loop of perception, decision-making, execution, and feedback.

[0007] A UHPC borehole enlargement anti-pull-out anchor bolt intelligent system for deep rock formations includes a data acquisition module, an intelligent decision-making module, an intelligent execution module, and a quality feedback module; The data acquisition module is integrated into the drilling equipment and is used to collect and analyze geological information in real time during construction. The data acquisition module includes a measurement-while-drilling (MWD) device, a borehole imager, and a micro acoustic testing device. The MWD device is used to acquire real-time drilling parameters, including drilling speed, drill bit torque, axial thrust, and rotational speed. The borehole imager is used to acquire rock mass structure images of the borehole wall, providing visual evidence for analyzing rock mass integrity. The micro acoustic testing device is used to perform acoustic testing at a specific depth within the borehole to obtain acoustic velocity data. The intelligent decision-making module includes at least one computing server and is equipped with a pre-trained neural network model. The neural network model takes real-time drilling parameters, rock mass structure image information, and acoustic velocity data acquired by the data acquisition module as input, and outputs a recommended set of construction parameters for the current construction area. The recommended set of construction parameters includes recommended parameters for borehole geometry, drilling process, UHPC material properties, and grouting process. The recommended parameters for borehole geometry include hole depth, diameter of the enlarged section, and anchor spacing. The recommended parameters for drilling process include drilling speed and axial thrust. The recommended parameters for UHPC material properties include target slump spread, water-cement ratio, and fiber volume fraction. The recommended parameters for grouting process include grouting pressure and grouting flow rate. Preferably, the core of the intelligent decision-making module lies in a specially designed and trained neural network model, which consists of an input layer, a hidden layer, and an input layer connected in sequence. The model's input layer receives preprocessed data and consists of 11 nodes, including: real-time drilling depth, drilling speed, drill bit torque, axial thrust, and rotational speed; joint and fracture density, average fracture width, dominant fracture dip angle, and borehole wall roughness coefficient extracted from rock mass image information; and acoustic velocity obtained through acoustic testing. Data preprocessing methods include: removing data containing obvious recording errors or sensor failures; filling in missing features using the mean of similar strata data from the same project; and performing Z-score standardization on both input and output data to scale each feature value to a distribution with a mean of 0 and a standard deviation of 1, thereby accelerating model convergence and improving training stability. The model employs a dual-hidden-layer structure. The first hidden layer contains 128 neurons and uses a linear rectified function as the activation function to learn complex underlying patterns from the input features. The second hidden layer contains 64 neurons and also uses a linear rectified function as the activation function to further combine features to form a higher-level abstract representation. This structure ensures that the model has strong expressive power while effectively controlling the risk of overfitting. The output layer of the model corresponds to the recommended set of construction parameters generated by intelligent decision-making. It contains 10 nodes and uses a linear activation function to directly output specific parameter values, including hole depth, diameter of the enlarged section, anchor spacing, drilling speed, axial thrust, target slump expansion, water-cement ratio, fiber volume content, grouting pressure, and grouting flow rate.

[0008] Preferably, the training of the neural network model relies on a high-quality historical engineering database. This database is constructed by collecting a large number of complete and high-quality anchor bolt projects, including: detailed logs of the construction process, borehole geological logging, drilling measurement data, borehole imaging data, acoustic test data, and the final anchor bolt pull-out force test report and grout quality non-destructive testing report. Each data record is reconstructed to form a sample pair. Each sample pair includes an input vector and a label vector. The input vector contains multi-dimensional geological information recorded during the construction of the borehole, corresponding to the features covered by the aforementioned real-time drilling parameters, rock mass structure image information, and acoustic velocity data. The label vector contains the construction parameter combination that was ultimately verified as successful for the borehole and can achieve the designed pull-out force, corresponding to the parameters covered by the aforementioned recommended set of construction parameters. The initial training set contains at least 5,000 valid sample pairs and is continuously expanded through successful cases generated by the system in practical applications. The neural network model training algorithm employs stochastic gradient descent with momentum, combined with backpropagation to calculate the gradient. The loss function uses mean squared error to minimize the gap between the model's predicted parameters and the actual successful parameters. The dataset is randomly divided into training and test sets, with 80% to 20% of the dataset allocated to each set. During training, the model's performance is validated on the test set after each training epoch. Training is terminated early when the model's loss function value on the test set no longer decreases for 10 consecutive training epochs to prevent overfitting.

