Intelligent detection method for verticality of rotary digging pile foundation based on laser measurement

Abstract

The present application relates to the technical field of intelligent vertical detection of pile foundation, in particular to a rotary pile verticality intelligent detection method based on laser measurement, which comprises the following steps: collecting original displacement data by continuously scanning a rotary descending drill pipe through a pile orifice laser measuring instrument, constructing a dynamic noise benchmark library to complete background noise mode matching and adaptive deduction, and obtaining a pure displacement data stream; calculating a pile body center horizontal projection track, separating low-frequency drift and high-frequency vibration components, inputting the low-frequency components into an attitude evolution model to predict the deviation trend and generate a pre-correction instruction with a time label, distinguishing normal vibration from abnormal vibration caused by hole wall collapse for the high-frequency components and triggering an emergency flag, and fusing to generate a comprehensive control instruction to issue a correction mechanism. The method can realize accurate noise removal, accurate pile body track restoration, early correction and real-time determination of hole wall abnormalities.

Description

Technical Field

[0001] This invention relates to the field of intelligent vertical detection technology for pile foundations, and in particular to an intelligent detection method for the verticality of rotary drilling pile foundations based on laser measurement. Background Technology

[0002] Current methods for detecting the verticality of rotary drilling piles primarily employ single-point laser measurement to acquire drill rod displacement data. Noise processing relies on static methods such as fixed filtering and mean denoising. Verticality detection only collects and determines the overall pile offset, without finely segmenting or parallel processing the displacement signal. The detection process only acquires discrete pile displacement data, failing to achieve continuous scanning and acquisition of the drill rod during its rotating descent. Control command generation depends solely on real-time offset data, without incorporating attitude trend prediction or borehole vibration status assessment.

[0003] Static noise processing methods cannot adapt to the dynamically changing background noise patterns in construction scenarios. Noise in the original pile displacement data cannot be accurately deducted, leading to distortion in the calculation of the pile center trajectory and insufficient reliability of the detection data. Merely detecting the overall offset cannot distinguish the signal characteristics corresponding to changes in pile posture and borehole wall vibration. It cannot predict subsequent deviations based on pile posture change patterns, nor can it determine the risk of borehole wall collapse through vibration signals. Correction command generation lacks precision, and abnormal conditions cannot be identified and responded to in a timely manner.

[0004] This invention aims to achieve dynamic noise pattern matching and adaptive subtraction of the original pile displacement data to obtain a pure displacement data stream without noise interference. At the same time, it separates the low-frequency drift component and the high-frequency vibration component from the projected trajectory of the pile center. Based on the low-frequency drift component, it predicts the pile deviation trend and generates a time-seriesd pre-correction command. It performs pattern recognition on the high-frequency vibration component to distinguish between normal vibration and abnormal vibration due to borehole wall collapse. Finally, it fuses the two types of data to generate a comprehensive control command. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an intelligent detection method for the verticality of rotary drilling pile foundations based on laser measurement.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent detection method for the verticality of rotary drilling pile foundations based on laser measurement, comprising: A laser beam is emitted by a laser measuring instrument installed at the borehole of the rotary drilling pile to continuously scan the surface of the rotating and descending drill rod, thereby collecting the original pile displacement data sequence. A dynamic noise benchmark library is constructed, and the collected original pile displacement data sequence is input into the dynamic noise benchmark library to perform background noise pattern matching and adaptive subtraction to obtain a clean pile displacement data stream. Based on the pure pile displacement data stream, the real-time projection trajectory of the pile center on the horizontal plane is calculated, and the low-frequency drift component and high-frequency vibration component of the pile are separated from the real-time projection trajectory. The low-frequency drift component is input into a preset pile attitude evolution model. The pile attitude evolution model predicts the future deviation trend of the pile based on the current stratum resistance parameters and historical attitude data, and outputs the predicted pile offset direction and offset speed. Based on the predicted pile offset direction and offset speed, a set of pre-correction instruction sequences containing time tags is generated; Simultaneously, pattern recognition is performed on the high-frequency vibration components to determine whether they belong to normal vibration of construction machinery or abnormal vibration caused by hole wall collapse. If an abnormal vibration pattern is identified, an emergency state flag is triggered. By combining the pre-correction instruction sequence with the emergency status flag, the final integrated control instruction is generated and issued to the pile correction execution mechanism in real time.

[0007] As a further aspect of the present invention, a dynamic noise benchmark library is constructed, and the collected original pile displacement data sequence is input into the dynamic noise benchmark library to perform background noise pattern matching and adaptive subtraction to obtain a clean pile displacement data stream. Specific implementation includes: The dynamic noise benchmark library was established by analyzing background noise data of laser measuring instruments acquired in an environment free from construction interference; Before the rotary drilling pile construction begins, the laser measuring instrument is activated and static measurements are performed at a fixed position. Background noise signals are continuously collected for a period of time. The collected background noise signals are analyzed in the time domain and frequency domain to extract a variety of typical background noise waveform patterns, including low-frequency fluctuation patterns caused by environmental airflow disturbances and high-frequency random patterns caused by the instrument's own electronic thermal noise. The extracted typical background noise waveform patterns are classified and stored according to their waveform characteristics, frequency range and occurrence probability to form an initial background noise pattern library. During construction, the original pile displacement data sequence is acquired in real time. At the same time, the background noise pattern library is dynamically retrieved from the background noise pattern library based on the current environmental sensor readings to find the set of background noise patterns that best match the current environmental conditions, which serves as the reference noise for the current moment. The original pile displacement data sequence is compared point by point with the retrieved reference noise at the current moment. By calculating the cross-correlation coefficient, the phase difference and amplitude ratio between the original pile displacement data sequence and the retrieved reference noise at the current moment are determined. Based on the determined phase difference and amplitude ratio, the matching reference noise is synchronously subtracted from the original pile displacement data sequence to complete the initial subtraction of background noise. After initially removing background noise, the data sequence is then subjected to local singularity detection to identify and remove abnormal data points caused by transient strong interference. Then, the data at the removed positions is repaired by using adjacent normal data points through a cubic spline interpolation algorithm, and finally, a continuous and smooth pure pile displacement data stream is output.

