A pile machine construction whole process intelligent monitoring method based on digital twinning

Abstract

The application discloses a kind of based on digital twinning pile machine construction whole process intelligent monitoring method, comprising: obtaining pile machine construction whole process data, constructs initial digital twin;Data depth segmentation generates construction memory atom, constructs heterogeneous construction memory graph, executes cross pile memory projection and forms memory field;Extract abnormal construction memory atom from memory field, construct topological graph based on space-time and abnormal association and generate abnormal topological ridge line;Determine the intersection relationship of current pile construction path and abnormal topological ridge line, update stratum and risk state in digital twin;Quality state vector is constructed and damage threshold surface is established, whether to enter irreversible damage critical state is judged and control strategy is generated;Perform recovery action to adjust construction parameter, generate quality locking result and backfill memory field to realize model continuous update.The present application realizes pile machine construction risk early identification and quality active control through pile group construction memory field and abnormal topological ridge line.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to an intelligent monitoring method for the entire process of pile driver construction based on digital twins. Background Technology

[0002] Piling is a crucial step in pile foundation engineering, involving multiple continuous processes such as drilling, drill lifting, grouting, re-mixing, and pile quality control. With the development of intelligent construction and the digitalization of engineering machinery, existing technologies have emerged that utilize sensors, industrial gateways, positioning systems, and construction management platforms to monitor the piling process. These systems can collect construction parameters such as drilling depth, drilling speed, drill rod rotation speed, power head torque, grouting pressure, grouting flow rate, and piling machine posture. Furthermore, digital twin models can be used to visualize the piling machine's operating status and the pile construction process.

[0003] However, existing methods for monitoring the entire pile driving process typically focus on individual piles, primarily relying on real-time construction parameters for threshold judgments or anomaly alarms. Construction data from completed piles is mostly used only to generate construction records or quality reports, making it difficult to further transform it into usable construction experience data for subsequent pile construction, and also hindering the uncovering of implicit correlations between different pile locations, depths, construction responses, anomaly types, and quality results. Existing digital twin models still largely depend on geological survey data and current real-time data before pile construction, failing to fully utilize the underground state information revealed by already constructed piles.

[0004] Underground anomalies such as grout leakage sensitive zones, hard interlayers, weak interlayers, and unstable borehole walls during pile driving often exhibit spatial continuity. However, current technologies mostly focus on identifying single-point anomalies or issuing alarms for single-pile parameters. This makes it difficult to reconstruct continuous underground anomalies in advance or to determine whether the current pile's future construction path will intersect with anomalies. Existing early warning methods typically only trigger after construction parameters exceed limits or quality defects form, lacking a mechanism to identify whether the quality status is approaching the boundary of irreversible damage. This leads to delayed construction adjustments and makes it difficult to promptly prevent irreversible quality defects such as diameter reduction, pile breakage, mud inclusion, grout leakage expansion, or pile discontinuity.

[0005] Therefore, how to provide an intelligent monitoring method for the entire process of pile driver construction based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent monitoring method for the entire pile driving construction process based on digital twins. This invention utilizes pile group construction memory field, improved metapath2vec cross-pile memory projection, underground abnormal topological ridge reconstruction, and reversible-irreversible quality damage threshold determination technology to transform the depth segmentation construction data of completed piles into reusable prior information on construction risks for the current pile and subsequent piles. During the construction of the current pile, it identifies underground continuous abnormal structures such as grout leakage sensitive zones, hard interlayers, and weak interlayers in advance, determines whether the construction state is close to the irreversible quality damage boundary, and then generates a set of prohibited construction actions and a minimum recovery action sequence. It has the advantages of strong data reuse capability of constructed piles, high advance recognition of underground abnormal structures, accurate dynamic calibration of digital twin models, timely construction risk warning, and strong targeted control of pile quality.

[0007] According to an embodiment of the present invention, a method for intelligent monitoring of the entire process of pile driver construction based on digital twin includes: Acquire data from the entire piling machine construction process to construct an initial digital twin; The data of the entire pile driver construction process is processed by depth segmentation, and construction memory atoms of corresponding depth segments are generated and a heterogeneous construction memory map is constructed. The improved metapath2vec model is used to perform cross-pile memory projection to obtain the prior state of construction risk of the current unconstructed depth segment and construct the pile group construction memory field. Abnormal construction memory atoms are extracted from the pile group construction memory field, and an underground abnormal topology map is constructed based on the abnormal construction memory atoms. The underground abnormal topology map is then connected and fitted to generate underground abnormal topology ridges. The current pile construction path is intersected with the underground abnormal topological ridge line. When the current pile construction path intersects with the underground abnormal topological ridge line, the stratum status, construction risk status and pile quality status in the current pile digital twin are updated. Based on the updated digital twin of the current pile, construct the quality state vector of the current depth segment, establish the reversible-irreversible quality damage threshold surface, determine whether the current construction state is in the irreversible damage critical state, and when the determination result is in the irreversible damage critical state, generate a set of prohibited construction actions and a minimum recovery action sequence. The construction control parameters during the current pile construction process are adjusted according to the minimum recovery action sequence. After the construction of the current depth section is completed, a quality lock result is generated. The quality lock result is converted into new construction memory atoms and reinjected into the pile group construction memory field to update the digital twin.

[0008] Optionally, the data for the entire piling process includes pile location coordinates, construction depth, drilling speed, drill lifting speed, drill rod rotation speed, power head torque, hydraulic pressure, grouting pressure, grouting flow rate, piling machine posture, drill rod vibration, construction time, construction process stages, and pile quality results.

[0009] Optionally, constructing the initial digital twin includes: By analyzing engineering design data, the pile location coordinates, design pile diameter, design pile length, and pile spacing are obtained, and a spatial twin of the pile location is established. By analyzing geological exploration data, stratigraphic types, stratigraphic thickness, stratigraphic strength, and groundwater conditions at different depths are obtained, and stratigraphic condition twins are established. By analyzing the parameters of the piling machine equipment, the parameters of the power head, drill rod, hydraulic system, and grouting system are obtained, and a twin of the piling machine equipment is established. By analyzing the construction process parameters, drilling parameters, drill lifting parameters, grouting parameters, and re-mixing parameters are obtained, and a construction process twin is established. Based on the design pile diameter, design pile length, pile verticality requirements, and grouting filling requirements, a pile quality twin is established.

[0010] Optionally, obtaining the prior state of construction risk for the currently unconstructed depth segment of the pile and constructing a pile group construction memory field includes: The construction process data of the completed piles is segmented according to the preset depth interval. The pile location identifier, depth segment identifier, construction response characteristics, stratum state characteristics, anomaly type characteristics and quality result characteristics corresponding to each depth segment are extracted and encapsulated into construction memory atoms. A heterogeneous construction memory graph is constructed based on construction memory atoms. The heterogeneous construction memory graph includes pile location nodes, depth segment nodes, construction response nodes, formation state nodes, anomaly type nodes, and quality result nodes. Spatial subordination edges are established between pile location nodes and depth segment nodes, response mapping edges are established between depth segment nodes and construction response nodes, anomaly triggering edges are established between construction response nodes and anomaly type nodes, quality influence edges are established between anomaly type nodes and quality result nodes, and formation correspondence edges are established between depth segment nodes and formation state nodes. An improved metapath2vec model is constructed, which includes a construction semantic metapath generation layer, a cross-pile memory projection layer, and a risk prior aggregation layer connected in sequence. In the construction semantic meta-path generation layer, three types of directed meta-paths are set. The first type of directed meta-path is the stratum migration meta-path composed of pile location node, depth segment node, stratum state node, depth segment node, and pile location node. The second type of directed meta-path is the abnormal quality meta-path composed of pile location node, depth segment node, construction response node, anomaly type node, and quality result node. The third type of directed meta-path is the defect backtracking meta-path composed of depth segment node, anomaly type node, quality result node, anomaly type node, and depth segment node. In the cross-pile memory projection layer, the current unconstructed depth segment of the pile is used as a query node and connected to the heterogeneous construction memory map. Based on the spatial distance of the pile location, the distance of the depth segment, the consistency of the directed element path matching result and the quality result, projectable memory atoms are selected from the construction memory atoms of the completed piles. The stratum state, anomaly type and quality result corresponding to the projectable memory atoms are projected to the query node to obtain the prior state of construction risk of the current unconstructed depth segment of the pile. In the risk prior aggregation layer, the construction risk prior state of the current unconstructed depth segment of the pile, the corresponding projectable memory atom, the spatial location of the pile, the depth segment location, and the memory projection source are written into the same spatial index structure to form a pile group construction memory field.