[0009] The intelligent execution module receives instructions from the intelligent decision-making module and executes corresponding construction operations, including an intelligent drilling rig, an automatic grouting device, and a UHPC automatic batching device. The intelligent drilling rig automatically adjusts and controls the drilling speed and axial thrust based on the recommended hole geometry parameters and drilling process parameters issued by the intelligent decision-making module, achieving the hole depth, enlarged section diameter, and anchor bolt spacing specified by the recommended hole geometry parameters. The automatic grouting device automatically adjusts and controls the grouting pressure and grouting flow rate based on the recommended grouting process parameters issued by the intelligent decision-making module. The UHPC automatic batching device automatically completes the precise metering, feeding, and mixing of materials based on the recommended UHPC material performance parameters issued by the intelligent decision-making module, preparing UHPC with specified target slump spread, water-cement ratio, and fiber volume fraction. The quality feedback module is used to collect construction quality data during and after construction, including anchor displacement sensors and grout ultrasonic testing instruments. The anchor displacement sensors are deployed on the anchor body to continuously monitor the axial displacement of the anchor during anchor tensioning and service, providing direct evidence for evaluating the anchor's pull-out resistance. The grout ultrasonic testing instrument is used to perform non-destructive scanning of the anchoring section after the grout has solidified to detect the internal density and uniformity of the grout. The quality feedback module transmits the collected data to the intelligent decision-making module in real time, forming a closed-loop control from geological perception, intelligent decision-making, precise execution to effect verification, providing data support for online optimization and iteration of construction parameters.

[0010] A construction method for a UHPC borehole enlargement anti-pull-out intelligent anchor system for deep rock formations, based on the aforementioned intelligent system, includes the following steps: S1. Construction of test holes: At the location of the designed anchor group, select no less than 10% of the total number of holes as test holes for construction, and drill to a depth slightly deeper than the designed hole depth. S2. Geological Information Acquisition: During the drilling of the test hole, geological information is collected in real time through the data acquisition module, including real-time drilling parameters, image information of the rock mass structure of the borehole wall, and acoustic velocity data. S3. Intelligent decision-making for construction parameters: The geological information of the current borehole location collected in step S2 is input into the pre-trained neural network model in the intelligent decision-making module for real-time analysis and calculation, and the recommended set of construction parameters is dynamically output and sent to the intelligent execution module. S4. Intelligent Execution of Construction Process: The intelligent execution module executes construction operations based on the received set of recommended construction parameters. The intelligent drilling rig completes drilling and reaming according to the recommended parameters for hole geometry and drilling process; the UHPC automatic batching equipment prepares UHPC material according to the recommended parameters for UHPC material properties; and the automatic grouting equipment completes grouting and anchor bolt installation according to the recommended parameters for grouting process. S5. Construction effect monitoring: Collect construction quality data during and after construction, including the axial displacement of anchor bolts and the internal density and uniformity of grout. S6. Feedback Optimization: The quality feedback module transmits the collected construction quality data to the intelligent decision-making module in real time, and performs feedback optimization and iteration on the neural network model.

[0011] In summary, compared with existing technologies, the UHPC borehole enlargement anti-pull-out anchor intelligent system and construction method for deep-penetrating rock formations provided by this invention have the following significant advantages: 1) Improve construction quality and reliability: The data acquisition module perceives multi-dimensional geological information revealed by the borehole in real time, such as drilling parameters, borehole wall structure and rock strength. The pre-trained neural network model in the intelligent decision-making module performs real-time analysis and dynamically generates an optimized set of construction parameters that is highly adapted to the current geological conditions. This realizes a fundamental shift from experience-driven to data and model-driven, ensuring that the parameters of key links such as borehole geometry, ultra-high performance concrete material performance and grouting process are accurately matched with the actual rock strata conditions. This effectively avoids problems such as borehole wall collapse, grout loss and insufficient anchoring force caused by improper parameters, and greatly improves the overall construction success rate and pull-out bearing capacity reliability of single anchors and anchor groups. 2) Intelligent Adaptation in the Construction Process: A complete closed loop of perception, decision-making, execution and feedback has been constructed. The intelligent execution module includes intelligent drilling rigs, automatic batching and grouting equipment, which can accurately execute customized instructions issued by the decision-making module, realize the automation and integrated collaborative operation of drilling, hole enlargement, ultra-high performance concrete preparation, grouting and other processes. This not only reduces the quality fluctuations caused by process connection, but also significantly improves construction efficiency and reduces labor intensity and dependence on highly skilled operators.