[0008] As a further aspect of the present invention, based on the pure pile displacement data stream, the real-time projection trajectory of the pile center on the horizontal plane is calculated, and the low-frequency drift component and high-frequency vibration component of the pile are separated from the real-time projection trajectory, specifically including: From the pure pile displacement data stream, a two-dimensional coordinate sequence of the pile center is extracted according to a fixed time window. The two-dimensional coordinate sequence is composed of displacement values ​​measured by a laser measuring instrument in two mutually perpendicular directions. The two-dimensional coordinate sequence is transformed and unified to a global coordinate system with the design center of the pile hole as the origin, resulting in a discrete set of position points of the pile center point on the horizontal plane. The discrete location points are sorted and connected based on timestamps to form a real-time projection trajectory that characterizes the change of the center position of the pile over time. The real-time projected trajectory is processed using a moving average filter with an adaptive window length. The adaptive window length is dynamically adjusted according to the current drill rod lowering speed. The faster the lowering speed, the shorter the window length. The smooth trajectory obtained after moving average filtering is the low-frequency drift component. The corresponding low-frequency drift component coordinate values ​​are subtracted point by point from the real-time projection trajectory to obtain the residual sequence, which is the high-frequency vibration component. The standard deviation of the high-frequency vibration component is calculated to obtain the vibration intensity index within the current time window.

[0009] As a further aspect of the present invention, the low-frequency drift component is input into a preset pile attitude evolution model. The pile attitude evolution model predicts the future deviation trend of the pile based on current ground resistance parameters and historical attitude data, and outputs the predicted pile offset direction and offset velocity, specifically including: The pile attitude evolution model is a recurrent neural network model based on a long short-term memory network. Its input layer receives the low-frequency drift component sequence arranged in time order, and at the same time receives the current formation resistance parameters reported in real time by the drilling rig sensors. The current formation resistance parameters include drilling pressure and torque. The pile attitude evolution model contains a state memory unit, which is used to store and update the historical characteristics of the pile attitude changes in the previous few time steps. The pile attitude evolution model calculates based on the input low-frequency drift component sequence and the current formation resistance parameters, combined with the historical characteristics stored in the state memory unit, through its internal multi-layer neural network. The output of the multilayer neural network is the predicted coordinates of the pile center position in the next few time steps. The predicted coordinates of the pile center position in the next few time steps are compared with the current pile center position coordinates to calculate the predicted displacement increment of the pile on the horizontal plane along the X and Y axes. Based on the predicted displacement increment, the direction angle of the composite vector is calculated, which is the predicted pile offset direction. At the same time, the ratio of the magnitude of the composite vector to the predicted time span is calculated, which is the predicted pile offset velocity.

[0010] As a further aspect of the present invention, based on the predicted pile offset direction and offset velocity, a set of pre-correction instruction sequences containing time tags is generated, specifically including: A pre-correction command mapping table is established, which defines the basic correction action parameters corresponding to different offset direction intervals and offset speed intervals. The basic correction action parameters include the extension and retraction direction and theoretical stroke of the correction cylinder. Query the pre-correction instruction mapping table, and determine the corresponding basic correction action parameters based on the azimuth interval of the predicted pile offset direction and the velocity threshold interval of the predicted pile offset velocity. Based on the current drill pipe lowering depth and lowering speed, calculate the time delay from issuing the command to the actual action of the correction mechanism. Based on the calculated time delay, add a future time stamp to the basic correction action parameters to form an initial pre-correction command with a future execution time. A feedback correction factor is introduced, which is dynamically calculated based on the deviation between the actual correction effect and the expected effect of the recently executed correction command. The feedback correction factor is multiplied by the theoretical travel in the initial pre-correction command to correct the theoretical travel and obtain the corrected theoretical travel. The corrected theoretical stroke, the extension and retraction direction of the correction cylinder, and the time stamp of the future moment are combined to generate a complete pre-correction instruction. Multiple such instructions are arranged in the order of the time stamps to form the pre-correction instruction sequence.

[0011] As a further aspect of the present invention, pattern recognition is performed on the high-frequency vibration components to determine whether they belong to normal vibration of construction machinery or abnormal vibration caused by hole wall collapse, specifically including: Two types of vibration samples were extracted and labeled from historical construction data. The first type is the normal vibration signal sample generated by the drill bit cutting the soil layer and the operation of the drill power head. The second type is the abnormal vibration signal sample generated by the collapse of the soil layer in the borehole wall impacting the drill rod. Feature extraction was performed on the two types of vibration signal samples. The extracted features included the energy ratio of the vibration signal in a specific frequency band, the zero-crossing rate, the steepness of the waveform envelope, and the probability distribution skewness of the signal amplitude. A support vector machine classifier was trained using the extracted features. The high-frequency vibration components within the current time window are collected in real time, and the same feature extraction operation as the training samples is performed on the collected high-frequency vibration components to obtain a set of real-time vibration feature vectors. The real-time vibration feature vector is input into a trained support vector machine classifier. The support vector machine classifier outputs a classification result label and a corresponding confidence score. If the classification result label is abnormal vibration and the confidence score exceeds the preset confidence threshold, the current vibration is determined to be an abnormal vibration mode, and the emergency state flag is triggered. If the emergency state flag is triggered, the vibration waveform characteristics and the corresponding pile depth information at the trigger time are recorded simultaneously and stored in the historical database as new abnormal samples for subsequent incremental learning of the support vector machine classifier.