[0011] Optionally, generating subsurface anomalous topological ridges includes: Read construction memory atoms, the prior state of construction risk of the current unconstructed depth section of the pile, the spatial location of the pile, the location of the depth section and the source of memory projection from the construction memory field of the pile group; extract abnormal construction memory atoms based on construction response abrupt changes, abnormal stratum state, abnormality type markers and abnormal quality results. Abnormal construction memory atoms are converted into underground abnormal nodes. Each underground abnormal node contains pile location identifier, depth segment identifier, spatial location, construction response characteristics, abnormality type, quality result and construction risk prior state. Candidate topology connections are established based on the spatial proximity relationship, depth continuity relationship, abnormality type association relationship and quality result association relationship between underground abnormal nodes. Candidate topological connections are screened for multi-relation consistency, and candidate topological connections that simultaneously satisfy spatial proximity, depth continuity, anomaly association, and quality association are retained. The retained underground anomaly nodes and topological connections are combined to form an underground anomaly topological map. A hierarchical connectivity search is performed on the underground anomaly topology map. First, underground anomaly nodes with the same anomaly type are connected within the same depth layer to form a horizontal anomaly segment. Then, horizontal anomaly segments with the same anomaly evolution direction are connected between adjacent depth layers to form a vertical anomaly chain. Finally, the horizontal anomaly segments and the vertical anomaly chain are merged into an underground anomaly structural unit. Based on the spatial arrangement order, depth change direction, anomaly evolution direction, and consistency of quality results of underground anomaly nodes in the underground anomaly structural unit, a central path representing the continuous direction of underground anomalies is extracted. The central path is then bound to the corresponding anomaly type, influence depth range, influence pile location range, anomaly intensity level, and quality correlation results to generate an underground anomaly topological ridge.

[0012] Optionally, updating the current pile digital twin's geological state, construction risk state, and pile quality state includes: The current pile construction path is generated based on the pile location coordinates, designed pile length, current construction depth and construction direction, and the current pile construction path is mapped to a digital twin space containing underground anomaly topological ridges. Read the anomaly type, affected depth range, affected pile location range, anomaly strength level and quality correlation results of the underground anomaly topological ridge line, and match the spatial coverage of the underground anomaly topological ridge line with the current pile construction path; When the current pile construction path enters the affected pile location range of the underground abnormal topological ridge and the depth interval corresponding to the current pile construction path falls into the affected depth range of the underground abnormal topological ridge, it is determined that the current pile construction path and the underground abnormal topological ridge have an intersection relationship, and the intersection depth segment, intersection anomaly type and intersection anomaly intensity level are recorded. Based on the intersection depth segment, intersection anomaly type, and intersection anomaly intensity level, the stratum state of the corresponding depth segment in the current pile digital twin is updated to the stratum state associated with the underground anomaly topological ridge line, and the construction risk state of the corresponding depth segment in the current pile digital twin is updated to the construction risk state associated with the intersection anomaly type. Update the pile quality status of the corresponding depth segment in the current pile digital twin based on the quality correlation results, and write the updated stratum status, construction risk status and pile quality status into the current pile digital twin.

[0013] Optionally, generating the set of prohibited construction actions and the minimum recovery action sequence includes: Read the current geological conditions, construction risk conditions, pile quality conditions, underground abnormal topological ridge intersection results and current construction control parameters from the updated current pile digital twin, and generate the current depth segment's quality state vector, which includes borehole wall stability, grouting filling, verticality, pile-soil disturbance, defect accumulation, and pile integrity. A quality status sample library is established based on the quality locking results of completed piles. Depth segments with qualified quality results and no triggering of recovery actions are marked as normal samples. Depth segments with qualified quality results after triggering recovery actions are marked as reversible abnormal samples. Depth segments with quality results of reduced diameter, broken piles, mud inclusion, hole wall collapse, grout leakage expansion, or pile discontinuity are marked as irreversible damage samples. The quality status sample library is grouped according to the stratum state, anomaly type and construction process stage. Within each group, the state boundary record between reversible anomaly samples and irreversible damage samples is extracted, and the state boundary record is mapped to the corresponding depth segment of the current pile digital twin to form an initial reversible-irreversible quality damage threshold surface. Based on the intersection depth, intersection anomaly type, and anomaly intensity level of the current pile construction path and the underground abnormal topological ridge, the initial reversible-irreversible quality damage threshold surface is corrected by ridge line disturbance. The depth segment within the influence range of the grout leakage sensitive ridge, hard interlayer ridge, soft interlayer ridge, or borehole wall instability ridge is set as the ridge line constraint depth segment. The corrected reversible-irreversible quality damage threshold surface is generated within the ridge line constraint depth segment. The current depth segment's quality state vector is compared with the corrected reversible-irreversible quality damage threshold surface to determine the state approximation. When the current quality state changes from a normal state to a reversible abnormal state and continues to evolve toward an irreversible damage state, the current construction state is determined to be in an irreversible damage critical state. When the current construction state is in an irreversible damage critical state, a set of prohibited construction actions and a sequence of minimum recovery actions are generated based on the current construction control parameters, intersection anomaly type and quality state vector. The set of prohibited construction actions includes construction actions that cause the quality state to cross the corrected reversible-irreversible quality damage threshold, and the sequence of minimum recovery actions includes construction actions that return the quality state to a reversible abnormal state or a normal state.

[0014] Optionally, generating a quality locking result after construction at the current depth segment is completed includes: The minimum recovery action sequence is parsed into ordered construction control instructions. The construction control parameters in the current pile construction process are adjusted according to the ordered construction control instructions. The construction control parameters include drilling speed, drill lifting speed, drill rod rotation speed, grouting pressure, grouting flow rate, drilling stop and pressure stabilization time, and re-stirring parameters. After adjusting the construction control parameters, the construction response data and quality status data of the current depth section are collected and written into the current pile digital twin to update the stratum status, construction risk status and pile quality status of the current depth section. After the construction at the current depth is completed, a quality lock result is generated. The quality lock result includes pile location identification, depth segment identification, construction response characteristics, underground abnormal topological ridge intersection results, prohibited construction action set, minimum recovery action sequence, adjusted construction control parameters, and final quality status. The quality locking result is converted into a new construction memory atom, and the new construction memory atom is fed back into the pile group construction memory field. This updates the cross-pile memory projection relationship between the completed piles and the current pile in the pile group construction memory field, and synchronizes the updated pile group construction memory field to the digital twin.

[0015] The beneficial effects of this invention are: This invention constructs a pile group construction memory field by deeply segmenting the entire construction process data of completed piles and generating construction memory atoms. This transforms completed piles from mere construction records or quality reports into readily accessible prior information on construction risks for the current and subsequent pile construction processes. By improving the metapath2vec model to perform cross-pile memory projection, the implicit correlations between pile location, depth segment, construction response, anomaly type, and quality results can be uncovered. This enhances the reusability of constructed pile data, allowing the digital twin to obtain prior support that more closely approximates the actual geological conditions before the current pile construction begins.

[0016] This invention extracts anomalous construction memory atoms from the pile group construction memory field and constructs an underground anomaly topology map based on the spatiotemporal continuity, anomaly correlation, and quality correlation of these atoms. This generates underground anomaly topological ridges, enabling early identification of spatially continuous underground anomalies such as grout leakage sensitive zones, hard interlayers, weak interlayers, and borehole wall instability zones. Compared to existing technologies that rely on local alarms based on real-time parameters of individual piles, this invention can determine the intersection between the current pile construction path and the underground anomaly topological ridges, and update the geological state, construction risk state, and pile quality state in the current pile digital twin in advance, improving the predictability and accuracy of construction risk identification.