[0012] 3) Enhanced adaptability to complex and variable geological conditions: The intelligent decision-making module has continuous learning and optimization capabilities; the neural network model is trained based on a large number of historical successful cases and can capture the complex nonlinear relationship between geological features and successful construction parameters; when faced with new or even rare geological conditions, the system can recommend construction schemes with high success rates based on real-time perception data, reducing trial and error costs and engineering risks. In particular, the system's dynamic adjustment advantages are more obvious for complex situations such as uneven rock weathering, well-developed joints and fissures, and the presence of weak interlayers. 4) It promotes the accumulation and iteration of construction experience and knowledge: The system collects construction effect data, such as anchor displacement and grout density, through the quality feedback module, and feeds it back to the historical database along with corresponding geological information and construction parameters as successful cases. Through regular incremental model learning, the neural network model is continuously optimized and upgraded, enabling the system's decision-making capabilities to continuously improve with the accumulation of engineering practice. This realizes the transformation of individual construction experience into replicable and iterative systematic knowledge, providing a platform for the overall improvement of the industry's technical level. Attached Figure Description

[0013] Figure 1 This is a connection diagram of a UHPC borehole enlargement anti-pull-out intelligent system for deep-penetrating rock strata, as shown in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the construction method of a UHPC borehole enlargement anti-pull-out intelligent anchor system for penetrating deep rock strata, as shown in an embodiment of the present invention. Figure labels: 1-Data acquisition module, 11-Measurement while drilling equipment, 12-In-hole imager, 13-Miniature acoustic testing device, 2-Intelligent decision-making module, 21-Neural network model, 3-Intelligent execution module, 31-Intelligent drilling rig, 32-Automatic grouting equipment, 33-UHPC automatic batching equipment, 4-Quality feedback module, 41-Anchor bolt displacement sensor, 42-Ultrasonic detector for grouting body. Detailed Implementation

[0014] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustration and explanation only and are not intended to limit the present invention.

[0015] Example 1: A UHPC borehole enlargement anti-pull-out intelligent system for deep-penetrating rock formations.