[0012] As a further aspect of the present invention, by combining the pre-correction command sequence and the emergency state flag, a final integrated control command is generated and issued to the pile correction execution mechanism in real time, specifically including: A command fusion scheduling center is established, which continuously receives a pre-correction command sequence from the pre-correction command generation module and an emergency status flag from the vibration pattern recognition module. The instruction fusion scheduling center maintains an instruction execution queue. The instruction execution queue stores pre-correction instructions to be executed in the order of the instruction timestamps. When a new pre-correction instruction sequence is received, the instructions in it are merged and inserted into the corresponding position of the instruction execution queue according to the timestamps. The instruction fusion scheduling center monitors the status of the emergency status flag in real time. When the emergency status flag is triggered to an effective state, the instruction fusion scheduling center immediately clears all pre-correction instructions in the current instruction execution queue. After clearing the instruction execution queue, the instruction fusion scheduling center generates a highest priority emergency correction instruction based on the energy magnitude of the high-frequency vibration component when the emergency state flag is triggered. The emergency correction instruction includes action parameters to quickly retract the correction cylinder to a safe position, and marks the emergency correction instruction as to be executed immediately. When the emergency status flag is invalid, the instruction fusion scheduling center retrieves the pre-correction instruction that arrives at the execution time first from the head of the instruction execution queue in chronological order, and uses it as the final integrated control instruction. The generated integrated control commands are sent to the controller of the pile body correction actuator in real time via fieldbus. The controller parses the commands and drives the correction cylinder to perform the corresponding extension and retraction actions.

[0013] As a further aspect of the present invention, after the integrated control command is sent to the pile body correction actuator, the method further includes an online evaluation of the correction action effect and a self-calibration of the model parameters: After the correction actuator starts to move, new pure pile displacement data streams fed back by the laser measuring instrument are continuously collected over a period of time. The displacement response curves corresponding to the correction action are extracted from the new pure pile displacement data streams. Calculate the actual characteristic parameters of the displacement response curve, including the response delay time, the settling time to reach steady state, and the overshoot. Compare the actual characteristic parameters with the expected ideal response characteristic parameters when issuing the integrated control command, and calculate the error between the actual characteristic parameters and the expected ideal response characteristic parameters when issuing the integrated control command. If the error exceeds the allowable range, the actual characteristic parameters, the content of the issued comprehensive control command, and the prediction result of the pile body attitude evolution model at that time will be used together as a training sample and stored in an effect evaluation sample library. Periodically call the recently accumulated sample data in the effect evaluation sample library to fine-tune the neural network connection weights in the pile posture evolution model, so as to reduce the error of the predicted pile offset direction and offset speed in the future. Meanwhile, based on the recent error trends of multiple correction actions, the correspondence between the offset speed range and the basic correction action parameters in the pre-correction command mapping table is dynamically adjusted to make the generated pre-correction command sequence more consistent with the dynamic characteristics of the current strata and equipment.

[0014] As a further aspect of the present invention, the step of emitting a laser beam through a laser measuring instrument deployed at the borehole of the rotary drilling pile to continuously scan the surface of the rotating and descending drill rod, thereby acquiring the original pile displacement data sequence, specifically includes: Two laser displacement sensors are installed on the drill frame crossbeam of the rotary drilling rig. The laser emitters of the two laser displacement sensors are arranged at a 90-degree angle and are vertically aligned with the outer surface of the drill rod below. Calibrate the measurement zero points of the two laser displacement sensors so that their intersection point coincides with the projection point of the designed central axis of the pile hole on the horizontal plane. Activate the laser displacement sensors so that the laser beams emitted by them continuously irradiate the same height annular band on the side of the drill rod during the uniform rotation and lowering of the drill rod. Two laser displacement sensors simultaneously measure the instantaneous distance from the center of their respective laser spots to the surface of the drill rod at a sampling frequency of no less than 100 Hz, thus obtaining two raw distance data streams in mutually perpendicular directions; The original distance data streams in both directions are timestamped and aligned. The two distance measurements at each sampling time are geometrically calculated with the installation position coordinates of the laser displacement sensor itself to obtain the two-dimensional relative displacement coordinates of the measured point on the side of the drill rod in the horizontal plane at the sampling time. Arranging the two-dimensional relative displacement coordinates calculated at all sampling times in chronological order constitutes the original pile displacement data sequence.

[0015] As a further aspect of the present invention, the step of calculating the time delay from issuing the command to the actual action of the correction mechanism based on the current drill rod lowering depth and lowering speed, and adding a future time stamp to the basic correction action parameters based on the calculated time delay, specifically includes: It receives real-time feedback from the drilling rig control system on the current drill rod lowering depth and lowering speed. Query the preset equipment response parameter table to obtain the inherent mechanical delay time of the correction actuator from receiving the command to starting the action, as well as the stress wave propagation time required for the correction action to be transmitted to the target depth on the drill pipe; The inherent mechanical delay time is added to the stress wave propagation time to obtain a fixed delay component; The dynamic delay component is calculated by multiplying the reciprocal of the lowering speed value by an adjustment coefficient related to the change in lowering depth. The fixed delay component is added to the dynamic delay component to obtain the total time delay from the issuance of the command to the actual action of the correction mechanism. Obtain the current system timestamp, add the total time delay to the current system timestamp, and calculate the future time when the correction action is expected to take effect; The calculated future moment is used as a time tag and bound to the extension and retraction direction and theoretical stroke of the correction cylinder in the basic correction action parameters to generate an initial pre-correction command with a specific future execution time.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By constructing a dynamic noise benchmark library and inputting the original pile displacement data sequence into it, real-time matching of background noise patterns in the construction scenario can be achieved. Based on the matching results, an adaptive subtraction operation is performed on the original data to remove the coupled dynamic background noise and random interference signals in the original data. The pure pile displacement data stream can truly reflect the actual displacement state of the pile. The real-time projection trajectory calculation of the pile center on the horizontal plane is completed based on the pure data stream. The trajectory reconstruction result is highly consistent with the actual movement state of the pile, eliminating the trajectory offset and data distortion problems caused by noise.

