A pile machine construction posture dynamic correction method based on multi-source perception fusion

Abstract

The application discloses a pile machine construction posture dynamic correction method based on multi-source sensing fusion, which comprises the following steps: collecting multi-source state data, completing time synchronization, filtering and abnormality elimination, and generating a construction state data set; constructing a virtual construction reference field, generating a target construction axis of a current construction stage, and calculating an actual construction axis; constructing a posture deviation matrix and a posture sensitivity matrix, extracting an observable subspace, and generating a projection operator; projecting a hydraulic actuator control input, generating a constraint control input, constructing and updating a posture memory field; constructing a memory field modulated adaptive Hammerstein model, generating a posture dynamic response prediction result; generating a hydraulic correction control amount and performing posture correction, and updating the reference field, the observable subspace and the memory field. Through the virtual construction reference field, the topological observability projection operator and the posture memory field modulated model, the dynamic stable correction of the pile machine construction posture is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for pile driver construction, and in particular to a dynamic correction method for pile driver construction posture based on multi-source perception fusion. Background Technology

[0002] Piling rigs are widely used in building foundations, bridge foundations, and municipal engineering construction. Their construction posture directly affects the quality of pile hole formation, pile verticality, and subsequent load-bearing capacity. Existing piling rig posture correction typically relies on tilt sensors, inertial measurement units, hydraulic cylinder displacement sensors, laser plumb bobs to collect equipment status data, and controls the hydraulic leveling mechanism to correct based on the instantaneous posture deviation of the vehicle body or mast, so as to keep the piling rig or drill rod as close as possible to the set construction posture.

[0003] However, in actual construction environments, foundation conditions are complex, especially in soft foundations, backfilled soil layers, or heterogeneous strata. The pile driver support points are affected by local settlement, eccentric loading, and construction vibrations. Existing methods typically use the detected attitude error directly as the basis for correction, without fully considering the coupling relationship between the foundation contact state, construction load state, drill rod stress state, and attitude deviation. They also fail to determine whether the hydraulic adjustment direction is a correctable direction that the current structure can effectively respond to, which can easily lead to the output of invalid control quantities in unobservable or low-response directions.

[0004] Existing attitude correction methods mostly focus on feedback control of the error at the current moment, lacking modeling and utilization of historical attitude deviation trends. When a deviation direction accumulates due to foundation plastic settlement, continuous stress on the drill pipe, or repeated leveling operations, correction based solely on instantaneous errors is prone to overcompensation, repeated oscillations, or correction failure.

[0005] Therefore, how to provide a dynamic correction method for pile driver construction posture based on multi-source perception fusion 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 a dynamic correction method for pile driver construction posture based on multi-source sensing fusion. This invention integrates posture data, hydraulic state data, construction load data, and vibration state data during pile driver construction to construct a virtual construction reference field, a topological observability projection operator, and an adaptive Hammerstein model modulated by the posture memory field. This enables dynamic modeling of the target construction axis, correctable direction, and historical deviation trend, and completes adaptive posture correction of the hydraulic actuator. It has the advantages of strong adaptability to complex foundations, high ability to suppress miscorrection of unobservable directions, strong ability to identify historical deviation accumulation, and high stability of dynamic correction of construction posture.

[0007] A method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to an embodiment of the present invention includes: Collect multi-source status data during the pile driver construction process, preprocess the multi-source status data, and generate a construction status dataset; A virtual construction reference field is constructed based on the construction status dataset. According to the foundation contact state, construction load state, drill rod stress state and historical attitude deviation trend, the target construction axis corresponding to the current construction stage is generated, and the actual construction axis is calculated. An attitude deviation matrix is ​​constructed based on the spatial deviation between the actual construction axis and the target construction axis. An attitude sensitivity matrix is ​​established based on the correspondence between the construction state dataset and the attitude deviation matrix. Singular value decomposition is performed on the attitude sensitivity matrix to extract the observable subspace and construct a topological observability projection operator. The control input of the hydraulic actuator is projected and mapped using the topological observability projection operator to generate the constrained control input located in the observable subspace. An attitude memory field is constructed based on historical construction attitude deviations, and the attitude memory field is used to accumulate and update the historical deviation trend. An adaptive Hammerstein model for attitude memory field modulation is constructed, with constraint control input as input variable, attitude memory field and virtual construction reference field as nonlinear module modulation parameters, and foundation contact state as linear dynamic module modulation parameters, to generate attitude dynamic response prediction results. Based on the attitude dynamic response prediction results, hydraulic correction control quantities are generated and attitude correction is performed. The virtual construction reference field, topological observable subspace, and attitude memory field are updated based on the corrected construction status data.

[0008] Optionally, the multi-source state data includes vehicle body posture data, mast posture data, power head posture data, hydraulic cylinder pressure data, hydraulic cylinder displacement data, drill pipe penetration depth data, power head torque data, feed pressure data, vibration acceleration data, laser alignment data, and visual axis data.

[0009] Optionally, the preprocessing of multi-source status data to generate a construction status dataset includes: Multi-source state data is timestamped to obtain time-synchronized multi-source state data. A sliding window is used to remove outliers from the time-synchronized multi-source state data to obtain valid state data. The valid state data is low-pass filtered to obtain filtered state data. The filtered state data is resampled according to the same sampling period to generate a construction state dataset.

[0010] Optionally, generating the target construction axis corresponding to the current construction stage and calculating the actual construction axis includes: Hydraulic cylinder pressure data, hydraulic cylinder displacement data, power head torque data, feed pressure data, drill rod penetration depth data, vibration acceleration data, mast attitude data, and power head attitude data are extracted from the construction status dataset to form the foundation contact status, construction load status, drill rod stress status, and historical attitude deviation trend, respectively. Using the design pile centerline as the initial reference, multiple continuous depth layers are divided in the direction of pile driver construction depth. Multiple candidate axis nodes are generated in each depth layer. Each candidate axis node is marked with support stability attribute, load disturbance attribute, drill rod force offset attribute and historical deviation accumulation attribute. Based on the support stability attribute, load disturbance attribute, drill pipe force offset attribute and historical deviation accumulation attribute of each candidate axis node, a node field value is generated. The candidate axis nodes and their node field values ​​in the same depth layer are combined to form a depth layer reference field. Multiple depth layer reference fields are connected in the order of construction depth to form a virtual construction reference field. In the virtual construction reference field, candidate axis nodes that satisfy the highest support stability attribute, lowest load disturbance attribute, lowest drill rod force offset attribute and limited historical deviation accumulation attribute are selected sequentially according to the construction depth direction. The selected candidate axis nodes in adjacent depth layers are connected to generate the target construction axis corresponding to the current construction stage. The theoretical axis of the mast is determined based on the mast posture data and power head posture data in the construction status dataset. The force offset of the drill rod is determined based on the drill rod penetration depth data, power head torque data, feed pressure data, and vibration acceleration data. The ground settlement offset is determined based on the hydraulic cylinder pressure data and hydraulic cylinder displacement data. The actual construction axis is obtained by combining the theoretical axis of the mast, the force offset of the drill rod, and the ground settlement offset.

[0011] Optionally, the step of performing singular value decomposition on the attitude sensitivity matrix, extracting the observable subspace, and constructing a topological observability projection operator includes: The differences between the actual construction axis and the target construction axis in lateral offset, longitudinal offset, roll attitude deviation and pitch attitude deviation are combined into an attitude deviation matrix, and the attitude deviation matrix is ​​layered and marked according to the construction depth. Hydraulic cylinder pressure data, hydraulic cylinder displacement data, foundation contact state and attitude deviation matrix are extracted from the construction status dataset. The correspondence between hydraulic state changes and attitude deviation changes is established, and an attitude sensitivity matrix is ​​generated based on the correspondence. The construction attitude transfer topology is constructed based on the attitude sensitivity matrix. The construction attitude transfer topology includes hydraulic actuator nodes, foundation support nodes, mast attitude nodes, power head attitude nodes and axis deviation nodes. The consistency of response direction, stability of response delay and continuity of response amplitude between nodes are used as the validity criteria for the topology edges. The observable paths of the construction attitude transfer topology are screened, and the topology paths that can be stably transferred from the hydraulic actuator node to the foundation support node, mast attitude node or power head attitude node are retained. Topology paths that repeatedly switch response directions, have response delays exceeding the preset sampling period range or have discontinuous response amplitudes are eliminated. The hydraulic control direction and attitude deviation direction corresponding to the retained topology paths are combined into an observable subspace. The extension direction, retraction direction, and holding direction of each hydraulic actuator are set as control direction units. Each control direction unit is matched with a topological path in the observable subspace. When a control direction unit has a corresponding observable topological path, the control direction unit is marked as a passable direction. Path allocation weights are generated based on the continuity of response amplitude, stability of response delay, and stability of foundation support of the corresponding topological path. When a control direction unit does not have a corresponding observable topological path, the control direction unit is marked as a shielded direction. The passable direction, shielded direction, and path allocation weights are combined into a topological observability projection operator.