[0016] like Figure 1 As shown in the figure, this embodiment provides a UHPC borehole enlargement anti-pull-out intelligent system for deep rock formations, including a data acquisition module 1, an intelligent decision-making module 2, an intelligent execution module 3, and a quality feedback module 4; The data acquisition module 1 is integrated into the drilling equipment and is used to collect and analyze geological information in real time during construction. The data acquisition module 1 includes a measurement-while-drilling device 11, a borehole imager 12, and a micro acoustic testing device 13. The measurement-while-drilling device 11 is used to acquire real-time drilling parameters, including drilling speed, drill bit torque, axial thrust, and rotational speed. The borehole imager 12 is used to acquire rock mass structure image information of the borehole wall, providing visual basis for analyzing rock mass integrity. The micro acoustic testing device 13 is used to conduct acoustic testing at a specific depth in the borehole to obtain acoustic velocity data. The acoustic velocity data can be converted into preliminary rock stratum strength data according to specifications. The intelligent decision-making module 2 includes at least one computing server and is equipped with a pre-trained neural network model 21. The neural network model 21 takes real-time drilling parameters, rock mass structure image information, and acoustic velocity data acquired by the data acquisition module as input, and outputs a set of recommended construction parameters for the current construction area through model calculation. The recommended construction parameter set includes recommended hole geometry parameters, recommended drilling process parameters, recommended UHPC material performance parameters, and recommended grouting process parameters. The recommended hole geometry parameters include hole depth, diameter of the enlarged section, and anchor spacing. The recommended drilling process parameters include drilling speed and axial thrust. The recommended UHPC material performance parameters include target slump expansion, water-cement ratio, and fiber volume fraction. The recommended grouting process parameters include grouting pressure and grouting flow rate. In specific implementation, the core of the intelligent decision-making module 2 lies in a specially designed and trained neural network model 21. By learning the complex nonlinear mapping relationship between geological features, construction parameters, and final results hidden in massive historical data, it can recommend construction parameters with a high probability of success when facing new geological conditions. The neural network model 21 consists of an input layer, a hidden layer, and an input layer connected in sequence. The model's input layer receives preprocessed data and consists of 11 nodes, including: real-time drilling depth, drilling speed, drill bit torque, axial thrust, and rotational speed; joint and fracture density, average fracture width, dominant fracture dip angle, and borehole wall roughness coefficient extracted from rock mass image information; and acoustic velocity obtained through acoustic testing. Data preprocessing methods include: removing data containing obvious recording errors or sensor failures; filling in missing features using the mean of similar strata data from the same project; and performing Z-score standardization on both input and output data to scale each feature value to a distribution with a mean of 0 and a standard deviation of 1, thereby accelerating model convergence and improving training stability. The model employs a dual-hidden-layer structure. The first hidden layer contains 128 neurons and uses a linear rectified function as the activation function to learn complex underlying patterns from the input features. The second hidden layer contains 64 neurons and also uses a linear rectified function as the activation function to further combine features to form a higher-level abstract representation. This structure ensures that the model has strong expressive power while effectively controlling the risk of overfitting. The output layer of the model corresponds to the recommended set of construction parameters generated by intelligent decision-making. It contains 10 nodes and uses a linear activation function to directly output specific parameter values, including hole depth, diameter of the enlarged section, anchor spacing, drilling speed, axial thrust, target slump expansion, water-cement ratio, fiber volume content, grouting pressure, and grouting flow rate.

[0017] Preferably, the training of the neural network model 21 relies on a high-quality historical engineering database. This database is constructed by systematically collecting a large number of complete and high-quality anchor bolt projects, including: detailed logs of the construction process, borehole geological logging, drilling measurement data, borehole imaging data, acoustic test data, and the final anchor bolt pull-out force test report and grout quality non-destructive testing report; each data record is reconstructed to form a sample pair; each sample pair includes an input vector and a label vector; the input vector contains multi-dimensional geological information recorded during the construction of the borehole, corresponding to the features covered by the aforementioned real-time drilling parameters, rock structure image information, and acoustic velocity data; the label vector contains the construction parameter combination that was ultimately verified as successful for the borehole and can achieve the design pull-out force, corresponding to the parameters covered by the aforementioned recommended set of construction parameters; the initial training set contains at least 5000 valid sample pairs and is continuously expanded through successful cases generated by the system of this invention in practical applications; In practice, the qualification standard for the tag vector is that it must simultaneously meet the standards for pull-out performance and grout quality; the pull-out performance standard is the measured value of the ultimate pull-out bearing capacity of the anchor bolt at the hole location during the acceptance tension test. P a Not lower than the design value P d The grout quality meets the standard by means of ultrasonic testing and other methods. The average density of the grout in the anchoring section is not less than 95%, and there are no large defects with a diameter greater than 50mm.

[0018] For each successful aperture that meets the above conditions, a sample pair is constructed ( X , Y );in, XAs the input vector, the real-time geological information acquired during the drilling process of the borehole is extracted and organized to form an 11-dimensional feature vector corresponding to the current system data acquisition module, which includes drilling depth, speed, torque, thrust, rotation speed, joint and fracture density, average fracture width, dominant fracture dip angle, borehole wall roughness coefficient, and sonic velocity. Y The label vector records the 10-dimensional construction parameters actually used at that borehole location and verified as successful, including borehole depth, diameter of the enlarged section, anchor bolt spacing, drilling speed, axial thrust, target slump spread, water-cement ratio, fiber volumetric content, grouting pressure, and grouting flow rate. Through these standards, discrete engineering experience is transformed into structured sample data suitable for machine learning. The initial training set should contain at least 5000 such valid sample pairs and can be continuously expanded through new successful cases generated in practical applications using the system of this invention.