[0017] Independent low-frequency drift and high-frequency vibration components are separated from the real-time projected trajectory of the pile center. After the low-frequency drift component is input into the pile attitude evolution model, the subsequent motion state of the pile can be deduced by combining the current stratum resistance parameters and historical attitude data, and the corresponding offset direction and offset velocity are output to form a pre-correction command sequence with time tags. The high-frequency vibration component can distinguish the inherent vibration characteristics of construction machinery from the abnormal vibration characteristics caused by borehole wall collapse through pattern recognition. When abnormal vibration is identified, an emergency state flag is triggered synchronously. The pre-correction command sequence and the emergency state flag are combined to directly generate comprehensive control commands adapted to the pile correction actuator, realizing the synchronous execution of attitude pre-correction and borehole wall abnormality emergency judgment. Attached Figure Description

[0018] Figure 1 This is a flowchart of the intelligent detection method for verticality of rotary drilling pile foundation based on laser measurement as described in this invention; Figure 2 A flowchart for real-time projection trajectory calculation and component separation; Figure 3 To correct the deviation trend of rotary drilling pile foundation; Figure 4 The transition curve for the working state of the instruction fusion scheduling center; Figure 5 The curves show the comparison between the original displacement of the pile and the pure low-frequency displacement. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides an intelligent detection method for the verticality of rotary drilling pile foundations based on laser measurement. The specific method includes: A laser measuring instrument is deployed at the borehole of the rotary drilling pile to continuously scan the surface of the rotating and descending drill rod, collecting the original pile displacement data sequence. A dynamic noise benchmark library is constructed, and the original data sequence is input into it for background noise pattern matching and adaptive subtraction to obtain a clean pile displacement data stream. Based on this clean data stream, the real-time projection trajectory of the pile center on the horizontal plane is calculated, and low-frequency drift components representing the trend deviation of the pile and high-frequency vibration components representing instantaneous disturbances are separated from it. The low-frequency drift component is input into a preset pile attitude evolution model, which combines current stratum resistance parameters and historical attitude data to predict the future deviation trend of the pile, output the predicted offset direction and offset velocity, and generate a set of pre-correction command sequences containing time tags. At the same time, pattern recognition is performed on the high-frequency vibration component to determine whether it belongs to normal vibration of construction machinery or abnormal vibration caused by borehole wall collapse. If it is identified as abnormal, an emergency state flag is triggered. The command fusion and dispatch center combines the pre-correction command sequence with emergency status signs to generate and issue comprehensive control commands to the pile body correction execution mechanism in real time, thereby realizing proactive control of verticality and emergency response to risks.

[0022] In one embodiment of the invention, two laser displacement sensors are installed on the crossbeam of the rotary drilling rig. The laser emitters of the two sensors are arranged at a 90-degree angle and vertically aligned with the outer surface of the drill rod below. The measurement zero points of the two laser displacement sensors are calibrated so that their intersection point coincides with the projection point of the designed central axis of the pile hole on the horizontal plane. The laser displacement sensors are then activated, and the laser beams emitted by them continuously illuminate the same height annular band on the side of the drill rod during the uniform rotation and lowering of the drill rod. The two laser displacement sensors synchronously measure the instantaneous distance values ​​from the center of their respective laser spots to the surface of the drill rod at a sampling frequency of not less than 100 Hz, obtaining two raw distance data streams in mutually perpendicular directions. The raw distance data streams in the two directions are timestamped and aligned, and the two distance measurements at each sampling moment are geometrically calculated with the installation position coordinates of the laser displacement sensors themselves to obtain the two-dimensional relative displacement coordinates of the measured point on the side of the drill rod in the horizontal plane at the sampling moment. The two-dimensional relative displacement coordinates calculated at all sampling moments are arranged in chronological order to form the original pile displacement data sequence.

[0023] The dynamic noise benchmark library was established by analyzing background noise data from laser measuring instruments acquired under conditions free from construction interference. Before the rotary drilling pile construction begins, the laser measuring instrument is activated and static measurements are performed at a fixed location, continuously collecting background noise signals for a period of time. Time-domain and frequency-domain analyses are performed on the collected background noise signals to extract various typical background noise waveform patterns, including low-frequency fluctuation patterns caused by environmental airflow disturbances and high-frequency random patterns caused by the instrument's own electronic thermal noise. These extracted typical background noise waveform patterns are categorized and stored according to their waveform characteristics, frequency range, and probability of occurrence, forming an initial background noise pattern library. During construction, the real-time acquired raw pile displacement data sequence is used, and simultaneously, based on current environmental sensor readings, a set of background noise patterns that best matches the current environmental conditions is dynamically retrieved from the background noise pattern library and used as the reference noise for that moment. The original pile displacement data sequence is compared point-by-point with the retrieved reference noise at the current moment. By calculating the cross-correlation coefficient, the phase difference and amplitude ratio between the original pile displacement data sequence and the retrieved reference noise at the current moment are determined. Based on the determined phase difference and amplitude ratio, the matched reference noise is synchronously subtracted from the original pile displacement data sequence to complete the initial subtraction of background noise. Local singularity detection is then performed on the data sequence after the initial subtraction of background noise to identify and remove abnormal data points caused by transient strong interference. Next, adjacent normal data points are used to repair the data at the removed locations using a cubic spline interpolation algorithm, ultimately outputting a continuous and smooth pure pile displacement data stream.