[0017] This invention constructs a quality state vector for the current depth segment and a reversible-irreversible quality damage threshold surface. This allows the system to determine whether the current construction state is at the critical point of irreversible damage before irreversible quality defects form, and generates a set of prohibited construction actions and a minimum recovery action sequence. Therefore, the system no longer only issues alarms after parameters exceed limits or quality defects form, but can proactively block and adjust construction parameters before quality problems such as diameter reduction, pile breakage, mud inclusion, grout leakage expansion, or pile discontinuity occur, improving the targeting, timeliness, and traceability of pile quality control. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an intelligent monitoring method for the entire piling machine construction process based on digital twins, as proposed in this invention. Figure 2 This is a schematic diagram illustrating the process of performing cross-pile memory projection and constructing a pile group construction memory field using the improved metapath2vec model of the intelligent monitoring method for the entire pile driver construction process based on digital twins proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2 A digital twin-based intelligent monitoring method for the entire piling machine construction process includes: Acquire data from the entire piling machine construction process to construct an initial digital twin; The data of the entire pile driver construction process is processed by depth segmentation, and construction memory atoms of corresponding depth segments are generated and a heterogeneous construction memory map is constructed. The improved metapath2vec model is used to perform cross-pile memory projection to obtain the prior state of construction risk of the current unconstructed depth segment and construct the pile group construction memory field. Abnormal construction memory atoms are extracted from the pile group construction memory field, and an underground abnormal topology map is constructed based on the abnormal construction memory atoms. The underground abnormal topology map is then connected and fitted to generate underground abnormal topology ridges. The current pile construction path is intersected with the underground abnormal topological ridge line. When the current pile construction path intersects with the underground abnormal topological ridge line, the stratum status, construction risk status and pile quality status in the current pile digital twin are updated. Based on the updated digital twin of the current pile, construct the quality state vector of the current depth segment, establish the reversible-irreversible quality damage threshold surface, determine whether the current construction state is in the irreversible damage critical state, and when the determination result is in the irreversible damage critical state, generate a set of prohibited construction actions and a minimum recovery action sequence. The construction control parameters during the current pile construction process are adjusted according to the minimum recovery action sequence. After the construction of the current depth section is completed, a quality lock result is generated. The quality lock result is converted into new construction memory atoms and reinjected into the pile group construction memory field to update the digital twin.

[0021] In this embodiment, the data for the entire piling process includes pile location coordinates, construction depth, drilling speed, drill lifting speed, drill rod rotation speed, power head torque, hydraulic pressure, grouting pressure, grouting flow rate, piling machine posture, drill rod vibration, construction time, construction process stage data, and pile quality results data.

[0022] In this embodiment, constructing the initial digital twin includes: By analyzing engineering design data, the pile location coordinates, design pile diameter, design pile length, and pile spacing are obtained, and a spatial twin of the pile location is established. By analyzing geological exploration data, stratigraphic types, stratigraphic thickness, stratigraphic strength, and groundwater conditions at different depths are obtained, and stratigraphic condition twins are established. By analyzing the parameters of the piling machine equipment, the parameters of the power head, drill rod, hydraulic system, and grouting system are obtained, and a twin of the piling machine equipment is established. By analyzing the construction process parameters, drilling parameters, drill lifting parameters, grouting parameters, and re-mixing parameters are obtained, and a construction process twin is established. Based on the design pile diameter, design pile length, pile verticality requirements, and grouting filling requirements, a pile quality twin is established.

[0023] In this embodiment, obtaining the prior state of construction risk for the current unconstructed depth section of the pile and constructing a pile group construction memory field includes: The construction process data of the completed piles is segmented according to the preset depth interval. The pile location identifier, depth segment identifier, construction response characteristics, stratum state characteristics, anomaly type characteristics and quality result characteristics corresponding to each depth segment are extracted and encapsulated into construction memory atoms. The preset depth interval is 0.2m. Encapsulated as construction memory atoms, specifically: Read the pile location identifier, depth range start and end values, average drilling speed, average drill lifting speed, average drill rod rotation speed, average power head torque, average hydraulic pressure, average grouting pressure, cumulative grouting flow rate, maximum pile driver inclination angle, and mean square value of drill rod vibration within the same depth range. Arrange the construction response data in a fixed field order to generate a construction response feature sequence. Read the stratum type, stratum strength grade, and borehole stability grade corresponding to the depth range to generate a stratum state feature sequence. Read the anomaly type marker and quality result marker corresponding to the depth range to generate an anomaly quality feature sequence. Combine the pile location identifier, depth range identifier, construction response feature sequence, stratum state feature sequence, and anomaly quality feature sequence in the field order, assign a unique memory atom number, and generate a construction memory atom corresponding to the depth range. A heterogeneous construction memory graph is constructed based on construction memory atoms. The heterogeneous construction memory graph includes pile location nodes, depth segment nodes, construction response nodes, formation state nodes, anomaly type nodes, and quality result nodes. Spatial subordination edges are established between pile location nodes and depth segment nodes, response mapping edges are established between depth segment nodes and construction response nodes, anomaly triggering edges are established between construction response nodes and anomaly type nodes, quality influence edges are established between anomaly type nodes and quality result nodes, and formation correspondence edges are established between depth segment nodes and formation state nodes. An improved metapath2vec model is constructed, which includes a construction semantic metapath generation layer, a cross-pile memory projection layer, and a risk prior aggregation layer connected in sequence. The improved metapath2vec model is constructed as follows: The metapath2vec model retains the metapath random walk module and skip-gram training module, and improves the traditional fixed metapath input method into a construction semantic metapath generation layer. The construction semantic metapath generation layer automatically generates the stratum migration metapath, abnormal quality metapath, and defect backtracking metapath based on the pile location node, depth segment node, construction response node, stratum state node, anomaly type node, and quality result node. The traditional random walk module is improved into a cross-pile memory projection layer. After obtaining the node context relationship through random walk, the current pile unconstructed depth segment query node is connected. Based on the consistency of pile location spatial distance, depth segment distance, metapath matching result, and quality result, the improved metapath2vec model is used to perform cross-pile memory projection to obtain the construction risk prior state of the current pile unconstructed depth segment and construct the pile group construction memory field. Abnormal construction memory atoms are extracted from the pile group construction memory field, and an underground abnormal topology map is constructed based on the abnormal construction memory atoms. The underground abnormal topology map is then connected and fitted to generate underground abnormal topology ridges. The current pile construction path is intersected with the underground abnormal topological ridge line. When the current pile construction path intersects with the underground abnormal topological ridge line, the stratum status, construction risk status and pile quality status in the current pile digital twin are updated. Based on the updated digital twin of the current pile, construct the quality state vector of the current depth segment, establish the reversible-irreversible quality damage threshold surface, determine whether the current construction state is in the irreversible damage critical state, and when the determination result is in the irreversible damage critical state, generate a set of prohibited construction actions and a minimum recovery action sequence. The construction control parameters during the current pile construction process are adjusted according to the minimum recovery action sequence. After the construction of the current depth section is completed, a quality lock result is generated. The quality lock result is converted into new construction memory atoms and reinjected into the pile group construction memory field to update the digital twin.

[0024] In this embodiment, the data for the entire piling process includes pile location coordinates, construction depth, drilling speed, drill lifting speed, drill rod rotation speed, power head torque, hydraulic pressure, grouting pressure, grouting flow rate, piling machine posture, drill rod vibration, construction time, construction process stage data, and pile quality results data.

[0025] In this embodiment, constructing the initial digital twin includes: By analyzing engineering design data, the pile location coordinates, design pile diameter, design pile length, and pile spacing are obtained, and a spatial twin of the pile location is established. By analyzing geological exploration data, stratigraphic types, stratigraphic thickness, stratigraphic strength, and groundwater conditions at different depths are obtained, and stratigraphic condition twins are established. By analyzing the parameters of the piling machine equipment, the parameters of the power head, drill rod, hydraulic system, and grouting system are obtained, and a twin of the piling machine equipment is established. By analyzing the construction process parameters, drilling parameters, drill lifting parameters, grouting parameters, and re-mixing parameters are obtained, and a construction process twin is established. Based on the design pile diameter, design pile length, pile verticality requirements, and grouting filling requirements, a pile quality twin is established.