[0019] The training algorithm for neural network model 21 employs stochastic gradient descent with momentum, combined with backpropagation to calculate gradients. The loss function uses mean squared error to minimize the gap between the model's predicted parameters and the actual successful parameters. The dataset is randomly divided into training and test sets, with a ratio of 80% to 20%. During training, the model's performance is validated on the test set after each training epoch. Training is terminated early when the model's loss function value on the test set no longer decreases for 10 consecutive training epochs to prevent overfitting.

[0020] The intelligent execution module 3 receives instructions from the intelligent decision-making module and executes corresponding construction operations. It includes an intelligent drilling rig 31, an automatic grouting device 32, and a UHPC automatic batching device 33. The intelligent drilling rig 31 automatically adjusts and controls the drilling speed and axial thrust based on the recommended hole geometry parameters and drilling process parameters issued by the intelligent decision-making module 2, achieving the hole depth, enlarged section diameter, and anchor spacing specified by the recommended hole geometry parameters. The automatic grouting device 32 automatically adjusts and controls the grouting pressure and grouting flow rate based on the recommended grouting process parameters issued by the intelligent decision-making module 2. The UHPC automatic batching device 33 automatically completes the precise metering, feeding, and mixing of materials based on the recommended UHPC material performance parameters issued by the intelligent decision-making module 2, preparing UHPC with specified target slump expansion, water-cement ratio, and fiber volume fraction. The quality feedback module 4 is used to collect construction quality data during and after construction, including an anchor displacement sensor 41 and a grout ultrasonic testing instrument 42. The anchor displacement sensor 41 is installed on the anchor body and is used to continuously monitor the axial displacement of the anchor during anchor tensioning and service, providing direct evidence for evaluating the pull-out performance of the anchor. The grout ultrasonic testing instrument 42 is used to perform non-destructive scanning on the anchoring section after the grout has solidified, in order to detect the internal density and uniformity of the grout. The quality feedback module 4 transmits the collected data to the intelligent decision-making module 2 and the historical project database in real time, forming a closed-loop control from geological perception, intelligent decision-making, precise execution to effect verification, providing data support for online optimization and iteration of construction parameters.

[0021] Example 2: A construction method for a UHPC borehole enlargement and pull-out resistance intelligent anchor system for deep-penetrating rock strata. For example... Figure 2 As shown, the implementation based on the aforementioned intelligent system includes the following steps: S1. Construction of test holes: At the location of the designed anchor group, select no less than 10% of the total number of holes as test holes for construction, and drill to a depth slightly deeper than the designed hole depth. In practice, based on the detailed geological survey report before construction, test boreholes are selected at the designed anchor bolt group locations following the principle of "controlled placement": First, key locations are prioritized, and test boreholes should cover at least each different key geological unit and the inferred unfavorable geological area. Key geological units include, but are not limited to, strongly weathered layers, moderately weathered layers, and fault-affected zones, while unfavorable geological areas include, but are not limited to, densely jointed zones and weak interlayers. Second, the quantity requirement is met. The total number of points selected according to the above principles should not be less than 10% of the total number of designed boreholes. If this is insufficient, additional boreholes are placed in areas with high geological uncertainty to reach the minimum proportion.