[0024] In practice, two laser displacement sensors are installed on the crossbeam of the rotary drilling rig. The laser emitters of the two sensors are arranged at a 90-degree angle and vertically aligned with the outer surface of the drill rod below. The measurement zero points of the two laser displacement sensors are calibrated so that the intersection point of the two sensors coincides with the projection point of the designed central axis of the pile hole on the horizontal plane. The laser displacement sensors are then activated, so that the laser beams emitted by the sensors continuously illuminate the same annular band on the side of the drill rod at the same height during the uniform rotation and lowering of the drill rod. In practice, the two laser displacement sensors simultaneously measure the instantaneous distance from their respective spot centers to the drill rod surface at a sampling frequency of not less than 100 Hz, obtaining two raw distance data streams in mutually perpendicular directions. In some embodiments, the raw distance data streams in the two directions are timestamped and aligned, and the two distance measurements at each sampling moment are geometrically calculated with the installation position coordinates of the laser displacement sensors themselves to obtain the two-dimensional relative displacement coordinates of the measured point on the side of the drill rod in the horizontal plane at the sampling moment. It is understandable that arranging the two-dimensional relative displacement coordinates calculated at all sampling times in chronological order constitutes the original pile displacement data sequence.

[0025] The dynamic noise reference library is established by analyzing background noise data from laser measuring instruments acquired under conditions free from construction interference. In specific implementations, before the rotary drilling pile construction begins, the laser measuring instrument is activated and static measurements are performed at a fixed location, continuously collecting background noise signals for a period of time. In specific implementations, the collected background noise signals are analyzed in both the time and frequency domains to extract various typical background noise waveform patterns. These patterns include low-frequency fluctuation patterns caused by environmental airflow disturbances and high-frequency random patterns caused by the instrument's own electronic thermal noise. Optionally, the extracted typical background noise waveform patterns are classified and stored according to waveform characteristics, frequency range, and probability of occurrence, forming an initial background noise pattern library. In some embodiments, during construction, the original pile displacement data sequence is acquired in real time. Simultaneously, from the background noise pattern library, based on current environmental sensor readings, a set of background noise patterns that best matches the current environmental conditions is dynamically retrieved as the reference noise for the current moment. It can be understood that the original pile displacement data sequence is compared point-by-point with the retrieved reference noise for the current moment, and the phase difference and amplitude ratio between the original pile displacement data sequence and the retrieved reference noise for the current moment are determined by calculating the cross-correlation coefficient. In the specific implementation, based on the determined phase difference and amplitude ratio, the matched reference noise is synchronously subtracted from the original pile displacement data sequence to complete the initial subtraction of background noise. In the specific implementation, local singularity detection is performed on the data sequence after the initial background noise subtraction to identify and remove abnormal data points caused by transient strong interference. In the specific implementation, adjacent normal data points are used to repair the data at the removed locations using a cubic spline interpolation algorithm, ultimately outputting a continuous and smooth pure pile displacement data stream. The cross-correlation coefficient is calculated using the following formula during the initial background noise subtraction process:

[0026] in: Indicates a delay of Cross-correlation coefficients at time This represents the first [number]th [unit] in the original pile displacement data sequence. Data points, This represents the mean of the original pile displacement data sequence. Indicates the current time in the reference noise sequence, the first... Data points, This represents the mean of the reference noise sequence at the current time. This indicates the length of the data window used for matching calculations.

[0027] In one embodiment of the present invention, see [reference] Figure 2From the pure pile displacement data stream, a two-dimensional coordinate sequence of the pile center is extracted according to a fixed time window. The two-dimensional coordinate sequence consists of displacement values ​​measured by a laser measuring instrument in two mutually perpendicular directions. In some embodiments, the two-dimensional coordinate sequence is transformed to unify the coordinate points in the two-dimensional coordinate sequence to a global coordinate system with the pile borehole design center as the origin, resulting in a discrete set of position points of the pile center on the horizontal plane. In a specific implementation, the discrete position point set is sorted and connected based on timestamps to form a real-time projection trajectory characterizing the change of the pile center position over time. In a specific implementation, a moving average filter with an adaptive window length is used to process the real-time projection trajectory. The adaptive window length is dynamically adjusted according to the current drill rod lowering speed; the faster the lowering speed, the shorter the window length. In a specific implementation, the smooth trajectory obtained after moving average filtering is the low-frequency drift component. In some embodiments, the corresponding low-frequency drift component coordinate values ​​are subtracted point by point from the real-time projection trajectory to obtain a residual sequence, which is the high-frequency vibration component. In a specific implementation, the standard deviation of the high-frequency vibration component is calculated to obtain the vibration intensity index within the current time window. The formula for calculating the adaptive window length is as follows:

[0028] in: This indicates the window length used by the moving average filter when calculating the low-frequency drift component at the current moment. Indicates the preset base window length. Indicates the speed influence coefficient. This indicates the current speed at which the drill pipe is lowered. When When the window length is increased, Reduce to adapt to rapidly changing operating conditions.

[0029] The pile attitude evolution model is a recurrent neural network model based on a long short-term memory network. The input layer of the model receives a sequence of low-frequency drift components arranged in chronological order, and simultaneously receives the current formation resistance parameters reported in real time by the drilling rig sensors. In specific implementations, the current formation resistance parameters include drilling pressure and torque. The model includes a state memory unit to store and update the historical characteristics of the pile's attitude changes over the previous few time steps. In specific implementations, the model calculates the pile attitude evolution based on the input low-frequency drift component sequence, the current formation resistance parameters, and the historical characteristics stored in the state memory unit, using a multi-layer neural network. In some embodiments, the predicted coordinates of the pile center position in the next few time steps are compared with the current pile center position coordinates to calculate the predicted displacement increments of the pile along the X and Y axes on the horizontal plane. In specific implementations, the direction angle of the composite vector is calculated based on the predicted displacement increments; this direction angle is the predicted pile offset direction.