[0026] In this embodiment, obtaining the prior state of construction risk for the current unconstructed depth section of the pile and constructing a pile group construction memory field includes: The construction process data of the completed piles is segmented according to the preset depth interval. The pile location identifier, depth segment identifier, construction response characteristics, stratum state characteristics, anomaly type characteristics and quality result characteristics corresponding to each depth segment are extracted and encapsulated into construction memory atoms. The preset depth interval is 0.2m. Encapsulated as construction memory atoms, specifically: Read the pile location identifier, depth range start and end values, average drilling speed, average drill lifting speed, average drill rod rotation speed, average power head torque, average hydraulic pressure, average grouting pressure, cumulative grouting flow rate, maximum pile driver inclination angle, and mean square value of drill rod vibration within the same depth range. Arrange the construction response data in a fixed field order to generate a construction response feature sequence. Read the stratum type, stratum strength grade, and borehole stability grade corresponding to the depth range to generate a stratum state feature sequence. Read the anomaly type marker and quality result marker corresponding to the depth range to generate an anomaly quality feature sequence. Combine the pile location identifier, depth range identifier, construction response feature sequence, stratum state feature sequence, and anomaly quality feature sequence in the field order, assign a unique memory atom number, and generate a construction memory atom corresponding to the depth range. A heterogeneous construction memory graph is constructed based on construction memory atoms. The heterogeneous construction memory graph includes pile location nodes, depth segment nodes, construction response nodes, formation state nodes, anomaly type nodes, and quality result nodes. Spatial subordination edges are established between pile location nodes and depth segment nodes, response mapping edges are established between depth segment nodes and construction response nodes, anomaly triggering edges are established between construction response nodes and anomaly type nodes, quality influence edges are established between anomaly type nodes and quality result nodes, and formation correspondence edges are established between depth segment nodes and formation state nodes. An improved metapath2vec model is constructed, which includes a construction semantic metapath generation layer, a cross-pile memory projection layer, and a risk prior aggregation layer connected in sequence. The improved metapath2vec model is constructed as follows: The metapath2vec model retains the metapath random walk module and skip-gram training module, but improves the traditional fixed metapath input method to a construction semantic metapath generation layer. This layer automatically generates stratum migration metapaths, abnormal quality metapaths, and defect backtracking metapaths based on pile location nodes, depth segment nodes, construction response nodes, stratum state nodes, anomaly type nodes, and quality result nodes. The traditional random walk module is also improved to a cross-pile memory projection layer, which, after obtaining node context relationships through random walks, accesses the query for the current pile's unconstructed depth segment. Nodes are selected for projectionable memory atoms based on pile location spatial distance, depth segment distance, metapath matching results, and consistency of quality results. The stratum state, anomaly type, and quality results of completed piles are projected to the query node. A risk prior aggregation layer is added after the traditional skip-gram training module. The risk prior aggregation layer receives the projection results output by the cross-pile memory projection layer, performs main projection determination, companion projection retention, confidence level marking, and spatial index writing for multiple projection results of the same query node, and finally outputs the construction risk prior state of the current unconstructed depth segment of the pile and forms the pile group construction memory field. In the construction semantic meta-path generation layer, three types of directed meta-paths are set. The first type of directed meta-path is the stratum migration meta-path composed of pile location node, depth segment node, stratum state node, depth segment node, and pile location node. The second type of directed meta-path is the abnormal quality meta-path composed of pile location node, depth segment node, construction response node, anomaly type node, and quality result node. The third type of directed meta-path is the defect backtracking meta-path composed of depth segment node, anomaly type node, quality result node, anomaly type node, and depth segment node. In the cross-pile memory projection layer, the current unconstructed depth segment of the pile is used as a query node and connected to the heterogeneous construction memory map. Based on the spatial distance of the pile location, the distance of the depth segment, the consistency of the directed element path matching result and the quality result, projectable memory atoms are selected from the construction memory atoms of the completed piles. The stratum state, anomaly type and quality result corresponding to the projectable memory atoms are projected to the query node to obtain the prior state of construction risk of the current unconstructed depth segment of the pile. In the risk prior aggregation layer, the construction risk prior state of the current unconstructed depth segment of the pile, the corresponding projectable memory atom, the spatial location of the pile, the depth segment location, and the memory projection source are written into the same spatial index structure to form a pile group construction memory field. Specifically, the formation of the pile group construction memory field is as follows: Read the horizontal and vertical coordinates of the current pile location, divide them by 1m and round down to obtain the horizontal and vertical index numbers. Read the start and end points of the current pile's unconstructed depth segment, divide them by 0.2m and round down to obtain the depth start and depth end index numbers. Concatenate the horizontal, vertical, depth start, and depth end index numbers in the order of horizontal, vertical, depth start, and depth end to form a spatial index key. Write the following field groups into the index cells corresponding to the spatial index key: Construction Risk Prior State Field, Main Projection Memory Atom Number Field, Accompanying Projection Memory Atom Number List Field, Projection Confidence Level Field, Source Pile Location Number Field, and Source Depth Segment Number Field. The system includes the current pile identifier field and the timestamp field. If the index unit does not exist, it is created; if it already exists, the projection confidence level is compared, and the record with the higher confidence level is retained while the remaining records are appended to the accompanying list. Multiple index units belonging to the same current pile with consecutive depth index numbers and consistent primary risk prior states are merged, with the depth start index number set to the minimum value and the depth end index number set to the maximum value. The continuous depth segment length field is updated. All index units are sorted according to the horizontal index number, vertical index number, and depth start index number and written to the main table of the pile group construction memory field. A fast retrieval index file is generated with the spatial index key as the primary key and the primary projection memory atom number as the secondary key, thus completing the establishment of the pile group construction memory field. The construction semantic meta-path generation layer includes: Node field parsing unit: Reads the pile location nodes, depth segment nodes, construction response nodes, stratum status nodes, anomaly type nodes, and quality result nodes in the heterogeneous construction memory diagram, and reads the node number, node type, and node field value of each node. Construction field coding unit: Convert the average drilling speed, average drill lifting speed, average drill rod rotation speed, average power head torque, average hydraulic pressure, average grouting pressure, cumulative grouting flow rate, maximum pile driver inclination angle, and mean square value of drill rod vibration in the construction response node into construction response codes; convert the stratum type, stratum strength grade, and borehole wall stability grade in the stratum state node into stratum state codes; convert the abnormal grout leakage, high resistance of hard interlayer, disturbance of soft interlayer, borehole wall instability, and pile deviation risk in the abnormal type node into abnormal type codes; and convert the qualified, local insufficient filling, diameter reduction risk, and pile continuity abnormality in the quality result node into quality result codes. Meta-path template storage unit: Stores three types of directed meta-path templates. The first type is a stratum migration meta-path composed of pile location nodes, depth segment nodes, stratum state nodes, depth segment nodes, and pile location nodes. The second type is an anomaly quality meta-path composed of pile location nodes, depth segment nodes, construction response nodes, anomaly type nodes, and quality result nodes. The third type is a defect backtracking meta-path composed of depth segment nodes, anomaly type nodes, quality result nodes, anomaly type nodes, and depth segment nodes. Meta-path validity verification unit: Read the adjacent nodes in each directed meta-path and check whether there are corresponding semantic edges between adjacent nodes. Among them, there are spatial subordinate edges between pile location nodes and depth segment nodes, response mapping edges between depth segment nodes and construction response nodes, abnormal triggering edges between construction response nodes and abnormal type nodes, quality impact edges between abnormal type nodes and quality result nodes, and stratum correspondence edges between depth segment nodes and stratum status nodes. Meta-path instance generation unit: Generates the node sequence that has passed the validity check into meta-path instances, and writes path number, path type, starting pile position, ending pile position, starting depth segment, ending depth segment, anomaly type field and quality result field for each meta-path instance; After receiving the heterogeneous construction memory graph, the construction semantic meta-path generation layer first reads the node fields one by one by the node field parsing unit and sends them to the construction field encoding unit according to the node type. The construction field encoding unit converts continuous construction parameters and discrete construction labels into searchable codes. The meta-path template storage unit calls three types of directed meta-path templates and traverses the nodes in the heterogeneous construction memory graph according to the template order. The meta-path validity verification unit checks whether the semantic edges in the path are complete. If any adjacent node is missing a corresponding semantic edge, the path instance is deleted. The remaining paths are written into the path instance table by the meta-path instance generation unit. Each record in the path instance table contains the path type, the node number passed by the path, the construction response code, the anomaly type code, and the quality result code corresponding to the path. The cross-stake memory projection layer includes: Query node access unit: Reads the pile position coordinates, depth range start and end values ​​and design stratum information of the current unconstructed depth segment of the pile, generates a query node, and connects the query node to the current pile position node and the corresponding depth segment node in the heterogeneous construction memory map; Projectable memory filtering unit: Filters candidate memory atoms from the construction memory atoms of completed piles. The filtering conditions include: the planar distance between the pile position of the candidate memory atom and the current pile position does not exceed 6.5m; the difference between the center value of the depth segment of the candidate memory atom and the center value of the depth segment of the query node does not exceed 0.6m; the candidate memory atom and the query node are in the same stratum type or adjacent stratum strength level; and the candidate memory atom contains anomaly type markers or quality result markers. Path matching counting unit: Reads the meta-path instances output by the construction semantic meta-path generation layer, and counts the number of stratigraphic migration meta-paths, abnormal quality meta-paths, and defect backtracking meta-paths that match between each candidate memory atom and the query node; Projection weight generation unit: Generates memory projection weights based on the planar distance level, depth difference level, number of metapath hits, and consistency of quality results of candidate memory atoms. Candidate memory atoms with a planar distance of no more than 3.5m are marked as first-level spatial weights, and candidate memory atoms with a planar distance greater than 3.5m but not more than 6.5m are marked as second-level spatial weights. Candidate memory atoms with a depth difference of no more than 0.2m are marked as first-level depth weights, and candidate memory atoms with a depth difference greater than 0.2m but not more than 0.6m are marked as second-level depth weights. Memory projection execution unit: Reads candidate memory atoms from high to low according to memory projection weight, and writes the formation state characteristics, anomaly type characteristics and quality result characteristics in the candidate memory atoms into the projection cache of the query node; Projection Conflict Marking Unit: When the same query node receives multiple results of different anomaly types or different quality, it reads the memory projection weight of each projection result, marks the result with the highest weight as the main projection result, marks the remaining results as accompanying projection results, and retains the source memory atom number of the accompanying projection results.