[0022] S2. Geological Information Acquisition: During the drilling of the test hole, geological information is collected in real time through the data acquisition module, including real-time drilling parameters, image information of the rock mass structure of the borehole wall, and acoustic velocity data. S3. Intelligent decision-making for construction parameters: The geological information of the current borehole location collected in step S2 is input into the pre-trained neural network model in the intelligent decision-making module for real-time analysis and calculation, and the recommended set of construction parameters is dynamically output and sent to the intelligent execution module. S4. Intelligent Execution of Construction Process: The intelligent execution module executes construction operations based on the received set of recommended construction parameters. The intelligent drilling rig completes drilling and reaming according to the recommended parameters for hole geometry and drilling process; the UHPC automatic batching equipment prepares UHPC material according to the recommended parameters for UHPC material properties; and the automatic grouting equipment completes grouting and anchor bolt installation according to the recommended parameters for grouting process. S5. Construction effect monitoring: Collect construction quality data during and after construction, including the axial displacement of anchor bolts and the internal density and uniformity of grout. S6. Feedback Optimization: The quality feedback module transmits the collected construction quality data to the intelligent decision-making module in real time and performs feedback optimization and iteration on the neural network model. Specifically, the feedback optimization and iteration of the neural network model includes the following steps: S601, Parameter Fine-tuning: For the current construction batch, if the monitoring data deviates from the expected range, the system will automatically fine-tune the recommended set of construction parameters for subsequent boreholes online based on the preset expert rule base; S602, Incremental Model Learning: After a construction project is completed, all successful cases are added to the historical database. The intelligent decision-making module uses this expanded database to incrementally train the neural network model after each project is completed, updating its weight parameters to achieve continuous iterative optimization of model performance. S603, Human-Machine Interaction Verification: The parameter optimization suggestions provided by the system must be confirmed by the on-site engineer before implementation to ensure that the entire process is under control.

[0023] The above describes one or more embodiments of the present invention in a relatively specific and detailed manner, but it should not be construed as limiting the scope of the present invention. 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 patent should be determined by the appended claims.

Claims

1. A UHPC borehole enlargement and pull-out anchor bolt intelligent system for penetrating deep into rock strata, characterized in that, It includes a data acquisition module, an intelligent decision-making module, an intelligent execution module, and a quality feedback module; The data acquisition module is integrated into the drilling equipment and includes a measurement-while-drilling device, a borehole imager, and a micro acoustic testing device. The measurement-while-drilling device is used to acquire real-time drilling parameters, including drilling speed, drill bit torque, axial thrust, and rotational speed. The borehole imager is used to acquire rock mass structure image information of the borehole wall. The micro acoustic testing device is used to perform acoustic testing at a specific depth inside the borehole to acquire acoustic velocity data. The intelligent decision-making module includes a computing server and is equipped with a pre-trained neural network model. The neural network model takes real-time drilling parameters, rock mass structure image information, and acoustic velocity data acquired by the data acquisition module as input, and outputs a set of recommended construction parameters for the current construction area through model calculation. The set of recommended construction parameters includes recommended parameters for hole geometry, recommended parameters for drilling process, recommended parameters for UHPC material properties, and recommended parameters for grouting process. The recommended geometric parameters for hole formation include hole depth, diameter of the enlarged section, and anchor bolt spacing; the recommended drilling process parameters include drilling speed and axial thrust. The recommended performance parameters for UHPC materials include target slump spread, water-cement ratio, and fiber volume fraction; the recommended parameters for grouting process include grouting pressure and grouting flow rate. The intelligent execution module receives instructions from the intelligent decision-making module and executes corresponding construction operations, including an intelligent drilling rig, an automatic grouting device, and a UHPC automatic batching device. The intelligent drilling rig automatically adjusts and controls the drilling speed and axial thrust based on the recommended hole geometry parameters and drilling process parameters issued by the intelligent decision-making module, achieving the hole depth, enlarged section diameter, and anchor bolt spacing specified by the recommended hole geometry parameters. The automatic grouting device automatically adjusts and controls the grouting pressure and grouting flow rate based on the recommended grouting process parameters issued by the intelligent decision-making module. The UHPC automatic batching device automatically completes the precise metering, feeding, and mixing of materials based on the recommended UHPC material performance parameters issued by the intelligent decision-making module, preparing UHPC with specified target slump spread, water-cement ratio, and fiber volume fraction. The quality feedback module is used to collect construction quality data during and after construction, including anchor displacement sensors and grout ultrasonic testing instruments. The anchor displacement sensors are installed on the anchor body and are used to continuously monitor the axial displacement of the anchor during anchor tensioning and service. The grout ultrasonic testing instrument is used to perform non-destructive scanning of the anchoring section after the grout has solidified. The quality feedback module transmits various types of collected data to the intelligent decision-making module in real time, providing data support for the online optimization and iteration of construction parameters.