[0012] Optionally, the generation of constraint control inputs located in the observable subspace, and the construction of an attitude memory field based on historical construction attitude deviations, includes: The control inputs of the hydraulic actuator are obtained, including the extension control amount, retraction control amount, and holding control amount of each hydraulic cylinder; The control inputs of the hydraulic actuator are filtered by the topological observability projection operator, retaining the control inputs that belong to the passable direction and removing the control inputs that belong to the shielded direction. Based on the path assignment weights in the topological observability projection operator, the retained control inputs are weighted to generate constrained control inputs located in the observable subspace. An attitude memory unit is established according to the construction depth layer, attitude deviation direction and deviation source. The attitude deviation direction includes the lateral offset direction, longitudinal offset direction, roll deviation direction and pitch deviation direction. The deviation source includes the foundation contact state deviation, construction load state deviation and drill rod stress state deviation. Write the attitude deviation matrix of the current construction cycle into the corresponding attitude memory unit, and accumulate and mark the attitude deviations that occur continuously in the same construction depth layer and have the same direction, mark the attitude deviations that change direction, and mark the attitude deviations that are self-locked by the shielding direction. The attitude memory units that have completed the accumulation marking, attenuation marking and self-locking marking in each construction depth layer are combined to form an attitude memory field, and the historical deviation trend in the attitude memory field is updated according to the attitude deviation matrix of the current construction cycle.

[0013] Optionally, the generation of attitude dynamic response prediction results includes: An adaptive Hammerstein model for attitude memory field modulation is constructed, comprising an attitude memory field index modulation layer, a static nonlinear input mapping layer, and a ground contact dynamic response layer. The constraint control input is input to the attitude memory field index modulation layer, and the corresponding attitude memory unit is read from the attitude memory field according to the current construction depth layer, attitude deviation direction and hydraulic actuator type. Based on the cumulative update flag, decay update flag, and self-locking flag in the attitude memory unit, memory modulation parameters for the corresponding hydraulic control channel are generated. The memory modulation parameters include input gain modulation parameters, directional offset modulation parameters, input limiting modulation parameters, and response hysteresis modulation parameters. The constraint control input and memory modulation parameters are input to the static nonlinear input mapping layer. The static nonlinear input mapping layer adjusts the input gain of the corresponding hydraulic control channel according to the input gain modulation parameters, corrects the correction direction of the corresponding hydraulic control channel according to the direction bias modulation parameters, and limits the input amplitude of the corresponding hydraulic control channel according to the input limiting modulation parameters to generate the intermediate control quantity after memory modulation. The virtual construction reference field is input into the static nonlinear input mapping layer. Based on the target construction axis corresponding to the current construction depth layer, the intermediate control quantities are screened for axis approach direction. The intermediate control quantities that approach the target construction axis are retained, and the intermediate control quantities that deviate from the target construction axis are suppressed, thereby generating axis constraint control quantities. The axis constraint control quantity is input into the foundation contact dynamic response layer, and the foundation contact state is used as the dynamic response parameter. The foundation contact dynamic response layer generates the attitude dynamic response prediction result corresponding to the current construction cycle based on the foundation support stability attribute, load disturbance attribute, drill rod force offset attribute and response hysteresis modulation parameter.

[0014] Optionally, updating the virtual construction reference field, topological observable subspace, and attitude memory field based on the corrected construction status data includes: The dynamic response prediction results of the attitude are compared with the attitude deviation matrix to determine the correction direction and correction magnitude corresponding to the lateral offset, longitudinal offset, roll attitude deviation and pitch attitude deviation, and generate the initial hydraulic correction control quantity. The initial hydraulic correction control quantity is filtered by the topological observability projection operator. The correction control quantity belonging to the observable subspace is retained, and the correction control quantity belonging to the unobservable direction is deleted, thereby generating an executable hydraulic correction control quantity. The executable hydraulic correction control quantity is decomposed into the extension control quantity, retraction control quantity, and holding control quantity of each hydraulic actuator, and a hydraulic execution sequence is generated according to the construction depth layer and the direction of attitude deviation. The hydraulic actuator is driven to perform attitude correction according to the hydraulic execution sequence, and multi-source state data after correction is collected during the execution process to generate corrected construction state data. Based on the corrected construction status data, the actual construction axis and attitude deviation matrix are recalculated, and the virtual construction reference field, topological observable subspace, and attitude memory field are updated based on the recalculated actual construction axis and attitude deviation matrix.

[0015] The beneficial effects of this invention are: This invention constructs a virtual construction reference field, transforming the traditional attitude correction method based on a fixed vertical reference into a correction method that dynamically generates the target construction axis based on the current construction state. This allows the piling rig to adjust its construction attitude in real time according to the foundation contact state, construction load state, drill rod stress state, and historical attitude deviation trends. In soft foundations, heterogeneous strata, and complex construction environments, this invention can reduce the impact of local foundation settlement, construction vibration, and drill rod stress deviation on the construction attitude, improve the matching degree between the actual construction axis and the target construction axis, and improve the quality of pile hole formation and construction accuracy.

[0016] This invention constructs a topological observability projection operator to filter and constrain the observable directions of the control inputs to the hydraulic actuator. It retains only control paths that can stably propagate to the axis deviation nodes and masks control inputs corresponding to unobservable directions. This avoids miscorrection problems caused by insufficient structural response, unstable foundation support, or invalid control directions in traditional attitude correction processes. By filtering response direction consistency, response delay stability, and response amplitude continuity, the effectiveness and stability of the hydraulic control path can be improved, reducing repeated leveling and attitude oscillations during construction.

[0017] This invention introduces an attitude memory field and combines it with an adaptive Hammerstein model modulated by the attitude memory field to continuously record and dynamically modulate historical construction attitude deviations. This enables the system to identify long-term deviation trends and attitude change patterns at different construction depths. By classifying and processing accumulated deviations, self-locking deviations, and attenuation deviations, the overcompensation problem caused by the accumulation of historical deviations can be effectively suppressed, improving the accuracy of attitude dynamic response prediction and the stability of hydraulic correction control. This enhances the continuous adaptive adjustment capability of the piling machine during construction and improves construction safety, efficiency, and attitude dynamic correction accuracy under complex working conditions. 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 a dynamic correction method for pile driver construction posture based on multi-source sensing fusion proposed in this invention; Figure 2 This is a schematic diagram illustrating the posture memory field construction and updating process of a dynamic correction method for pile driver construction posture based on multi-source sensing fusion 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 dynamic correction method for pile driver construction posture based on multi-source sensing fusion, comprising: Collect multi-source status data during the pile driver construction process, preprocess the multi-source status data, and generate a construction status dataset; A virtual construction reference field is constructed based on the construction status dataset. According to the foundation contact state, construction load state, drill rod stress state and historical attitude deviation trend, the target construction axis corresponding to the current construction stage is generated, and the actual construction axis is calculated. An attitude deviation matrix is ​​constructed based on the spatial deviation between the actual construction axis and the target construction axis. An attitude sensitivity matrix is ​​established based on the correspondence between the construction state dataset and the attitude deviation matrix. Singular value decomposition is performed on the attitude sensitivity matrix to extract the observable subspace and construct a topological observability projection operator. The control input of the hydraulic actuator is projected and mapped using the topological observability projection operator to generate the constrained control input located in the observable subspace. An attitude memory field is constructed based on historical construction attitude deviations, and the attitude memory field is used to accumulate and update the historical deviation trend. An adaptive Hammerstein model for attitude memory field modulation is constructed, with constraint control input as input variable, attitude memory field and virtual construction reference field as nonlinear module modulation parameters, and foundation contact state as linear dynamic module modulation parameters, to generate attitude dynamic response prediction results. Based on the attitude dynamic response prediction results, hydraulic correction control quantities are generated and attitude correction is performed. The virtual construction reference field, topological observable subspace, and attitude memory field are updated based on the corrected construction status data.

[0021] In this embodiment, the multi-source state data includes vehicle body posture data, mast posture data, power head posture data, hydraulic cylinder pressure data, hydraulic cylinder displacement data, drill rod penetration depth data, power head torque data, feed pressure data, vibration acceleration data, laser alignment data, and visual axis data.

[0022] In this embodiment, the preprocessing of multi-source status data to generate a construction status dataset includes: Multi-source state data is timestamped to obtain time-synchronized multi-source state data. A sliding window is used to remove outliers from the time-synchronized multi-source state data to obtain valid state data. The valid state data is low-pass filtered to obtain filtered state data. The filtered state data is resampled according to the same sampling period to generate a construction state dataset.