[0027] When the cross-pile memory projection layer is running, the query node access unit first writes the current unconstructed depth segment of the pile into the heterogeneous construction memory map; the projectable memory filtering unit retrieves construction memory atoms from the completed piles with a plane distance not exceeding 6.5m and a depth difference not exceeding 0.6m; the path matching counting unit reads the meta-path instances between these candidate memory atoms and the query node, and counts whether they hit the stratum migration meta-path, abnormal quality meta-path, and defect backtracking meta-path; the projection weight generation unit sorts the candidate memory atoms according to spatial distance, depth difference, path hit count, and quality result consistency; the memory projection execution unit writes the stratum state characteristics, abnormality type characteristics, and quality result characteristics of the top-ranked candidate memory atoms into the query node projection cache; when there are multiple abnormal types, the projection conflict marking unit retains the main projection result and the accompanying projection result, and generates the cross-pile memory projection result for the current unconstructed depth segment of the pile. The risk prior aggregation layer includes: Projection result caching unit: Receives the projection results output by the cross-pile memory projection layer, and caches the main projection result, accompanying projection result, source memory atom number, projection weight, source pile location and source depth segment according to the query node number; Risk field synthesis unit: Reads multiple projection results under the same query node and synthesizes the formation state projection result, anomaly type projection result and quality result projection result into a construction risk prior state, which includes a prior formation state field, a prior anomaly type field, a prior quality risk field, projection confidence level field and source memory field; Multi-source conflict resolution unit: When there are two or more main projection candidate results for the same query node, the projection weights are compared first, then the quality result levels of the source memory atoms are compared, and then the spatial distance between the source pile and the current pile is compared. The final main risk prior state is determined in the order of high projection weight, high quality result risk level, and close spatial distance. Spatial index writing unit: The current pile position coordinates, the start and end values ​​of the unconstructed depth section, the prior state of construction risk, the source memory atom number and the projection confidence level are written into the three-dimensional spatial index structure. The three-dimensional spatial index structure is a spatial index unit of 1m×1m×0.2m. Memory field update unit: Associates the prior state of construction risk in the three-dimensional spatial index structure with the memory atoms of completed pile construction to generate a pile group construction memory field, and synchronizes the pile group construction memory field to the digital twin.

[0028] When the risk prior aggregation layer is running, the projection result caching unit first aggregates the projection results from different completed piles according to the query node number; the risk field synthesis unit synthesizes the stratum state, anomaly type, and quality results under the same query node into a unified construction risk prior state; the multi-source conflict resolution unit sorts the results from multiple sources and determines the main risk prior state and the accompanying risk prior state according to the projection weight, quality risk level, and spatial distance; the spatial index writing unit combines the plane index number corresponding to the current pile position coordinates and the depth index number corresponding to the depth segment start and end values ​​into a spatial index key, and writes the construction risk prior state into the corresponding spatial index unit; the memory field update unit arranges all spatial index units in order of pile position and depth to form a pile group construction memory field that can be called by the current pile and subsequent piles.

[0029] In this embodiment, generating underground anomalous topological ridges includes: Read construction memory atoms, the prior state of construction risk of the current unconstructed depth section of the pile, the spatial location of the pile, the location of the depth section and the source of memory projection from the construction memory field of the pile group; extract abnormal construction memory atoms based on construction response abrupt changes, abnormal stratum state, abnormality type markers and abnormal quality results. Abnormal construction memory atoms are converted into underground abnormal nodes. Each underground abnormal node contains pile location identifier, depth segment identifier, spatial location, construction response characteristics, abnormality type, quality result and construction risk prior state. Candidate topology connections are established based on the spatial proximity relationship, depth continuity relationship, abnormality type association relationship and quality result association relationship between underground abnormal nodes. Candidate topological connections are screened for multi-relation consistency, and candidate topological connections that simultaneously satisfy spatial proximity, depth continuity, anomaly association, and quality association are retained. The retained underground anomaly nodes and topological connections are combined to form an underground anomaly topological map. A hierarchical connectivity search is performed on the underground anomaly topology map. First, underground anomaly nodes with the same anomaly type are connected within the same depth layer to form horizontal anomaly segments. Then, horizontal anomaly segments with the same anomaly evolution direction are connected between adjacent depth layers to form vertical anomaly chains. Finally, the horizontal anomaly segments and vertical anomaly chains are merged into underground anomaly structural units. Specifically, the fusion of horizontal anomaly segments and vertical anomaly chains into underground anomaly structural units is as follows: Read the segment number, start and end coordinates, and anomaly type label of all horizontal anomaly segments. Enclose horizontal anomaly segments of the same type according to the minimum bounding rectangle of their vertical projection in the plane coordinates to generate a horizontal envelope. Read the chain number and chain node sequence of the vertical anomaly chain. Perform node-by-node downsampling on the chain nodes to ensure that the vertical distance between nodes is not less than 0.15m and not greater than 0.35m. After downsampling, calculate the Euclidean distance from each chain node to all horizontal envelopes. If the distance is less than 0.45m, record the association relationship between the node and the corresponding horizontal envelope. For each vertical anomaly chain, start from the top of the chain and proceed along the node sequence towards the depth direction. During traversal, when a node related to the horizontal envelope is encountered, the chain node sequence and the envelope fragment number are written into the same fusion list. After traversal, if the same fusion list contains two or more horizontal envelope numbers, they are sorted in ascending order of envelope number. The start and end coordinates of each envelope fragment are merged into a joint coverage interval, which is then combined with the chain node sequence to form an underground anomaly structure unit. The unit number, anomaly type, joint coverage interval, chain node sequence, and horizontal fragment number list are written into the fusion list. If the fusion list contains only one horizontal envelope number, the underground anomaly structure unit is generated using the start and end coordinates of the envelope fragment and the chain node sequence. Based on the spatial arrangement order, depth change direction, anomaly evolution direction, and quality result consistency of underground anomaly nodes in the underground anomaly structural unit, a central path representing the continuous direction of the underground anomaly is extracted. This central path is then bound to the corresponding anomaly type, influence depth range, influence pile location range, anomaly intensity level, and quality correlation results to generate an underground anomaly topological ridge line. Specifically, the generation of the underground anomaly topological ridge line involves: The three-dimensional coordinates, anomaly type labels, and quality result levels of all underground anomaly nodes within each underground anomaly structural unit are read. Nodes are then sorted by depth value from smallest to largest. A three-dimensional Dijkstra algorithm is used to search for the shortest connected path covering all nodes within a subgraph containing only topological connecting edges, yielding the first candidate path. A three-dimensional least-squares spline smoothing process is then performed on the candidate path node sequence using a sliding window of length 5 nodes to eliminate local jitter and maintain overall path continuity. If the Euclidean distance between adjacent nodes exceeds 0.7m after smoothing, intermediate nodes are interpolated between line segments to ensure path density. Noise segments are pruned from the smoothed path, removing segments with a length less than 0.4m. Branches with a radius of curvature greater than 15m or m are marked as redundant and removed, and the remaining main node sequence is recorded as the central path. The depth values ​​of the nodes at both ends of the central path are counted as the influence depth range. The planar distance from each node of the central path to the adjacent pile is calculated, and the set of piles with a distance of less than 2m is taken as the influence pile range. The average anomaly intensity of the abnormal nodes on the central path is calculated as the anomaly intensity level. The proportion of severe nodes is counted as the quality correlation result. Finally, the coordinate sequence of the central path, together with the anomaly type, influence depth range, influence pile range, anomaly intensity level, and quality correlation result, are encapsulated and written into the underground anomaly topological ridge data object.