[0030] In one embodiment of the present invention, a pre-correction command mapping table is established. This table defines basic correction action parameters corresponding to different offset direction and offset speed ranges. These parameters include the extension / retraction direction and theoretical stroke of the correction cylinder. The pre-correction command mapping table is queried, and the corresponding basic correction action parameters are determined based on the calculated azimuth angle range of the predicted pile offset direction and the speed threshold range of the predicted pile offset speed. Based on the current drill rod lowering depth and speed, the time delay from issuing the command to the actual action of the correction mechanism is calculated. Based on this calculated time delay, a future time stamp is added to the basic correction action parameters, forming an initial pre-correction command with a future execution time. A feedback correction factor is introduced. This factor is dynamically calculated based on the deviation between the actual correction effect and the expected effect of recently executed correction commands. The feedback correction factor is multiplied by the theoretical stroke in the initial pre-correction command to correct the theoretical stroke, resulting in the corrected theoretical stroke. The corrected theoretical stroke, the extension and retraction direction of the correction cylinder, and the time stamp of the future moment are combined to generate a complete pre-correction instruction. Multiple such instructions are arranged in the order of the time stamps to form a pre-correction instruction sequence.

[0031] The system receives real-time feedback from the drilling rig control system regarding the current drill pipe lowering depth and speed. It queries a preset equipment response parameter table to obtain the inherent mechanical delay time of the correction actuator from receiving the command to initiating action, as well as the stress wave propagation time required for the correction action to reach the target depth on the drill pipe. The inherent mechanical delay time and stress wave propagation time are added together to obtain a fixed delay component. The reciprocal of the lowering speed is multiplied by an adjustment coefficient related to changes in lowering depth to calculate the dynamic delay component. The fixed and dynamic delay components are added together to obtain the total time delay from issuing the command to the actual action of the correction mechanism. The system obtains the current timestamp and adds the total time delay to the current timestamp to calculate the expected future time when the correction action is expected to take effect. This calculated future time is used as a timestamp and bound to the correction cylinder extension / retraction direction and theoretical stroke in the basic correction action parameters to generate an initial pre-correction command with a specific future execution time.

[0032] In practical implementation, a pre-correction command mapping table is established. This table defines the basic correction action parameters corresponding to different offset direction and offset speed ranges. The basic correction action parameters include the extension and retraction direction and theoretical stroke of the correction cylinder. See Table 1.

[0033] Table 1: Pre-correction instruction mapping table

[0034] In some embodiments, a pre-correction command mapping table is queried, and the corresponding basic correction action parameters are determined based on the calculated azimuth interval of the predicted pile offset direction and the velocity threshold interval of the predicted pile offset velocity. In a specific implementation, the time delay from issuing the command to the actual action of the correction mechanism is calculated based on the current drill rod lowering depth and speed. Based on the calculated time delay, a future time stamp is added to the basic correction action parameters to form an initial pre-correction command with a future execution time. A feedback correction factor is introduced, which is dynamically calculated based on the deviation between the actual correction effect and the expected effect of recently executed correction commands. This can be understood as multiplying the feedback correction factor by the theoretical stroke in the initial pre-correction command to correct the theoretical stroke, resulting in a corrected theoretical stroke. In a specific implementation, the corrected theoretical stroke, the extension / retraction direction of the correction cylinder, and the future time stamp are combined to generate a complete pre-correction command. In a specific implementation, multiple such commands are arranged in chronological order according to their time stamps to form a pre-correction command sequence. The formula for calculating the corrected theoretical stroke is:

[0035] in: This indicates the revised theoretical travel distance. This represents the initial theoretical travel obtained from the pre-correction instruction mapping table. This represents the feedback correction factor. The value depends on the trend of the ratio of the actual displacement correction to the theoretical displacement correction in the most recent correction actions. If the actual correction remains consistently small, then... If the actual correction amount remains excessively large, then .

[0036] The system receives real-time feedback from the drilling rig control system regarding the current drill rod lowering depth and speed. It queries a preset equipment response parameter table to obtain the inherent mechanical delay time of the correction actuator from receiving the command to initiating action, as well as the stress wave propagation time required for the correction action to reach the target depth on the drill rod. In specific implementations, the inherent mechanical delay time and stress wave propagation time are added to obtain a fixed delay component. In another specific implementation, the reciprocal of the lowering speed is multiplied by an adjustment coefficient related to changes in lowering depth to calculate the dynamic delay component. In some embodiments, the fixed delay component and the dynamic delay component are added to obtain the total time delay from issuing the command to the actual action of the correction mechanism. The system obtains its current timestamp; this can be understood as adding the total time delay to the current timestamp to calculate the expected future time when the correction action is expected to take effect. In a specific implementation, this calculated future time is used as a timestamp and bound to the correction cylinder extension / retraction direction and theoretical stroke in the basic correction action parameters to generate an initial pre-correction command with a specific future execution time.

[0037] See Figure 3In the dynamic feedback control of the correction stroke trend, the comparison between the theoretical stroke and the corrected stroke intuitively reflects the real-time correction effect of the dynamic feedback factor. The dashed line in the figure represents the theoretical stroke obtained from the pre-correction command mapping table, which remains constant at 12.0 mm; the solid line represents the corrected stroke after introducing the dynamic feedback factor, and its fluctuation characteristics reflect the adaptive adjustment to the actual correction effect. From the data characteristics, the corrected stroke exhibits high-frequency oscillations around the theoretical stroke, with an amplitude range of approximately 10.2 mm to 12.6 mm. This fluctuation stems from the dynamic scaling of the feedback correction factor on the initial theoretical stroke: when the recent actual correction amount is consistently less than expected, the feedback factor is greater than 1, making the corrected stroke higher than the theoretical value to enhance the correction force; when the actual correction amount is consistently greater than expected, the feedback factor is less than 1, making the corrected stroke lower than the theoretical value to avoid over-correction. This dynamic adjustment mechanism effectively compensates for the influence of uncertainties such as changes in ground resistance and equipment mechanical delays on the correction effect, making the actual pile posture closer to the preset control target. From an engineering perspective, this curve verifies the core role of the dynamic feedback factor in the closed-loop correction control: by evaluating the effect error of the executed correction command in real time and dynamically adjusting the stroke of the correction cylinder, the correction control is upgraded from "open-loop lookup table type" to "closed-loop adaptive type", which improves the robustness and accuracy of the verticality control of rotary drilling pile foundation.