2. The UHPC borehole enlargement and pull-out anchor bolt intelligent system for penetrating deep rock strata according to claim 1, characterized in that, The core of the intelligent decision-making module lies in a specially designed and trained neural network model, which consists of an input layer, a hidden layer, and an input layer connected in sequence. The model's input layer is used to receive data and has a total of 11 nodes, including: real-time drilling depth, drilling speed, drill bit torque, axial thrust and rotation speed, joint and fracture density, average fracture width, dominant fracture dip angle and borehole roughness coefficient extracted from rock mass image information, and acoustic velocity obtained through acoustic testing. The model employs a dual-hidden-layer structure, where the first hidden layer contains 128 neurons and uses a linear rectified function as the activation function; the second hidden layer contains 64 neurons and also uses a linear rectified function as the activation function. The output layer of the model corresponds to the recommended set of construction parameters generated by intelligent decision-making. It contains 10 nodes and uses a linear activation function to directly output specific parameter values, including hole depth, diameter of the enlarged section, anchor spacing, drilling speed, axial thrust, target slump expansion, water-cement ratio, fiber volume content, grouting pressure, and grouting flow rate.

3. The UHPC borehole enlargement and pull-out anchor bolt intelligent system for penetrating deep rock strata according to claim 1, characterized in that, The training of the neural network model relies on a historical engineering database, which is constructed by collecting a large number of complete and high-quality anchor bolt projects. This database includes: detailed construction logs, borehole geological records, drilling measurement data, borehole imaging data, acoustic testing data, and final anchor bolt pull-out strength test reports and grout quality non-destructive testing reports. Each data record is reconstructed to form sample pairs. Each sample pair includes an input vector and a label vector. The input vector contains multi-dimensional geological information recorded during the construction process of the borehole, corresponding to the features covered by the aforementioned real-time drilling parameters, rock structure image information, and acoustic velocity data. The label vector contains the construction parameter combination that was ultimately verified as successful at the borehole, achieving the designed pull-out strength, corresponding to the parameters covered by the aforementioned recommended set of construction parameters. The initial training set contains at least 5000 valid sample pairs and is continuously expanded using successful cases generated in practical applications of this invention's system. The neural network model training algorithm employs stochastic gradient descent with momentum, combined with backpropagation to calculate the gradient. The loss function uses mean squared error to minimize the gap between the model's predicted parameters and the actual successful parameters. The dataset is randomly divided into training and test sets, with 80% to 20% of the dataset allocated to each set. During training, the model's performance is validated on the test set after each training epoch. Training is terminated early when the model's loss function value on the test set no longer decreases for 10 consecutive training epochs to prevent overfitting.

4. A construction method for a UHPC borehole enlargement and pull-out anchor system for deep-penetrating rock formations, applied to the UHPC borehole enlargement and pull-out anchor system for deep-penetrating rock formations as described in claim 1, characterized in that, Includes the following steps: S1. Construction of test holes: At the location of the designed anchor group, select no less than 10% of the total number of holes as test holes for construction, and drill to a depth slightly deeper than the designed hole depth. S2. Geological Information Acquisition: During the drilling of the test hole, geological information is collected in real time through the data acquisition module, including real-time drilling parameters, image information of the rock mass structure of the borehole wall, and acoustic velocity data. S3. Intelligent decision-making for construction parameters: The geological information of the current borehole location collected in step S2 is input into the pre-trained neural network model in the intelligent decision-making module for real-time analysis and calculation, and the recommended set of construction parameters is dynamically output and sent to the intelligent execution module. S4. Intelligent Execution of Construction Process: The intelligent execution module executes construction operations based on the received set of recommended construction parameters. The intelligent drilling rig completes drilling and reaming according to the recommended parameters for hole geometry and drilling process; the UHPC automatic batching equipment prepares UHPC material according to the recommended parameters for UHPC material properties; and the automatic grouting equipment completes grouting and anchor bolt installation according to the recommended parameters for grouting process. S5. Construction effect monitoring: Collect construction quality data during and after construction, including the axial displacement of anchor bolts and the internal density and uniformity of grout. S6. Feedback Optimization: The quality feedback module transmits the collected construction quality data to the intelligent decision-making module in real time, and performs feedback optimization and iteration on the neural network model.

Images