[0030] In this embodiment, updating the geological state, construction risk state, and pile quality state in the current pile digital twin includes: The current pile construction path is generated based on the pile location coordinates, designed pile length, current construction depth and construction direction, and the current pile construction path is mapped to a digital twin space containing underground anomaly topological ridges. Read the anomaly type, affected depth range, affected pile location range, anomaly strength level and quality correlation results of the underground anomaly topological ridge line, and match the spatial coverage of the underground anomaly topological ridge line with the current pile construction path; When the current pile construction path enters the affected pile location range of the underground abnormal topological ridge and the depth interval corresponding to the current pile construction path falls into the affected depth range of the underground abnormal topological ridge, it is determined that the current pile construction path and the underground abnormal topological ridge have an intersection relationship, and the intersection depth segment, intersection anomaly type and intersection anomaly intensity level are recorded. Based on the intersection depth segment, intersection anomaly type, and intersection anomaly intensity level, the stratum state of the corresponding depth segment in the current pile digital twin is updated to the stratum state associated with the underground anomaly topological ridge line, and the construction risk state of the corresponding depth segment in the current pile digital twin is updated to the construction risk state associated with the intersection anomaly type. Update the pile quality status of the corresponding depth segment in the current pile digital twin based on the quality correlation results, and write the updated stratum status, construction risk status and pile quality status into the current pile digital twin.

[0031] In this embodiment, generating the set of prohibited construction actions and the minimum recovery action sequence includes: The updated digital twin of the current pile reads the geological state, construction risk state, pile quality state, underground anomaly topological ridge intersection results, and current construction control parameters for the current depth segment, and generates a quality state vector for the current depth segment. This quality state vector includes borehole wall stability, grouting filling, verticality, pile-soil disturbance, defect accumulation, and pile integrity. Specifically, the generation of the quality state vector for the current depth segment is as follows: Read the data buffer corresponding to the start and end depths of the current depth segment, and write the drilling speed sequence, power head torque sequence, hydraulic pressure sequence, grouting pressure sequence, grouting flow rate sequence, pile driver inclination angle sequence, and drill rod vibration sequence into the local array in the order of sampling time; when the array length is not less than 30 data entries, perform the following calculations on the corresponding sequences: The sum of the standard deviation of drilling speed and the standard deviation of power head torque is calculated to obtain the borehole wall disturbance index. The value is then mapped to a 0-100 scale to obtain the borehole wall stability state. The percentage of the difference between the cumulative grouting flow rate in the calculated depth section and the theoretical grouting volume is used to map the percentage to the theoretical grouting volume. This percentage is then mapped to a 0-100 point scale to obtain the grouting filling status. Take the maximum absolute value of the pile driver inclination angle and compare it with the design inclination angle limit. Map the percentage to a 0-100 point system to obtain the verticality status. The absolute value of the difference between the mean square value of drill rod vibration and the mean square value of vibration of the same type of stratum in history is calculated and mapped to a 0-100 scale to obtain the pile-soil disturbance state. The ratio of the number of anomaly type markers to the number of samples within the statistical depth segment is mapped to a 0–100 point scale to obtain the defect accumulation status. The integrity status of the pile is calculated by weighted average of five scores: borehole wall stability, grouting filling, verticality, pile-soil disturbance, and defect accumulation. The hole wall stability, grout filling, verticality, pile-soil disturbance, defect accumulation, and pile integrity are sequentially written into six fields to form the quality state vector of the current depth segment. A quality status sample library is established based on the quality locking results of completed piles. Depth segments with qualified quality results and no triggering of recovery actions are marked as normal samples. Depth segments with qualified quality results after triggering recovery actions are marked as reversible abnormal samples. Depth segments with quality results of reduced diameter, broken piles, mud inclusion, hole wall collapse, grout leakage expansion, or pile discontinuity are marked as irreversible damage samples. The quality status sample library is grouped according to geological conditions, anomaly types, and construction technology stages. Within each group, state boundary records between reversible anomaly samples and irreversible damage samples are extracted, and these state boundary records are mapped to the corresponding depth segments of the current pile digital twin to form an initial reversible-irreversible quality damage threshold surface. Specifically, the formation of the initial reversible-irreversible quality damage threshold surface is as follows: Read the formation state label, anomaly type label, construction process stage label, and corresponding six-dimensional quality state vector of each sample in the quality state sample library. Generate a grouping key by combining the three types of labels, and write all samples into a hash grouping table. For the sample set corresponding to each grouping key, sort it from high to low according to the pile integrity state. Mark samples with a pile integrity state below 60 and a defect accumulation state above 70 as irreversible damage samples, and mark samples with a pile integrity state between 60 and 80 and a defect accumulation state between 50 and 70 as reversible anomaly samples. If there are no less than 8 reversible anomaly samples and 8 irreversible damage samples under the same grouping key. The mean of the six-dimensional quality state vectors of both samples is calculated, and the midpoint of the two mean vectors in each dimension is taken as the state boundary record. If there are fewer than 8 irreversible damage samples, the midpoint of the mean vector of reversible abnormal samples and the mean vector of all irreversible damage samples in the database is taken to generate the state boundary record. Then, the depth segment number matching the grouping key in the current piling digital twin is read, the state boundary record is written into the threshold buffer of the corresponding depth segment, and the six-dimensional threshold vector is written with the depth segment number and quality state dimension index as the key. After traversing all grouping keys, the threshold vectors of each depth segment are written into the initial reversible-irreversible quality damage threshold table in ascending order of depth. Based on the intersection depth, intersection anomaly type, and anomaly intensity level of the current pile construction path and the underground abnormal topological ridge, the initial reversible-irreversible quality damage threshold surface is corrected by ridge line disturbance. The depth segment within the influence range of the grout leakage sensitive ridge, hard interlayer ridge, soft interlayer ridge, or borehole wall instability ridge is set as the ridge line constraint depth segment. The corrected reversible-irreversible quality damage threshold surface is generated within the ridge line constraint depth segment. The current depth segment's quality state vector is compared with the corrected reversible-irreversible quality damage threshold surface to determine the state approximation. When the current quality state changes from a normal state to a reversible abnormal state and continues to evolve toward an irreversible damage state, the current construction state is determined to be in an irreversible damage critical state. When the current construction state is in an irreversible damage critical state, a set of prohibited construction actions and a sequence of minimum recovery actions are generated based on the current construction control parameters, intersection anomaly type and quality state vector. The set of prohibited construction actions includes construction actions that cause the quality state to cross the corrected reversible-irreversible quality damage threshold, and the sequence of minimum recovery actions includes construction actions that return the quality state to a reversible abnormal state or a normal state.

[0032] In this embodiment, generating a quality lock result after the construction of the current depth segment is completed includes: The minimum recovery action sequence is parsed into ordered construction control instructions. The construction control parameters in the current pile construction process are adjusted according to the ordered construction control instructions. The construction control parameters include drilling speed, drill lifting speed, drill rod rotation speed, grouting pressure, grouting flow rate, drilling stop and pressure stabilization time, and re-stirring parameters. After adjusting the construction control parameters, the construction response data and quality status data of the current depth section are collected and written into the current pile digital twin to update the stratum status, construction risk status and pile quality status of the current depth section. After the construction at the current depth is completed, a quality lock result is generated. The quality lock result includes pile location identification, depth segment identification, construction response characteristics, underground abnormal topological ridge intersection results, prohibited construction action set, minimum recovery action sequence, adjusted construction control parameters, and final quality status. The quality locking result is converted into a new construction memory atom, and the new construction memory atom is fed back into the pile group construction memory field. This updates the cross-pile memory projection relationship between the completed piles and the current pile in the pile group construction memory field, and synchronizes the updated pile group construction memory field to the digital twin.

[0033] Example 1: In a continuous construction scenario of a pile group foundation, the construction area includes 60 grouting piles with a designed pile diameter of 600mm, a designed pile length of 22m, and a pile spacing between 2.4m and 3.2m. The system receives construction process data from the pile driver PLC, grouting pump, attitude sensor, and positioning terminal at a sampling frequency of 1Hz. The average construction time for a single pile is 42 minutes, generating an average of 2520 raw data points per pile. The raw data includes construction depth, drilling speed, drill lifting speed, drill rod rotation speed, power head torque, hydraulic pressure, grouting pressure, grouting flow rate, pile driver inclination angle, and vibration response. After the initial digital twin is established, the system maps the designed pile location, designed pile length, stratum layering, pile driver equipment parameters, and construction process parameters to the digital twin space. The initial stratum state is divided into five categories: backfill soil layer, silty clay layer, sand interlayer, hard interlayer, and bearing layer.

[0034] After the completion of the first 24 piles, the system segmented the entire construction process data at 0.2m depth intervals, forming 110 depth segments for each pile, resulting in a total of 2640 depth segment samples. After quality inspection and backfilling, 2148 samples were found to be normal, 126 samples were found to have abnormal grout leakage, 84 samples had high resistance due to hard interlayers, 63 samples had disturbance due to weak interlayers, 41 samples had unstable borehole walls, and 52 samples had abnormal quality results. Taking one depth segment as an example, the average torque of the power head in the 8.6m–8.8m depth segment of P18 pile was 44.1kN·m, the average grouting pressure decreased from 1.79MPa to 1.21MPa, and the average grouting flow rate increased from 71L / min to 96L / min. The final quality result was localized insufficient filling. The system encapsulated this depth segment into a construction memory atom containing pile location identifier, depth segment identifier, construction response characteristics, formation state characteristics, anomaly type characteristics, and quality result characteristics.