[0038] In one embodiment of the present invention, two types of vibration samples are extracted and labeled from historical construction data. The first type consists of normal vibration signal samples generated by the drill bit cutting the soil layer and the operation of the drill rig's power head. The second type consists of abnormal vibration signal samples generated by the collapse of the borehole wall soil layer impacting the drill rod. Feature extraction is performed on the two types of vibration signal samples. The extracted features include the energy ratio of the vibration signal in a specific frequency band, the zero-crossing rate, the steepness of the waveform envelope, and the probability distribution skewness of the signal amplitude. A support vector machine classifier is trained using the extracted features. High-frequency vibration components within the current time window are collected in real time. The same feature extraction operation as for the training samples is performed on the real-time collected high-frequency vibration components to obtain a set of real-time vibration feature vectors. The real-time vibration feature vectors are input into the trained support vector machine classifier. The support vector machine classifier outputs a classification result label and a corresponding confidence score. If the classification result label is abnormal vibration and the confidence score exceeds a preset confidence threshold, the current vibration is determined to be an abnormal vibration mode, and an emergency state flag is triggered. If an emergency state flag is triggered, the vibration waveform characteristics and corresponding pile depth information at the trigger time are recorded simultaneously and stored as new abnormal samples in the historical database for subsequent incremental learning of the support vector machine classifier.

[0039] A command fusion and scheduling center is established, continuously receiving pre-correction command sequences from the pre-correction command generation module and emergency status flags from the vibration pattern recognition module. Internally, the command fusion and scheduling center maintains a command execution queue, storing pre-correction commands to be executed in chronological order according to their timestamps. When a new pre-correction command sequence is received, the commands within it are merged and inserted into the corresponding positions in the command execution queue according to their timestamps. The command fusion and scheduling center monitors the status of the emergency status flags in real time. When the emergency status flag is triggered and becomes active, the command fusion and scheduling center immediately clears all pre-correction commands in the current command execution queue. After clearing the command execution queue, the command fusion and scheduling center generates a highest-priority emergency correction command based on the energy magnitude of the high-frequency vibration component that triggered the emergency status flag. This emergency correction command includes action parameters to rapidly retract the correction cylinder to a safe position and is marked for immediate execution. When the emergency status flag is inactive, the command fusion and scheduling center retrieves the pre-correction command that arrives at its execution time earliest from the head of the command execution queue, using it as the final integrated control command. The generated integrated control commands are sent to the controller of the pile body correction actuator in real time via fieldbus. The controller parses the commands and drives the correction cylinder to perform the corresponding extension and retraction actions.

[0040] In specific implementations, two types of vibration samples are extracted and labeled from historical construction data. The first type consists of normal vibration signals generated by the drill bit cutting through the soil and the operation of the drill rig's power head. The second type consists of abnormal vibration signals generated by the collapse of the borehole wall soil impacting the drill rod. In some embodiments, feature extraction is performed on the two types of vibration signal samples. The extracted features include the energy proportion of the vibration signal within a specific frequency band, the zero-crossing rate, the steepness of the waveform envelope, and the skewness of the probability distribution of the signal amplitude. In specific implementations, a support vector machine classifier is trained using the extracted features. In specific implementations, high-frequency vibration components within the current time window are collected in real time. The same feature extraction operation as for the training samples is performed on the real-time collected high-frequency vibration components to obtain a set of real-time vibration feature vectors. In one specific implementation, the extracted real-time vibration feature vectors can be represented as a set of numerical examples in Table 2. (Refer to Table 2.)

[0041] Table 2: Examples of Real-Time Vibration Feature Vectors

[0042] In specific implementations, the real-time vibration feature vector is input into a pre-trained support vector machine (SVM) classifier. The SVM classifier outputs a classification result label and a corresponding confidence score. In specific implementations, if the classification result label is "abnormal vibration" and the confidence score exceeds a preset confidence threshold, the current vibration is determined to be an abnormal vibration mode, and an emergency state flag is triggered. In some embodiments, if an emergency state flag is triggered, the vibration waveform features and corresponding pile depth information at the triggering time are simultaneously recorded and stored as new abnormal samples in the historical database for subsequent incremental learning of the SVM classifier. The decision function of the SVM classifier can be expressed as:

[0043] in: This represents the output symbol of the decision function, used to indicate the classification label. It is a symbolic function. This represents the total number of support vectors. Indicates the first The Lagrange multipliers corresponding to each support vector. Indicates the first The training labels are known for the support vectors. Indicates the first The feature vectors of the support vectors. This represents the input real-time vibration feature vector. Represents the kernel function. Indicates the bias term. When An emergency state flag is triggered when the value is -1 and the corresponding classification confidence score exceeds the threshold.

[0044] A command fusion and scheduling center is established, which continuously receives pre-correction command sequences from the pre-correction command generation module and emergency status flags from the vibration pattern recognition module. In specific implementations, the command fusion and scheduling center maintains an internal command execution queue, which stores pre-correction commands to be executed according to their timestamps. In some embodiments, when a new pre-correction command sequence is received, the commands within it are merged according to their timestamps and inserted into the corresponding positions in the command execution queue. The command fusion and scheduling center monitors the status of the emergency status flags in real time. When the emergency status flag is triggered and becomes active, the command fusion and scheduling center immediately clears all pre-correction commands in the current command execution queue. After clearing the command execution queue, the command fusion and scheduling center generates a highest-priority emergency correction command based on the energy magnitude of the high-frequency vibration component that triggered the emergency status flag. In specific implementations, the emergency correction command includes action parameters to rapidly retract the correction cylinder to a safe position, and the emergency correction command is marked for immediate execution. It is understandable that when the emergency status flag is invalid, the instruction fusion scheduling center will take the pre-correction instruction that arrives at the execution time first from the head of the instruction execution queue in chronological order and use it as the final integrated control instruction.