[0023] In this embodiment, generating the target construction axis corresponding to the current construction stage and calculating the actual construction axis includes: Hydraulic cylinder pressure data, hydraulic cylinder displacement data, power head torque data, feed pressure data, drill pipe penetration depth data, vibration acceleration data, mast attitude data, and power head attitude data are extracted from the construction status dataset to form data on foundation contact status, construction load status, drill pipe stress status, and historical attitude deviation trends. Specifically, the formation of foundation contact status, construction load status, drill pipe stress status, and historical attitude deviation trends are as follows: Formation of foundation contact state: The pressure and displacement data of each outrigger or track hydraulic cylinder are synchronously compared, the change range of support reaction force and displacement response of each support point at the current construction depth layer are calculated, and the softness, settlement rate and stability level of the support point are determined by combining the change of mast attitude, and the foundation contact state at the current depth layer is obtained by summarizing. Formation of construction load state: Extract power head torque data, feed pressure data and drill rod penetration depth data, establish torque-depth curve and pressure-depth curve in time sequence, combine the frequency of peak vibration acceleration, divide stable drilling section and high disturbance section, obtain the construction load magnitude, load direction and disturbance intensity at the current moment, and generate construction load state. Formation of drill pipe stress state: The torque data of the power head, the attitude data of the mast and the vibration acceleration data are correlated to analyze the coupling relationship of the drill pipe under torsion, compression and lateral vibration, identify the bending trend and main force direction of the drill pipe, and obtain three indicators: drill pipe force offset, force direction and force amplitude, and generate the drill pipe stress state. The formation of historical attitude deviation trends: According to the construction depth layer and construction cycle, the attitude deviation matrix recorded in the previous cycle and earlier cycles is arranged in time series. The cumulative occurrence, cumulative deviation amplitude and duration of each deviation direction in the same depth layer are counted. Combined with the rate of change of the foundation contact state in the current depth layer, the long-term deviation direction, short-term deviation direction and stable deviation direction are extracted to generate historical attitude deviation trends. Using the designed pile centerline as an initial reference, multiple continuous depth layers are divided along the pile driver's construction depth direction. Multiple candidate axis nodes are generated within each depth layer. Each candidate axis node is marked with support stability attributes, load disturbance attributes, drill rod force offset attributes, and historical deviation accumulation attributes. Specifically, the division of multiple continuous depth layers along the pile driver's construction depth direction and the generation of multiple candidate axis nodes within each depth layer are as follows: Using the center line of the designed pile location as the Z-axis, longitudinal division is carried out from the top of the pile downwards. The initial layer thickness is taken as the reference thickness set in the construction parameter library. The drilling resistance increase and support settlement rate are read in real time. When any indicator shows a sharp increase or a sharp decrease and continues for more than a preset time window, the current layer thickness is reduced to half of the reference thickness so that the layer boundary fits the hard and soft interface of the stratum, and the depth layer division is completed with the beginning and end connected and without overlap. After entering a depth layer, a polar coordinate system is established within the cross-section of the layer center. With the design center point as the pole, the circumference is discretized into several radial directions at equal angular intervals. The extreme value of the support stiffness of the depth layer is detected and proportionally calculated with the average value of the support stiffness of the same layer to obtain the support stiffness ratio. The first radial step size of the current layer is selected based on the support stiffness ratio. If the support stiffness ratio is less than the preset threshold, a larger step size is taken; otherwise, a smaller step size is taken. Then, multiple concentric rings are generated by increasing outward by two levels according to the selected step size. Candidate axis nodes are created at the intersections of each radial direction and each ring. The depth layer number, radial distance, azimuth angle and ring number of the node are recorded to generate node geometric identifiers. If a unilateral torque peak occurs in the drill pipe stress state of the depth layer and the peak duration exceeds the set threshold, an axial expansion node is added along the outermost ring radius by a fixed distance in the peak direction to capture possible lateral bending trends. All ring nodes and expansion nodes in the same depth layer are summarized and written into the node index table according to the node geometric identifiers. Based on the support stability properties, load disturbance properties, drill pipe force offset properties, and historical deviation accumulation properties of each candidate axis node, node field values ​​are generated. Candidate axis nodes and their node field values ​​within the same depth layer are combined to form a depth layer reference field. Multiple depth layer reference fields are connected in sequence according to construction depth to form a virtual construction reference field, wherein: The node field values ​​are generated as follows: The support stability attribute, load disturbance attribute, drill rod force offset attribute, and historical deviation accumulation attribute of the candidate axis nodes are normalized to map the four attributes to a unified dimension range. Adjustable weight coefficients are applied to the normalized four attributes. The weight coefficients are retrieved from the preset weight library according to the foundation type and construction technology of the current construction stage. After the weight allocation is completed, the four weighted attributes are weighted and summed and linearly smoothed to obtain the single numerical field value of the node. The value is then added to the node index table as the node field value. The deep-layer reference field is formed as follows: Within the same depth layer, using the ring number and azimuth angle as two-dimensional indices, the corresponding candidate axis nodes and their node field values ​​are mapped to a polar coordinate grid. For the position of missing nodes in the polar coordinate grid, the field values ​​of the neighboring nodes are used for directional consistency interpolation to fill the complete polar coordinate grid. The polar coordinate grid is then converted into a two-dimensional matrix structure, and the ring number and azimuth angle order corresponding to the matrix rows and columns are recorded to form the depth layer reference field. The upper depth layer reference field and the lower depth layer reference field are connected in the order of construction depth. The reference fields of each depth layer are spliced ​​together in the longitudinal direction. During the splicing process, the field values ​​at the boundary of adjacent depth layers are subjected to extreme value smoothing to make the virtual construction reference field continuous in the depth direction without step change. In the virtual construction reference field, candidate axis nodes are selected sequentially along the construction depth direction, satisfying the highest support stability attribute, lowest load disturbance attribute, lowest drill pipe force offset attribute, and limited historical deviation accumulation attribute. The selected candidate axis nodes in adjacent depth layers are connected to generate the target construction axis corresponding to the current construction stage. Specifically, generating the target construction axis corresponding to the current construction stage involves: Read the node field value matrix of each depth layer in the virtual construction reference field sequentially from the pile top to the pile bottom. For the current depth layer, sort the node field values ​​from largest to smallest, and sequentially search for candidate axis nodes whose node field values ​​meet the criteria of highest support stability attribute, lowest load disturbance attribute, lowest drill rod force offset attribute, and historical deviation accumulation attribute has not triggered the locking flag. Stop searching when encountering a node that meets the conditions during sorting and traversal, and determine the optimal node of the depth layer. Then, the search process is repeated in the next depth layer. When the azimuth difference between the optimal node found and the optimal node in the previous depth layer is less than the preset angle tolerance, the two nodes are directly connected by a line segment. When the azimuth difference exceeds the angle tolerance, a candidate node with the middle azimuth and the top three node field values ​​is inserted between the optimal node in the current depth layer and the optimal node in the previous depth layer. The node is connected by a broken line to smoothly transition the axis direction. Continue traversing all depth layers downwards until the final design depth layer is reached, and complete the sequential connection of nodes; after the connection is completed, smooth the entire node connection line in the depth direction to make the turning angles of adjacent line segments continuous and meet the maximum allowable deflection angle requirements of the designed pile body, and generate a target construction axis that matches the current construction stage. The theoretical mast axis is determined based on the mast attitude data and power head attitude data in the construction status dataset. The drill rod force offset is determined based on the drill rod penetration depth data, power head torque data, feed pressure data, and vibration acceleration data. The foundation settlement offset is determined based on the hydraulic cylinder pressure data and hydraulic cylinder displacement data. The actual construction axis is obtained by combining the theoretical mast axis, drill rod force offset, and foundation settlement offset. The theoretical axis of the mast is determined as follows: Read the mast attitude data and power head attitude data from the construction status dataset, project the mast pitch angle and roll angle onto the mast body coordinate system, and transform the power head pitch angle and roll angle into the mast body coordinate system to perform attitude consistency verification. Among the qualified attitude segments, take the segment with the smallest attitude change within the continuous time window as the statically stable section, calculate the spatial direction of the longitudinal baseline of the mast body after rotation by pitch and roll angles within the statically stable section, push the direction downward and outward from the lower hinge point of the mast, and use the center point of the power head as the verification benchmark to correct the hinge point error to obtain the theoretical axis of the mast. The specific method for determining the force offset of the drill pipe is as follows: Select the current depth layer from the drill pipe penetration depth data, and read the power head torque data, feed pressure data, and vibration acceleration data corresponding to the depth layer; pair the feed pressure and vibration amplitude at the moment corresponding to the torque peak to determine the torsional-compression coupling point of the drill pipe, and then count the frequency of occurrence of the same-direction coupling point according to the depth interval and torque direction between the coupling points; when the same-direction coupling point appears in three consecutive depth intervals and the frequency of occurrence is greater than the preset threshold, it is determined that the drill pipe has a stable lateral bending trend. Take the torque direction as the lateral offset direction, and use the load index formed by the feed pressure and torque to convert the offset amplitude to obtain the force offset of the drill pipe. The specific method for determining the foundation settlement offset is as follows: Read the hydraulic cylinder pressure and displacement data, and subtract the real-time displacement value of each outrigger or track cylinder from the reference displacement value at the initial alignment to obtain the vertical settlement of each support point; classify the vertical settlement by orientation, calculate the average settlement of support points in the same orientation, and obtain the overall settlement difference of the piling machine chassis on the two horizontal axes of transverse and longitudinal; based on the chassis settlement difference and the height of the mast hinge point, use the rigid body rotation geometry to calculate the spatial displacement of the top of the mast, and project the displacement onto the vertical plane of the mast's theoretical axis to obtain the foundation settlement offset.