[0035] The system constructs a heterogeneous construction memory map based on 2640 construction memory atoms. The map forms 60 pile location nodes, 6600 depth segment nodes, 18 types of construction response nodes, 5 types of soil state nodes, 6 types of anomaly type nodes, and 4 types of quality result nodes. The improved metapath2vec model consists of a construction semantic metapath generation layer, a cross-pile memory projection layer, and a risk prior aggregation layer. The construction semantic metapath generation layer generates paths for pile location, depth segment, construction response, anomaly type, quality result, and pile location, depth segment, soil state, depth segment, and pile location. The cross-pile memory projection layer connects the unconstructed depth segment of the current pile P25 to the heterogeneous construction memory map. The risk prior aggregation layer aggregates the memory atoms of adjacent completed piles. After the model runs, the 8.6m–9.4m depth segment of pile P25 is assigned a prior state of grout leakage sensitive construction risk, and the 12.0m–12.6m depth segment of pile P25 is assigned a prior state of high-resistivity construction risk due to hard interlayer.

[0036] When the P25 pile was actually constructed to 7.8m, the traditional method only showed that the current grouting pressure was 1.76MPa and the grouting flow rate was 73L / min, neither of which reached the alarm threshold. The system of this invention has extracted anomalous construction memory atoms located at 8.6m–9.4m from the pile group construction memory field of P18, P19, and P23, and converted them into underground anomalous nodes. The system first connects nodes with the same anomalous type within the same depth layer to form a transverse anomalous segment at 8.6m–9.0m, and then connects segments with the same anomalous evolution direction between adjacent depth layers, ultimately generating a grout leakage-sensitive underground anomalous topological ridge. This ridge affects a depth range of 8.6m–9.4m, covering six adjacent piles, with an anomalous strength level of medium to high, and the quality correlation result indicating insufficient grouting.

[0037] When the construction path of pile P25 intersects with the underground anomaly ridge, the system confirms that P25 will cross the ridge within the range of 8.7m–9.3m. Therefore, before P25 enters the 8.7m range, the system updates the corresponding depth segment of the current pile digital twin's stratum status to a grout leakage sensitive state, the construction risk status to a grout diffusion anomaly risk, and the pile quality status to an insufficient filling risk. Traditional methods at this stage still only rely on the real-time parameters of the current single pile point for judgment, without calling the historical depth segment anomaly data of P18, P19, and P23, thus failing to generate an early warning.

[0038] When the P25 pile reached 8.7m, the grouting pressure decreased from 1.78MPa to 1.35MPa, while the grouting flow rate increased from 72L / min to 91L / min, still within the traditional alarm threshold of 1.20MPa. Based on the updated digital twin of the current pile, the system generated a quality state vector: borehole wall stability score was 72, grout filling score was 65, verticality score was 93, pile-soil disturbance score was 69, defect accumulation score was 64, and pile integrity score was 71. After comparing this state with the reversible-irreversible quality damage threshold, the system determined it to be in a critical state of irreversible damage. The system generates a set of prohibited construction actions, including prohibiting increasing the drilling speed, prohibiting further increasing the grouting pressure, prohibiting maintaining continuous high-flow grouting, and prohibiting rapid crossing of the depth section. At the same time, it generates a minimum recovery action sequence, adjusting the drilling speed from 0.80m / min to 0.45m / min, stabilizing the grouting pressure at 1.48MPa–1.55MPa, controlling the grouting flow rate at 78L / min–85L / min, and extending the pressure stabilization grouting time within the range of 8.8m–9.2m.

[0039] After executing the minimum recovery action sequence, the grouting pressure of pile P25 stabilized at 1.51 MPa at 9.2m, the grouting flow rate stabilized at 82 L / min, the borehole wall stability score improved to 82, the grout filling score improved to 86, and the pile integrity score improved to 88. After this depth segment was completed, the system generated a quality locking result, recording the intersection anomaly type as grout leakage sensitive, recovery action effective, and final quality status qualified. This result was then converted back into construction memory atoms and reinjected into the pile group construction memory field. Subsequently, before the construction of pile P26, the system automatically set 8.6m–9.5m as the grout leakage sensitive risk prior depth segment, enabling P26 to adopt a pressure-stabilizing grouting strategy before entering this depth.

[0040] In the comparative verification, the first 24 piles were used as training samples, and the last 16 piles were used as test samples. The traditional method identified 0 continuous underground anomalous structures, while this invention identified 4 anomalous underground topographic ridges. The early detection rate of the grout leakage sensitive depth segment using the traditional method was 18.7%, while this invention's rate was 87.5%. The average early warning depth using the traditional method was 0.18m, while this invention's was 1.46m. The average early warning time using the traditional method was 1.9min, while this invention's was 14.8min. The traditional method missed 7 grouting anomalies, while this invention missed 1. The traditional method generated 19 segments of locally insufficient filling depth, while this invention generated 6 segments. The first-pass yield rate of pile formation quality using the traditional method was 84.4%, while this invention's was 96.9%. The average time for single-pile anomaly handling using the traditional method was 28.5min, while this invention's was 12.6min.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of the entire piling machine construction process based on digital twins, characterized in that, include: Acquire data from the entire piling machine construction process to construct an initial digital twin; The data of the entire pile driver construction process is processed by depth segmentation, and construction memory atoms of corresponding depth segments are generated and a heterogeneous construction memory map is constructed. The improved metapath2vec model is used to perform cross-pile memory projection to obtain the prior state of construction risk of the current unconstructed depth segment and construct the pile group construction memory field. Abnormal construction memory atoms are extracted from the pile group construction memory field, and an underground abnormal topology map is constructed based on the abnormal construction memory atoms. The underground abnormal topology map is then connected and fitted to generate underground abnormal topology ridges. The current pile construction path is intersected with the underground abnormal topological ridge line. When the current pile construction path intersects with the underground abnormal topological ridge line, the stratum status, construction risk status and pile quality status in the current pile digital twin are updated. Based on the updated digital twin of the current pile, construct the quality state vector of the current depth segment, establish the reversible-irreversible quality damage threshold surface, determine whether the current construction state is in the irreversible damage critical state, and when the determination result is in the irreversible damage critical state, generate a set of prohibited construction actions and a minimum recovery action sequence. The construction control parameters during the current pile construction process are adjusted according to the minimum recovery action sequence. After the construction of the current depth section is completed, a quality lock result is generated. The quality lock result is converted into new construction memory atoms and reinjected into the pile group construction memory field to update the digital twin.

2. The intelligent monitoring method for the entire piling machine construction process based on digital twins according to claim 1, characterized in that, The data for the entire piling machine construction process includes pile location coordinates, construction depth, drilling speed, drill lifting speed, drill rod rotation speed, power head torque, hydraulic pressure, grouting pressure, grouting flow rate, piling machine posture, drill rod vibration, construction time, construction process stages, and pile quality results.

3. The intelligent monitoring method for the entire piling machine construction process based on digital twins according to claim 1, characterized in that, The construction of the initial digital twin includes: By analyzing engineering design data, the pile location coordinates, design pile diameter, design pile length, and pile spacing are obtained, and a spatial twin of the pile location is established. By analyzing geological exploration data, stratigraphic types, stratigraphic thickness, stratigraphic strength, and groundwater conditions at different depths are obtained, and stratigraphic condition twins are established. By analyzing the parameters of the piling machine equipment, the parameters of the power head, drill rod, hydraulic system, and grouting system are obtained, and a twin of the piling machine equipment is established. By analyzing the construction process parameters, drilling parameters, drill lifting parameters, grouting parameters, and re-mixing parameters are obtained, and a construction process twin is established. Based on the design pile diameter, design pile length, pile verticality requirements, and grouting filling requirements, a pile quality twin is established.