[0024] In this embodiment, the step of performing singular value decomposition on the attitude sensitivity matrix, extracting the observable subspace, and constructing a topological observability projection operator includes: The differences between the actual construction axis and the target construction axis in lateral offset, longitudinal offset, roll attitude deviation, and pitch attitude deviation are combined to form an attitude deviation matrix. This attitude deviation matrix is ​​then layered and labeled according to the construction depth order. Specifically, the combined attitude deviation matrix is ​​as follows: During the current construction period, using the longitudinal drilling depth of the pile driver as an index, the lateral offset value, longitudinal offset value, roll attitude deviation value, and pitch attitude deviation value are extracted according to the sampling order from the top of the pile to the bottom of the pile. The four deviation values ​​of the same depth sampling point are written into a row of data in a fixed column order to form a row vector containing four deviation columns. The depth layer number of the depth sampling point is added to the first column of the row to form a record row with a five-column structure. Traverse all sampling depths sequentially downwards, appending the record rows of each depth sampling point in ascending order of depth, accumulating row by row to form a two-dimensional matrix structure. After completing the recording of all depth sampling points, write the depth layer number field name in the first column of the first row of the matrix, and write the lateral offset, longitudinal offset, roll deviation, and pitch deviation field names in the first row of the remaining four columns, respectively, to form an attitude deviation matrix with clear row and column meanings. Hydraulic cylinder pressure data, hydraulic cylinder displacement data, foundation contact state, and attitude deviation matrix are extracted from the construction status dataset. A correspondence is established between changes in hydraulic state and changes in attitude deviation. An attitude sensitivity matrix is ​​generated based on this correspondence. Specifically, the generation of the attitude sensitivity matrix based on the correspondence is as follows: The construction status dataset is divided into several continuous sampling windows along the time axis. Each sampling window is bounded by the time period during which the hydraulic actuator keeps the input constant until the next adjustment action takes effect. Within any sampling window, the pressure increment and displacement increment between the start and end points of the window are calculated for each hydraulic cylinder, generating an independent hydraulic status increment vector for each cylinder. Simultaneously extract the lateral offset increment, longitudinal offset increment, roll deviation increment, and pitch deviation increment between the start and end points of the attitude deviation matrix within the same sampling window, combine them into an attitude deviation increment vector, establish a one-to-one correspondence between the hydraulic state increment vector and the attitude deviation increment vector, construct a hydraulic-attitude increment pair sample set, accumulate the sample set for each hydraulic cylinder channel, and use the least squares linear fitting method to obtain the linear increment response coefficient of the cylinder channel to lateral offset, longitudinal offset, roll deviation, and pitch deviation. After the coefficients of all hydraulic cylinder channels are calculated, the response coefficients of each channel are filled into the corresponding positions according to the column order of the attitude deviation matrix, and an attitude sensitivity matrix is ​​constructed with the attitude deviation direction as the row and the hydraulic cylinder channel as the column. A construction attitude transfer topology is constructed based on the attitude sensitivity matrix. This topology includes hydraulic actuator nodes, foundation support nodes, mast attitude nodes, power head attitude nodes, and axis deviation nodes. The consistency of response direction, stability of response delay, and continuity of response amplitude between nodes are used as the validity criteria for topology edges. Specifically, the construction of the construction attitude transfer topology based on the attitude sensitivity matrix is ​​as follows: Based on the attitude sensitivity matrix, each hydraulic cylinder channel is mapped to a hydraulic actuator node. The four types of attitude deviations—lateral, longitudinal, roll, and pitch—are mapped to mast attitude nodes, power head attitude nodes, or axis deviation nodes. The support stiffness distribution is mapped to a foundation support node. Hydraulic actuator nodes and foundation support nodes are connected according to the same support orientation. Then, based on the consistency of the response coefficient sign, they are connected to mast attitude nodes or power head attitude nodes, and finally to axis deviation nodes. The response time difference of nodes within a continuous sampling window is statistically analyzed. Only the connections that fall within the time delay tolerance and whose response amplitude variance is lower than a set threshold are retained, forming a multi-level connected path that satisfies the requirements of directional consistency, time delay stability, and amplitude continuity, i.e., the construction attitude transfer topology. The construction attitude transfer topology is screened for observable paths. Topology paths that stably transfer from the hydraulic actuator node through the foundation support node, mast attitude node, or power head attitude node to the axis deviation node are retained. Topology paths with repeated response direction switching, response delays exceeding the preset sampling period, or discontinuous response amplitudes are removed. The hydraulic control direction and attitude deviation direction corresponding to the retained topology paths are combined into an observable subspace. Specifically, the combination of the hydraulic control direction and attitude deviation direction corresponding to the retained topology paths into an observable subspace is as follows: Read the starting hydraulic actuator node, ending axis deviation node, and path direction label of each preserved topology path, extract the hydraulic cylinder number, extension direction, retraction direction, or holding direction corresponding to the starting hydraulic actuator node, and generate a hydraulic control direction index. Extract the lateral offset, longitudinal offset, roll deviation, or pitch deviation corresponding to the endpoint axis deviation node, generate an attitude deviation direction index, bind the hydraulic control direction index and attitude deviation direction index corresponding to the same retained topology path into a group of observable mapping units, repeat the process for all retained topology paths, delete duplicate observable mapping units, and sort them according to construction depth layer, hydraulic cylinder number, and attitude deviation direction; define the sorted set of observable mapping units as an observable subspace. The extension direction, retraction direction, and holding direction of each hydraulic actuator are respectively set as control direction units. Each control direction unit is matched with a topological path in the observable subspace. When a control direction unit has a corresponding observable topological path, the control direction unit is marked as a passable direction. Path allocation weights are generated based on the response amplitude continuity, response delay stability, and foundation support stability of the corresponding topological path. When a control direction unit does not have a corresponding observable topological path, the control direction unit is marked as a shielded direction. The passable direction, shielded direction, and path allocation weights are combined into a topological observability projection operator. Specifically, the combination of the passable direction, shielded direction, and path allocation weights into a topological observability projection operator is as follows: For each hydraulic cylinder, the extension direction, retraction direction, and holding direction are sequentially numbered, and a control direction index table is established. A weight field and a masking flag field are reserved in the index table. Following the order of the control direction index table, the topological path corresponding to the support orientation in the observable subspace is retrieved. If a matching observable topology path exists, the masking flag field in the corresponding row of the control direction is written as passable, and weight values ​​are assigned according to three indicators: response amplitude variance, time delay fluctuation range, and foundation support stiffness level of the topology path. If no matching observable topology path exists, write "mask" in the masking flag field, set the weight field to zero, and after completing the weight filling and masking flags for all control direction index tables, write the weight values ​​of the passable directions into the weight vector according to the row order in the table, write zero to the corresponding position of the masked direction, and generate the weight vector. Write the passable and maskable markers into the direction mask vector in the same row order, and encapsulate the weight vector and the direction mask vector into one to form a topological observability projection operator.

[0025] In this embodiment, generating the constraint control input located in the observable subspace and constructing an attitude memory field based on historical construction attitude deviations includes: The control inputs of the hydraulic actuator are obtained, including the extension control amount, retraction control amount, and holding control amount of each hydraulic cylinder; The control inputs of the hydraulic actuator are filtered by the topological observability projection operator, retaining the control inputs that belong to the passable direction and removing the control inputs that belong to the shielded direction. Based on the path assignment weights in the topological observability projection operator, the retained control inputs are weighted to generate constrained control inputs located within the observable subspace. Specifically, the weighting of the retained control inputs to generate constrained control inputs within the observable subspace is performed as follows: Each control input is read according to the hydraulic control direction index order and matched with the path allocation weight at the same index position in the topological observability projection operator. If the path allocation weight is greater than zero, the control input is multiplied by the corresponding weight value to obtain the weighted control quantity. If the path allocation weight is zero, the control input is set to zero and marked as a masked control quantity. All weighted control quantities are rearranged according to the hydraulic cylinder number and control direction to form a control vector consistent with the actual channel sequence of the hydraulic actuator. The control vector is categorized by direction, and the weighted control quantities of the same cylinder's extension direction, retraction direction, and holding direction are accumulated to obtain the final extension, retraction, and holding amounts of the cylinder. The extension, retraction, and holding amounts of all cylinders are amplitude-verified. The portion exceeding the hydraulic system safety threshold is proportionally reduced to within the threshold. The holding direction quantity is adjusted synchronously with the threshold ratio to ensure that the control quantity is within the executable range, forming a constrained control input that satisfies the topological constraints and has a limited amplitude. An attitude memory unit is established according to the construction depth layer, attitude deviation direction, and deviation source. The attitude deviation direction includes the lateral offset direction, longitudinal offset direction, roll deviation direction, and pitch deviation direction. The deviation source includes foundation contact state deviation, construction load state deviation, and drill rod stress state deviation. Specifically, the attitude memory unit is established according to the construction depth layer, attitude deviation direction, and deviation source as follows: The attitude deviation matrix, which has been stratified and marked, is grouped according to the depth layer number. Within each depth layer, it is further subdivided into subgroups according to four attitude deviation directions: lateral offset, longitudinal offset, roll deviation, and pitch deviation. The ground contact state, construction load state, and drill pipe stress state corresponding to the depth layer are read. The primary source of deviation for each of the four attitude deviation directions is determined: if the deviation in the same direction is highly correlated with the change in support stiffness, it is marked as ground contact state deviation; if it is synchronized with the feed pressure or power head torque fluctuation, it is marked as construction load state deviation; if it is consistent with the direction of torque peak and occurs continuously, it is marked as drill pipe stress state deviation. Based on the triple index of depth layer number, attitude deviation direction, and deviation source, corresponding attitude memory units are generated sequentially. In each attitude memory unit, a cumulative count field, a decay count field, a self-locking mark field, and a latest deviation amplitude field are preset. Write the attitude deviation matrix of the current construction cycle into the corresponding attitude memory unit, and accumulate and mark the attitude deviations that occur continuously in the same construction depth layer and have the same direction, mark the attitude deviations that change direction, and mark the attitude deviations that are self-locked by the shielding direction. The attitude memory field is formed by combining attitude memory units that have completed cumulative marking, attenuation marking, and self-locking marking within each construction depth layer. The historical deviation trend in the attitude memory field is then updated based on the attitude deviation matrix of the current construction cycle. Specifically, the combination of attitude memory units that have completed cumulative marking, attenuation marking, and self-locking marking within each construction depth layer to form the attitude memory field is as follows: Traversing from top to bottom according to the depth layer numbering, within each depth layer, in the order of lateral offset, longitudinal offset, roll deviation, and pitch deviation, retrieve the attitude memory units that have completed accumulation, attenuation, or self-locking marking, and establish a memory linked list for the depth layer. Arrange the attitude memory units in the four directions according to the source of deviation, prioritizing the insertion of foundation contact state deviation units, followed by construction load state deviation units, and finally drill pipe stress state deviation units. The linked list nodes are sequentially written with the corresponding unit's accumulated count value, attenuation count value, and self-locking mark. Record the state and the latest deviation amplitude. After completing the construction of the current depth layer linked list, connect the head pointer of the linked list to the tail pointer of the previous depth layer linked list to form a vertically connected multi-level linked list structure. After traversing all depth layers, write the attitude memory field head mark at the head end of the linked list and the attitude memory field tail mark at the tail end of the linked list. Record the total number of depth layers in the head mark and the total number of nodes in the tail mark. The final multi-level connected linked list is the attitude memory field, which includes the depth layer index, attitude deviation direction index, deviation source index, and various deviation evolution marks.

[0026] In this embodiment, generating the attitude dynamic response prediction result includes: An adaptive Hammerstein model for attitude memory field modulation is constructed, comprising an attitude memory field index modulation layer, a static nonlinear input mapping layer, and a ground contact dynamic response layer. The constraint control input is input to the attitude memory field index modulation layer, and the corresponding attitude memory unit is read from the attitude memory field according to the current construction depth layer, attitude deviation direction and hydraulic actuator type. Based on the cumulative update flag, decay update flag, and self-locking flag in the attitude memory unit, memory modulation parameters for the corresponding hydraulic control channel are generated. The memory modulation parameters include input gain modulation parameters, directional offset modulation parameters, input limiting modulation parameters, and response hysteresis modulation parameters. The constraint control input and memory modulation parameters are input to the static nonlinear input mapping layer. The static nonlinear input mapping layer adjusts the input gain of the corresponding hydraulic control channel according to the input gain modulation parameters, corrects the correction direction of the corresponding hydraulic control channel according to the direction bias modulation parameters, and limits the input amplitude of the corresponding hydraulic control channel according to the input limiting modulation parameters to generate the intermediate control quantity after memory modulation. The virtual construction reference field is input into the static nonlinear input mapping layer. Based on the target construction axis corresponding to the current construction depth layer, the intermediate control quantities are screened for axis approach direction. The intermediate control quantities that approach the target construction axis are retained, and the intermediate control quantities that deviate from the target construction axis are suppressed, thereby generating axis constraint control quantities. The axis constraint control quantity is input into the foundation contact dynamic response layer, and the foundation contact state is used as the dynamic response parameter. The foundation contact dynamic response layer generates the attitude dynamic response prediction result corresponding to the current construction cycle based on the foundation support stability attribute, load disturbance attribute, drill rod force offset attribute and response hysteresis modulation parameter. The attitude memory field index modulation layer includes: Control Channel Index Unit: Reads the hydraulic cylinder number, extension direction, retraction direction, holding direction, and current construction depth layer from the constraint control input, and generates a control channel index; Attitude memory unit matching unit: Based on the control channel index, attitude deviation direction and deviation source, it retrieves the corresponding attitude memory unit in the attitude memory field; Memory marker reading unit: Reads the cumulative update marker, decay update marker, self-locking marker, and latest deviation amplitude from the attitude memory unit; Modulation parameter generation unit: generates input gain modulation parameters based on cumulative update flags, generates directional bias modulation parameters based on attenuation update flags, generates input limiting modulation parameters based on self-locking flags, and generates response hysteresis modulation parameters based on the latest deviation amplitude; Parameter binding unit: Binds the input gain modulation parameter, directional offset modulation parameter, input limiting modulation parameter and response hysteresis modulation parameter to the corresponding hydraulic control channel to form memory modulation control data; After receiving the constraint control input, the attitude memory field index modulation layer first determines the depth layer, control direction, and hydraulic cylinder number corresponding to each hydraulic control channel by the control channel index unit, and then the attitude memory unit matching unit searches for the corresponding attitude memory unit in the attitude memory field; the memory mark reading unit reads the accumulation, attenuation, and self-locking states in the attitude memory unit; the modulation parameter generation unit converts the above states into the gain, direction, limiting, and hysteresis parameters of the corresponding hydraulic control channel; the parameter binding unit binds the modulation parameters to the constraint control input one by one and outputs the memory modulation control data. The static nonlinear input mapping layer includes: Input gain mapping unit: Adjusts the control amplitude of the corresponding hydraulic control channel according to the input gain modulation parameters; Directional offset correction unit: corrects the correction direction of the corresponding hydraulic control channel according to the directional offset modulation parameters; Amplitude limiting unit: limits the maximum control amplitude of the corresponding hydraulic control channel according to the input amplitude limiting modulation parameters; Reference field approach determination unit: Reads the target construction axis in the virtual construction reference field and determines whether the current control direction makes the actual construction axis approach the target construction axis; Intermediate control quantity generation unit: Combines the control quantities after gain adjustment, direction correction, amplitude limiting constraint, and approach determination into an intermediate control quantity. After receiving the memory modulation control data, the static nonlinear input mapping layer first amplifies or compresses the control amplitude of each hydraulic control channel. The direction offset correction unit then corrects the control direction with historical offset trends. The amplitude limiting constraint unit limits the amplitude of the control quantity in the self-locking direction or high-risk direction. The reference field approach determination unit reads the target construction axis corresponding to the current depth layer, retains the control quantity that can make the actual construction axis approach the target construction axis, and suppresses the control quantity that makes the actual construction axis deviate from the target construction axis. The intermediate control quantity generation unit rearranges the retained control quantities according to the hydraulic cylinder channel sequence to generate intermediate control quantities. The foundation contact dynamic response layer includes: Foundation condition coding unit: Reads the support stability attribute, load disturbance attribute and drill rod force offset attribute in the foundation contact state, and generates the foundation response state for the current construction cycle; Dynamic response parameter update unit: Updates the dynamic response parameters corresponding to each hydraulic control channel based on the foundation response state and response hysteresis modulation parameters; Linear dynamic recursive unit: It takes the intermediate control quantity as input and calculates the response change of each hydraulic control channel to the attitude deviation according to the updated dynamic response parameters; Multi-channel response fusion unit: merges the response changes of each hydraulic control channel according to lateral offset, longitudinal offset, roll deviation and pitch deviation; Prediction result output unit: Generates attitude dynamic response prediction results for the current construction cycle based on the merged response changes; After receiving the intermediate control quantity, the foundation contact dynamic response layer first converts the current foundation support stability, construction load disturbance, and drill rod force offset into the foundation response state; the dynamic response parameter update unit adjusts the response speed, response amplitude, and response lag of each hydraulic control channel according to the foundation response state; the linear dynamic recursion unit calculates the attitude change caused by the intermediate control quantity under the current foundation conditions; the multi-channel response fusion unit merges the attitude changes generated by each hydraulic cylinder channel into the corresponding attitude deviation direction; and the prediction result output unit outputs the attitude dynamic response prediction result of the actual construction axis relative to the target construction axis in the next construction cycle.