4. The intelligent monitoring method for the entire piling machine construction process based on digital twins according to claim 1, characterized in that, The process of obtaining the prior state of construction risk for the currently unconstructed depth section of the pile and constructing a pile group construction memory field includes: The construction process data of the completed piles is segmented according to the preset depth interval. The pile location identifier, depth segment identifier, construction response characteristics, stratum state characteristics, anomaly type characteristics and quality result characteristics corresponding to each depth segment are extracted and encapsulated into construction memory atoms. A heterogeneous construction memory graph is constructed based on construction memory atoms. The heterogeneous construction memory graph includes pile location nodes, depth segment nodes, construction response nodes, formation state nodes, anomaly type nodes, and quality result nodes. Spatial subordination edges are established between pile location nodes and depth segment nodes, response mapping edges are established between depth segment nodes and construction response nodes, anomaly triggering edges are established between construction response nodes and anomaly type nodes, quality influence edges are established between anomaly type nodes and quality result nodes, and formation correspondence edges are established between depth segment nodes and formation state nodes. An improved metapath2vec model is constructed, which includes a construction semantic metapath generation layer, a cross-pile memory projection layer, and a risk prior aggregation layer connected in sequence. In the construction semantic meta-path generation layer, three types of directed meta-paths are set. The first type of directed meta-path is the stratum migration meta-path composed of pile location node, depth segment node, stratum state node, depth segment node, and pile location node. The second type of directed meta-path is the abnormal quality meta-path composed of pile location node, depth segment node, construction response node, anomaly type node, and quality result node. The third type of directed meta-path is the defect backtracking meta-path composed of depth segment node, anomaly type node, quality result node, anomaly type node, and depth segment node. In the cross-pile memory projection layer, the current unconstructed depth segment of the pile is used as a query node and connected to the heterogeneous construction memory map. Based on the spatial distance of the pile location, the distance of the depth segment, the consistency of the directed element path matching result and the quality result, projectable memory atoms are selected from the construction memory atoms of the completed piles. The stratum state, anomaly type and quality result corresponding to the projectable memory atoms are projected to the query node to obtain the prior state of construction risk of the current unconstructed depth segment of the pile. In the risk prior aggregation layer, the construction risk prior state of the current unconstructed depth segment of the pile, the corresponding projectable memory atom, the spatial location of the pile, the depth segment location, and the memory projection source are written into the same spatial index structure to form a pile group construction memory field.

5. The intelligent monitoring method for the entire piling machine construction process based on digital twins according to claim 1, characterized in that, The generation of underground anomalous topological ridges includes: Read construction memory atoms, the prior state of construction risk of the current unconstructed depth section of the pile, the spatial location of the pile, the location of the depth section and the source of memory projection from the construction memory field of the pile group; extract abnormal construction memory atoms based on construction response abrupt changes, abnormal stratum state, abnormality type markers and abnormal quality results. Abnormal construction memory atoms are converted into underground abnormal nodes. Each underground abnormal node contains pile location identifier, depth segment identifier, spatial location, construction response characteristics, abnormality type, quality result and construction risk prior state. Candidate topology connections are established based on the spatial proximity relationship, depth continuity relationship, abnormality type association relationship and quality result association relationship between underground abnormal nodes. Candidate topological connections are screened for multi-relation consistency, and candidate topological connections that simultaneously satisfy spatial proximity, depth continuity, anomaly association, and quality association are retained. The retained underground anomaly nodes and topological connections are combined to form an underground anomaly topological map. A hierarchical connectivity search is performed on the underground anomaly topology map. First, underground anomaly nodes with the same anomaly type are connected within the same depth layer to form a horizontal anomaly segment. Then, horizontal anomaly segments with the same anomaly evolution direction are connected between adjacent depth layers to form a vertical anomaly chain. Finally, the horizontal anomaly segments and the vertical anomaly chain are merged into an underground anomaly structural unit. Based on the spatial arrangement order, depth change direction, anomaly evolution direction, and consistency of quality results of underground anomaly nodes in the underground anomaly structural unit, a central path representing the continuous direction of underground anomalies is extracted. The central path is then bound to the corresponding anomaly type, influence depth range, influence pile location range, anomaly intensity level, and quality correlation results to generate an underground anomaly topological ridge.

6. The intelligent monitoring method for the entire process of pile driver construction based on digital twins according to claim 1, characterized in that, The updating of the current pile digital twin's geological state, construction risk state, and pile quality state includes: The current pile construction path is generated based on the pile location coordinates, designed pile length, current construction depth and construction direction, and the current pile construction path is mapped to a digital twin space containing underground anomaly topological ridges. Read the anomaly type, affected depth range, affected pile location range, anomaly strength level and quality correlation results of the underground anomaly topological ridge line, and match the spatial coverage of the underground anomaly topological ridge line with the current pile construction path; When the current pile construction path enters the affected pile location range of the underground abnormal topological ridge and the depth interval corresponding to the current pile construction path falls into the affected depth range of the underground abnormal topological ridge, it is determined that the current pile construction path and the underground abnormal topological ridge have an intersection relationship, and the intersection depth segment, intersection anomaly type and intersection anomaly intensity level are recorded. Based on the intersection depth segment, intersection anomaly type, and intersection anomaly intensity level, the stratum state of the corresponding depth segment in the current pile digital twin is updated to the stratum state associated with the underground anomaly topological ridge line, and the construction risk state of the corresponding depth segment in the current pile digital twin is updated to the construction risk state associated with the intersection anomaly type. Update the pile quality status of the corresponding depth segment in the current pile digital twin based on the quality correlation results, and write the updated stratum status, construction risk status and pile quality status into the current pile digital twin.

7. The intelligent monitoring method for the entire piling machine construction process based on digital twins according to claim 1, characterized in that, The generation of the prohibited construction action set and the minimum recovery action sequence includes: Read the current geological conditions, construction risk conditions, pile quality conditions, underground abnormal topological ridge intersection results and current construction control parameters from the updated current pile digital twin, and generate the current depth segment's quality state vector. The quality state vector includes the borehole wall stability state, grouting filling state, verticality state, pile-soil disturbance state, defect accumulation state and pile integrity state. A quality status sample library is established based on the quality locking results of completed piles. Depth segments with qualified quality results and no triggering of recovery actions are marked as normal samples. Depth segments with qualified quality results after triggering recovery actions are marked as reversible abnormal samples. Depth segments with quality results of reduced diameter, broken piles, mud inclusion, hole wall collapse, grout leakage expansion, or pile discontinuity are marked as irreversible damage samples. The quality status sample library is grouped according to the stratum state, anomaly type and construction process stage. Within each group, the state boundary record between reversible anomaly samples and irreversible damage samples is extracted, and the state boundary record is mapped to the corresponding depth segment of the current pile digital twin to form an initial reversible-irreversible quality damage threshold surface. Based on the intersection depth, intersection anomaly type, and anomaly intensity level of the current pile construction path and the underground abnormal topological ridge, the initial reversible-irreversible quality damage threshold surface is corrected by ridge line disturbance. The depth segment within the influence range of the grout leakage sensitive ridge, hard interlayer ridge, soft interlayer ridge, or borehole wall instability ridge is set as the ridge line constraint depth segment. The corrected reversible-irreversible quality damage threshold surface is generated within the ridge line constraint depth segment. The current depth segment's quality state vector is compared with the corrected reversible-irreversible quality damage threshold surface to determine the state approximation. When the current quality state changes from a normal state to a reversible abnormal state and continues to evolve toward an irreversible damage state, the current construction state is determined to be in an irreversible damage critical state. When the current construction state is in an irreversible damage critical state, a set of prohibited construction actions and a sequence of minimum recovery actions are generated based on the current construction control parameters, intersection anomaly type and quality state vector. The set of prohibited construction actions includes construction actions that cause the quality state to cross the corrected reversible-irreversible quality damage threshold, and the sequence of minimum recovery actions includes construction actions that return the quality state to a reversible abnormal state or a normal state.

8. The intelligent monitoring method for the entire piling machine construction process based on digital twins according to claim 1, characterized in that, The process of generating a quality lock result after construction at the current depth section is completed includes: The minimum recovery action sequence is parsed into ordered construction control instructions. The construction control parameters in the current pile construction process are adjusted according to the ordered construction control instructions. The construction control parameters include drilling speed, drill lifting speed, drill rod rotation speed, grouting pressure, grouting flow rate, drilling stop and pressure stabilization time, and re-stirring parameters. After adjusting the construction control parameters, the construction response data and quality status data of the current depth section are collected and written into the current pile digital twin to update the stratum status, construction risk status and pile quality status of the current depth section. After the construction at the current depth is completed, a quality lock result is generated. The quality lock result includes pile location identification, depth segment identification, construction response characteristics, underground abnormal topological ridge intersection results, prohibited construction action set, minimum recovery action sequence, adjusted construction control parameters, and final quality status. The quality locking result is converted into a new construction memory atom, and the new construction memory atom is fed back into the pile group construction memory field. This updates the cross-pile memory projection relationship between the completed piles and the current pile in the pile group construction memory field, and synchronizes the updated pile group construction memory field to the digital twin.

Images