[0027] In this embodiment, updating the virtual construction reference field, topological observable subspace, and attitude memory field based on the corrected construction status data includes: The attitude dynamic response prediction results are compared with the attitude deviation matrix to determine the correction direction and correction magnitude corresponding to lateral offset, longitudinal offset, roll attitude deviation, and pitch attitude deviation, and initial hydraulic correction control quantities are generated. Specifically, the generation of initial hydraulic correction control quantities is as follows: The dynamic response prediction results of the attitude are compared with the attitude deviation matrix in the same construction cycle item by item. The sign relationship between the predicted attitude change and the actual deviation residual amount is calculated in the four directions of lateral, longitudinal, roll and pitch. If the residual amount is in the same direction as the predicted change, the direction still needs to be corrected by adding a correction input. If the residual amount is in the opposite direction to the predicted change, the direction can be reduced or the existing control input can be maintained. Based on the response weights of each hydraulic cylinder channel in the attitude sensitivity matrix to the four attitude directions, the two hydraulic actuation channels with the greatest positive impact on the residual quantity are selected as the primary correction channels. The unity gain values ​​of the control quantities in the extension direction, retraction direction, and holding direction are read. For each primary correction channel, the control amplitude is amplified or reduced proportionally according to the magnitude of the residual quantity. The direction depends on the sign of the residual quantity: if the residual quantity is positive, the extension direction is selected; if the residual quantity is negative, the retraction direction is selected. If the absolute value of the residual quantity is lower than the threshold, the direction control quantity is set to the holding direction. Search for secondary hydraulic channels in the same attitude direction, and supplement auxiliary control quantities in descending order of their response weights until the absolute value of the predicted remaining quantity drops to within the target tolerance or the control amplitude of the hydraulic actuation channel reaches the safety limit. Write the control commands of the preferred correction channel and auxiliary channel into the control vector according to the hydraulic cylinder number and direction code. Write the holding command into the control vector for channels that are not involved in the correction. The overall vector is the initial hydraulic correction control quantity. The initial hydraulic correction control quantity is directionally filtered using a topological observability projection operator. Correction control quantities belonging to the observable subspace are retained, while those belonging to unobservable directions are deleted, generating executable hydraulic correction control quantities. Specifically, the directional filtering of the initial hydraulic correction control quantity using the topological observability projection operator is as follows: Read the orientation mask vector and weight vector in the topological observability projection operator. Correspond the initial hydraulic correction control quantity to the hydraulic cylinder number and control direction one by one. Traverse the control vector one by one. For control quantities marked as passable by the orientation mask, retain the original amplitude and multiply it by the weight value at the same index position. For control quantities marked as masked, set them to zero directly. After the traversal is completed, the control quantities of the extension, retraction and holding directions in the same hydraulic cylinder are merged again. If the total amplitude after merging exceeds the safety limit of the hydraulic system, it is scaled up to the safety limit. The extension, retraction and holding amounts of all hydraulic cylinders are rearranged in the preset channel order to form an executable hydraulic correction control quantity that only contains observable directions and whose amplitude meets the safety limit. The executable hydraulic correction control quantity is decomposed into the extension control quantity, retraction control quantity, and holding control quantity of each hydraulic actuator, and a hydraulic execution sequence is generated according to the construction depth layer and the direction of attitude deviation. Specifically, the hydraulic execution sequence generated according to the construction depth layer and the direction of attitude deviation is as follows: Read the extension, retraction and holding commands of each hydraulic cylinder in the executable hydraulic correction control quantity, map them to the current construction depth layer of the pile driver according to the position of the outrigger or track to which the cylinder belongs, and sort the remaining quantities in the four directions of lateral offset, longitudinal offset, roll deviation and pitch deviation in the same depth layer by amplitude. The deviation direction with the largest amplitude is given the highest execution priority, and the other directions decrease in order. In each depth layer, hydraulic cylinder commands associated with the deviation direction are selected sequentially according to the priority of the deviation direction. The extension command is placed before the retraction command, and the holding command is placed at the end of the queue, forming a direction priority queue for the corresponding depth layer. After the direction priority queues of all depth layers are completed, the direction priority queues are sequentially connected from the top of the pile to the bottom of the pile to obtain a hydraulic execution sequence that takes into account both the depth layer order and the priority of the attitude deviation direction. The hydraulic execution sequence is checked for safety gaps. If adjacent commands may cause hydraulic channel conflicts, a holding time gap is inserted between commands or the order of adjacent commands is swapped to ensure that the execution order is safe and feasible, and the final hydraulic execution sequence is generated. The hydraulic actuator is driven to perform attitude correction according to the hydraulic execution sequence, and multi-source state data after correction is collected during the execution process to generate corrected construction state data. Specifically, the attitude correction is performed by driving the hydraulic actuator according to the hydraulic execution sequence as follows: The hydraulic execution sequence is imported into the controller's task buffer. An execution counter and safety interlock flag are set. The controller sequentially reads the first instruction, sending extension, retraction, or holding control signals to the corresponding hydraulic valve group to drive the designated cylinder. Simultaneously, real-time feedback monitoring is initiated, continuously reading cylinder pressure and displacement sensor data. When the sensor feedback reaches the instruction's target value and the holding time exceeds a preset dwell threshold, the instruction is considered complete. The controller automatically calls the next instruction, driving the corresponding hydraulic cylinder according to the same process. When multiple cylinders are in parallel action, the controller checks the interlock flag. If adjacent cylinder actions may cause structural interference, a holding delay is inserted, waiting for the previous cylinder to stabilize before execution. After the entire execution sequence is completed, the controller triggers global attitude stability detection, confirming that the real-time deviations in the lateral, longitudinal, roll, and pitch directions all fall within the set thresholds before ending the correction cycle. Based on the corrected construction status data, the actual construction axis and attitude deviation matrix are recalculated, and the virtual construction reference field, topological observable subspace, and attitude memory field are updated based on the recalculated actual construction axis and attitude deviation matrix.

[0028] Example 1: In a pile driver drilling operation in a soft, heterogeneous stratum, the target drilling depth was 30m, the designed pile diameter was 800mm, the initial alignment error was 6.8mm laterally and 7.4mm longitudinally, the initial mast roll deviation was 0.08°, and the pitch deviation was 0.07°. During construction, the system received multi-source status data from vehicle body attitude sensors, mast attitude sensors, power head attitude sensors, hydraulic cylinder pressure sensors, hydraulic cylinder displacement sensors, power head torque acquisition unit, feed pressure acquisition unit, drill rod penetration depth acquisition unit, vibration acceleration acquisition unit, and laser alignment unit. The sampling frequency was 20Hz, and a total of 216,000 sets of valid data were collected in one complete construction cycle.

[0029] After the data enters the processing flow, the system first synchronizes the time of all sensor data, correcting the maximum time deviation from the original 86ms to within 9ms. Then, a sliding window method is used to remove outliers. From 216,000 sets of data, 3,784 sets of data with pressure mutations, vibration spikes, and attitude jumps are removed, with an outlier rate of 1.75%. After low-pass filtering, the standard deviation of mast attitude noise is reduced from 0.031° to 0.012°, and the hydraulic cylinder pressure fluctuation noise is reduced from 0.18MPa to 0.07MPa, finally generating the construction status dataset.

[0030] During the construction of the virtual construction benchmark field, the system divided the 30m construction depth into 60 continuous depth layers along the design pile centerline, with each layer having a thickness of 0.5m. When drilling reached approximately 9.5m, the torque of the power head increased from 52.6kN·m to 68.9kN·m, and the displacement of the right rear support cylinder increased by 3.7mm within 6 consecutive seconds. The system further subdivided this section into 4 sub-depth layers. Within each depth layer, 24 circular candidate axis nodes were generated, and 2 additional outward-expanding nodes were generated in the force offset direction, resulting in a total of 1548 candidate axis nodes across all depth layers. The system wrote support stability attributes, load disturbance attributes, drill rod force offset attributes, and historical deviation accumulation attributes for each candidate axis node. The average support stability attribute value in the right rear direction between 9.5m and 12m was 0.42, significantly lower than the average of 0.76 in other directions; the average torque disturbance attribute value in the same section was 0.69, higher than the average of 0.31 in the stable section. Based on this, the system generates a depth layer reference field and connects multiple depth layer reference fields in depth order to form a virtual construction reference field.

[0031] During the generation of the target construction axis, the system reads the node field values ​​layer by layer from the pile top to the pile bottom. In the 0m to 9m section, the target construction axis basically fits the design pile centerline, with an average offset of 4.2mm. In the 9.5m to 12m section, the system avoids the right rear low-support stable node and selects a nearby candidate node with lower load disturbance, resulting in a maximum mild offset of 9.6mm relative to the design centerline for the target construction axis. After 12m, as the support stability property recovers to above 0.71, the target construction axis gradually returns to the design centerline. Simultaneously, the system calculates the theoretical mast axis based on the mast attitude and power head attitude, calculates the drill pipe force offset based on torque, feed pressure, and vibration acceleration, and calculates the foundation settlement offset based on cylinder pressure and displacement. When drilling reaches 10.2m, the lateral offset of the theoretical mast axis is 13.4mm, the drill pipe force offset is 7.8mm, and the foundation settlement offset is 5.6mm. After combination, the actual construction axis lateral offset reaches 26.8mm.

[0032] The system compares the actual construction axis with the target construction axis to generate an attitude deviation matrix. Taking a 10.2m depth layer as an example, the attitude deviation matrix records a lateral offset of 17.2mm, a longitudinal offset of 11.5mm, a roll deviation of 0.23°, and a pitch deviation of 0.19°. Subsequently, the system establishes an attitude sensitivity matrix based on the hydraulic cylinder pressure increment, displacement increment, and attitude deviation increment. In 80 consecutive sampling windows, the average response of the right rear outrigger extension channel to the roll deviation was 0.006° / mm, but the response delay fluctuated between 0.4s and 1.8s, with an amplitude variance of 0.014; the average response of the left front outrigger retraction channel to the roll deviation was 0.019° / mm, with a response delay stable between 0.5s and 0.7s, and an amplitude variance of 0.004. Based on this, the system constructs a construction attitude transfer topology map, which includes 12 hydraulic execution nodes, 4 foundation support nodes, 2 mast attitude nodes, 2 power head attitude nodes, 4 axis deviation nodes, and 96 initial topology paths.

[0033] During observable path screening, the system removed 18 paths with repeatedly switching response directions, 21 paths with unstable response delays, and 14 paths with discontinuous response amplitudes, ultimately retaining 43 observable topology paths. The right rear outrigger extension direction was marked as a shielded direction due to unstable response delays, while the left front outrigger retraction direction, the right front outrigger extension direction, and the rear outrigger holding direction were marked as passable directions. The system uses passable directions, shielded directions, and path assignment weights to form a topology observability projection operator, where the left front outrigger retraction channel has a weight of 0.84, the right front outrigger extension channel has a weight of 0.76, and the right rear outrigger extension channel has a weight of 0.

[0034] During the construction of the attitude memory field, the system establishes attitude memory units according to depth layer, attitude deviation direction, and deviation source. In the 9.5m to 12m section, roll deviation occurs continuously for 11 sampling periods, and the source of the deviation is determined to be ground contact state deviation; the system marks this direction as cumulative update. The right rear support direction is shielded by the topological observability projection operator for 5 consecutive periods; the system marks it as self-locking. The longitudinal offset changes from positive to negative within 3 adjacent periods; the system marks it as decay update. After marking, 64 attitude memory units are formed in this section, including 19 cumulative update units, 27 decay update units, and 6 self-locking units.

[0035] After the adaptive Hammerstein model of attitude memory field modulation starts running, the attitude memory field index modulation layer reads the current depth layer, deviation direction, and hydraulic actuator type to generate memory modulation parameters. For the right rear self-locking direction, the input gain is compressed to the original 0.22, and the input limit is restricted to 35% of the normal upper limit; for the left front passable direction, the input gain is adjusted to 1.18. The static nonlinear input mapping layer further combines the virtual construction reference field to filter the approach direction and suppress the control amount that deviates from the target construction axis. Based on the support stability attribute of 0.42, the load disturbance attribute of 0.69, and the drill rod force offset attribute of 0.58, the foundation contact dynamic response layer predicts that the roll deviation in the next construction cycle can be reduced from 0.23° to 0.11°, and the lateral offset can be reduced from 17.2mm to 8.9mm.

[0036] The controller generates hydraulic correction control quantities based on the prediction results and performs attitude correction. In the first correction, the left front outrigger retracts by 1.6 mm, the right front outrigger extends by 1.2 mm, and the right rear outrigger remains stationary. After execution, the actual roll deviation decreases from 0.23° to 0.13°, and the lateral offset decreases from 17.2 mm to 10.1 mm. In the second correction, the system continues to maintain the right rear shield, the left front outrigger retracts by 0.8 mm, and the right front outrigger extends by 0.6 mm. After execution, the roll deviation decreases to 0.08°, and the lateral offset decreases to 6.4 mm. The total time for both corrections is 14.6 seconds, during which the peak hydraulic pressure is 12.9 MPa, and no reverse overshoot occurs.

[0037] To verify the effectiveness, this embodiment compares the method of the present invention with the traditional instantaneous error leveling method under the same construction parameters. The training sample size for both methods is 600 sets, and the test construction window is 200 sets. The traditional method does not construct a virtual construction reference field or a topological observability projection operator; it directly outputs hydraulic control quantities based solely on the mast tilt angle. Within the same batch of tests, the maximum lateral offset of the traditional method was 34.7 mm, while that of this invention was 16.3 mm; the maximum longitudinal offset of the traditional method was 29.8 mm, while that of this invention was 14.1 mm; the maximum roll deviation of the traditional method was 0.46°, while that of this invention was 0.21°; the maximum pitch deviation of the traditional method was 0.41°, while that of this invention was 0.19°; the number of unobservable direction control outputs of the traditional method was 32, while that of this invention was 4; the number of overcompensation times of the traditional method was 11, while that of this invention was 2; the average single-test stabilization time of the traditional method was 19.4 s, while that of this invention was 11.7 s; the final comprehensive deviation of the pile bottom of the traditional method was 37.6 mm, while that of this invention was 15.8 mm; the peak hydraulic pressure during the construction cycle of the traditional method was 15.8 MPa, while that of this invention was 13.1 MPa; and the attitude stability pass rate of the traditional method was 84.5%, while that of this invention was 96.2%.

[0038] 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 dynamic correction of pile driver construction posture based on multi-source sensing fusion, characterized in that, include: Collect multi-source status data during the pile driver construction process, preprocess the multi-source status data, and generate a construction status dataset; A virtual construction reference field is constructed based on the construction status dataset. According to the foundation contact state, construction load state, drill rod stress state and historical attitude deviation trend, the target construction axis corresponding to the current construction stage is generated, and the actual construction axis is calculated. An attitude deviation matrix is ​​constructed based on the spatial deviation between the actual construction axis and the target construction axis. An attitude sensitivity matrix is ​​established based on the correspondence between the construction state dataset and the attitude deviation matrix. Singular value decomposition is performed on the attitude sensitivity matrix to extract the observable subspace and construct a topological observability projection operator. The control input of the hydraulic actuator is projected and mapped using the topological observability projection operator to generate the constrained control input located in the observable subspace. An attitude memory field is constructed based on historical construction attitude deviations, and the attitude memory field is used to accumulate and update the historical deviation trend. An adaptive Hammerstein model for attitude memory field modulation is constructed, with constraint control input as input variable, attitude memory field and virtual construction reference field as nonlinear module modulation parameters, and foundation contact state as linear dynamic module modulation parameters, to generate attitude dynamic response prediction results. Based on the attitude dynamic response prediction results, hydraulic correction control quantities are generated and attitude correction is performed. The virtual construction reference field, topological observable subspace, and attitude memory field are updated based on the corrected construction status data.

2. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The multi-source state data includes vehicle body posture data, mast posture data, power head posture data, hydraulic cylinder pressure data, hydraulic cylinder displacement data, drill rod penetration depth data, power head torque data, feed pressure data, vibration acceleration data, laser alignment data, and visual axis data.

3. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The preprocessing of multi-source status data to generate a construction status dataset includes: Multi-source state data is timestamped to obtain time-synchronized multi-source state data. A sliding window is used to remove outliers from the time-synchronized multi-source state data to obtain valid state data. The valid state data is low-pass filtered to obtain filtered state data. The filtered state data is resampled according to the same sampling period to generate a construction state dataset.

4. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The process of generating the target construction axis corresponding to the current construction stage and calculating the actual construction axis includes: Hydraulic cylinder pressure data, hydraulic cylinder displacement data, power head torque data, feed pressure data, drill rod penetration depth data, vibration acceleration data, mast attitude data, and power head attitude data are extracted from the construction status dataset to form the foundation contact status, construction load status, drill rod stress status, and historical attitude deviation trend, respectively. Using the design pile centerline as the initial reference, multiple continuous depth layers are divided in the direction of pile driver construction depth. Multiple candidate axis nodes are generated in each depth layer. Each candidate axis node is marked with support stability attribute, load disturbance attribute, drill rod force offset attribute and historical deviation accumulation attribute. Based on the support stability attribute, load disturbance attribute, drill pipe force offset attribute and historical deviation accumulation attribute of each candidate axis node, a node field value is generated. The candidate axis nodes and their node field values ​​in the same depth layer are combined to form a depth layer reference field. Multiple depth layer reference fields are connected in the order of construction depth to form a virtual construction reference field. In the virtual construction reference field, candidate axis nodes that satisfy the highest support stability attribute, lowest load disturbance attribute, lowest drill rod force offset attribute and limited historical deviation accumulation attribute are selected sequentially according to the construction depth direction. The selected candidate axis nodes in adjacent depth layers are connected to generate the target construction axis corresponding to the current construction stage. The theoretical axis of the mast is determined based on the mast posture data and power head posture data in the construction status dataset. The force offset of the drill rod is determined based on the drill rod penetration depth data, power head torque data, feed pressure data, and vibration acceleration data. The ground settlement offset is determined based on the hydraulic cylinder pressure data and hydraulic cylinder displacement data. The actual construction axis is obtained by combining the theoretical axis of the mast, the force offset of the drill rod, and the ground settlement offset.

5. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The process of performing singular value decomposition on the attitude sensitivity matrix, extracting the observable subspace, and constructing a topological observability projection operator includes: The differences between the actual construction axis and the target construction axis in lateral offset, longitudinal offset, roll attitude deviation and pitch attitude deviation are combined into an attitude deviation matrix, and the attitude deviation matrix is ​​layered and marked according to the construction depth. Hydraulic cylinder pressure data, hydraulic cylinder displacement data, foundation contact state and attitude deviation matrix are extracted from the construction status dataset. The correspondence between hydraulic state changes and attitude deviation changes is established, and an attitude sensitivity matrix is ​​generated based on the correspondence. The construction attitude transfer topology is constructed based on the attitude sensitivity matrix. The construction attitude transfer topology includes hydraulic actuator nodes, foundation support nodes, mast attitude nodes, power head attitude nodes and axis deviation nodes. The consistency of response direction, stability of response delay and continuity of response amplitude between nodes are used as the validity criteria for the topology edges. The observable paths of the construction attitude transfer topology are screened, and the topology paths that can be stably transferred from the hydraulic actuator node to the foundation support node, mast attitude node or power head attitude node are retained. Topology paths that repeatedly switch response directions, have response delays exceeding the preset sampling period range or have discontinuous response amplitudes are eliminated. The hydraulic control direction and attitude deviation direction corresponding to the retained topology paths are combined into an observable subspace. The extension direction, retraction direction, and holding direction of each hydraulic actuator are set as control direction units. Each control direction unit is matched with a topological path in the observable subspace. When a control direction unit has a corresponding observable topological path, the control direction unit is marked as a passable direction. Path allocation weights are generated based on the continuity of response amplitude, stability of response delay, and stability of foundation support of the corresponding topological path. When a control direction unit does not have a corresponding observable topological path, the control direction unit is marked as a shielded direction. The passable direction, shielded direction, and path allocation weights are combined into a topological observability projection operator.

6. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The generation of constraint control inputs located within the observable subspace, and the construction of an attitude memory field based on historical construction attitude deviations, includes: The control inputs of the hydraulic actuator are obtained, including the extension control amount, retraction control amount, and holding control amount of each hydraulic cylinder; The control inputs of the hydraulic actuator are filtered by the topological observability projection operator, retaining the control inputs that belong to the passable direction and removing the control inputs that belong to the shielded direction. Based on the path assignment weights in the topological observability projection operator, the retained control inputs are weighted to generate constrained control inputs located in the observable subspace. An attitude memory unit is established according to the construction depth layer, attitude deviation direction and deviation source. The attitude deviation direction includes the lateral offset direction, longitudinal offset direction, roll deviation direction and pitch deviation direction. The deviation source includes the foundation contact state deviation, construction load state deviation and drill rod stress state deviation. Write the attitude deviation matrix of the current construction cycle into the corresponding attitude memory unit, and accumulate and mark the attitude deviations that occur continuously in the same construction depth layer and have the same direction, mark the attitude deviations that change direction, and mark the attitude deviations that are self-locked by the shielding direction. The attitude memory units that have completed the accumulation marking, attenuation marking and self-locking marking in each construction depth layer are combined to form an attitude memory field, and the historical deviation trend in the attitude memory field is updated according to the attitude deviation matrix of the current construction cycle.

7. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The generated attitude dynamic response prediction results include: An adaptive Hammerstein model for attitude memory field modulation is constructed, comprising an attitude memory field index modulation layer, a static nonlinear input mapping layer, and a ground contact dynamic response layer. The constraint control input is input to the attitude memory field index modulation layer, and the corresponding attitude memory unit is read from the attitude memory field according to the current construction depth layer, attitude deviation direction and hydraulic actuator type. Based on the cumulative update flag, decay update flag, and self-locking flag in the attitude memory unit, memory modulation parameters for the corresponding hydraulic control channel are generated. The memory modulation parameters include input gain modulation parameters, directional offset modulation parameters, input limiting modulation parameters, and response hysteresis modulation parameters. The constraint control input and memory modulation parameters are input to the static nonlinear input mapping layer. The static nonlinear input mapping layer adjusts the input gain of the corresponding hydraulic control channel according to the input gain modulation parameters, corrects the correction direction of the corresponding hydraulic control channel according to the direction bias modulation parameters, and limits the input amplitude of the corresponding hydraulic control channel according to the input limiting modulation parameters to generate the intermediate control quantity after memory modulation. The virtual construction reference field is input into the static nonlinear input mapping layer. Based on the target construction axis corresponding to the current construction depth layer, the intermediate control quantities are screened for axis approach direction. The intermediate control quantities that approach the target construction axis are retained, and the intermediate control quantities that deviate from the target construction axis are suppressed, thereby generating axis constraint control quantities. The axis constraint control quantity is input into the foundation contact dynamic response layer, and the foundation contact state is used as the dynamic response parameter. The foundation contact dynamic response layer generates the attitude dynamic response prediction result corresponding to the current construction cycle based on the foundation support stability attribute, load disturbance attribute, drill rod force offset attribute and response hysteresis modulation parameter.

8. The method for dynamic correction of pile driver construction posture based on multi-source sensing fusion according to claim 1, characterized in that, The step of updating the virtual construction reference field, topological observable subspace, and attitude memory field based on the corrected construction status data includes: The dynamic response prediction results of the attitude are compared with the attitude deviation matrix to determine the correction direction and correction magnitude corresponding to the lateral offset, longitudinal offset, roll attitude deviation and pitch attitude deviation, and generate the initial hydraulic correction control quantity. The initial hydraulic correction control quantity is filtered by the topological observability projection operator. The correction control quantity belonging to the observable subspace is retained, and the correction control quantity belonging to the unobservable direction is deleted, thereby generating an executable hydraulic correction control quantity. The executable hydraulic correction control quantity is decomposed into the extension control quantity, retraction control quantity, and holding control quantity of each hydraulic actuator, and a hydraulic execution sequence is generated according to the construction depth layer and the direction of attitude deviation. The hydraulic actuator is driven to perform attitude correction according to the hydraulic execution sequence, and multi-source state data after correction is collected during the execution process to generate corrected construction state data. Based on the corrected construction status data, the actual construction axis and attitude deviation matrix are recalculated, and the virtual construction reference field, topological observable subspace, and attitude memory field are updated based on the recalculated actual construction axis and attitude deviation matrix.

Images