[0045] See Figure 4 In the workflow of the command fusion scheduling center, the evolution of the system's working state strictly follows the logic of priority scheduling and abnormal response. Specifically, the system initially receives pre-correction instructions and monitors emergency flags, maintaining a working state of 1 (running state). This indicates that the instruction execution queue is normally receiving and buffering the pre-correction instruction sequence, while continuously monitoring the emergency flag output by the vibration pattern recognition module. When the emergency flag is invalid, the system enters the normal execution phase, and the working state drops to 0.8. At this point, the scheduling center extracts the pre-correction instructions from the instruction queue according to the time tag sequence and issues them as integrated control instructions to the correction execution mechanism, demonstrating the stable scheduling characteristics under normal operating conditions. If the emergency flag is triggered and becomes valid, the system immediately switches to the emergency clearing phase, and the working state drops sharply to 0. The command fusion scheduling center performs a queue clearing operation, terminating all pending pre-correction instructions to respond to abnormal vibration risks such as borehole wall collapse. After the system is cleared, it enters the emergency command issuance phase, and the working status returns to 1. The dispatch center generates and issues the highest priority emergency correction command, driving the correction cylinder to retract to a safe position, completing the emergency response closed loop under abnormal conditions. The state curve fully depicts the entire process of state transition from routine pre-correction scheduling to emergency anomaly response, intuitively reflecting the state switching mechanism and priority design logic of the command fusion dispatch center under both stable production and safety emergency modes.

[0046] In one embodiment of the present invention, after the integrated control command is issued to the pile body correction actuator, the effect of the correction action is evaluated online and the model parameters are self-calibrated. After the correction actuator starts its action, new pure pile body displacement data streams fed back by the laser measuring instrument are continuously collected over a period of time. The displacement response curve corresponding to the correction action is extracted from the new pure pile body displacement data stream. The actual characteristic parameters of the displacement response curve are calculated, including the response delay time, the adjustment time to reach steady state, and the overshoot. The actual characteristic parameters are compared with the expected ideal response characteristic parameters when the integrated control command is issued, and the error between the actual characteristic parameters and the expected ideal response characteristic parameters when the integrated control command is issued is calculated. If the error exceeds the allowable range, the actual characteristic parameters, the content of the issued integrated control command, and the prediction result of the pile body attitude evolution model at that time are used together as a training sample and stored in an effect evaluation sample library. The recently accumulated sample data in the effect evaluation sample library is periodically called to fine-tune the neural network connection weights in the pile body attitude evolution model to reduce the error in the predicted pile body offset direction and offset speed in the future. Meanwhile, based on the recent error trends of multiple correction actions, the correspondence between the offset speed range and the basic correction action parameters in the pre-correction command mapping table is dynamically adjusted to make the generated pre-correction command sequence more consistent with the dynamic characteristics of the current strata and equipment.

[0047] In practical implementation, after the integrated control command is issued to the pile correction actuator, the effect of the correction action is evaluated online and the model parameters are self-calibrated. After the correction actuator begins its action, new, clean pile displacement data streams fed back by the laser measuring instrument are continuously collected over a period of time. In practical implementation, the displacement response curve corresponding to the correction action is extracted from the new clean pile displacement data stream. In some embodiments, the actual characteristic parameters of the displacement response curve are calculated, including the response delay time, the settling time to reach steady state, and the overshoot. In practical implementation, the actual characteristic parameters are compared with the ideal response characteristic parameters expected when the integrated control command is issued, and the error between the actual characteristic parameters and the ideal response characteristic parameters expected when the integrated control command is issued is calculated. The formula for calculating the error is as follows:

[0048] in: This represents the overall response error. This represents the actual measured response delay time. This indicates the expected ideal response time. This indicates the actual measured settling time. This indicates the expected ideal settling time. This represents the actual measured overshoot. This indicates the expected ideal overshoot. , , Let represent the weighting coefficients of response delay time, settling time, and overshoot in the comprehensive error calculation, respectively, and satisfy . .

[0049] In practical implementation, if the error exceeds the allowable range, the actual characteristic parameters, the content of the issued integrated control commands, and the prediction results of the pile attitude evolution model at that time are all used as a training sample and stored in an effect evaluation sample library. In practice, recently accumulated sample data from the effect evaluation sample library is periodically called to fine-tune the neural network connection weights in the pile attitude evolution model to reduce the error in the predicted pile offset direction and offset velocity. It can be understood that the fine-tuning training of the neural network connection weights can employ a backpropagation algorithm based on gradient descent. In some embodiments, the correspondence between the offset velocity range and the basic correction action parameters in the pre-correction command mapping table is dynamically adjusted based on the effect error trend of recent correction actions, making the generated pre-correction command sequence more consistent with the dynamic characteristics of the current strata and equipment. The strategy for adjusting the correspondence can be achieved by statistically analyzing the deviation between the actual correction effect of recent commands and the theoretical values ​​in the pre-correction command mapping table.

[0050] See Figure 5 In the noise adaptive subtraction process of pile displacement data, the separation of the original signal from the clean low-frequency component relies on a dynamic noise reference library and smoothing filtering technology. Specifically, the original X-displacement sequence represents a high-frequency fluctuating signal containing environmental disturbances and instrument noise, characterized by random spikes and instantaneous peaks superimposed on a smooth trend. After background noise pattern matching, singular value removal, and cubic spline interpolation repair, a clean X-displacement (low-frequency) curve is obtained. This curve fully preserves the macroscopic drift trend of pile attitude changes while filtering out high-frequency vibrations and instantaneous interference. The difference between the two displacement sequences is quantified through time-domain residual analysis and frequency-domain energy proportion: the original signal is bandpass filtered to extract high-frequency noise components above 10Hz, and their cross-correlation coefficient with the reference noise mode is calculated. This coefficient serves as an evaluation index for the noise subtraction effect. During parameter configuration, the window length of the moving average filter is adaptively adjusted according to the drill rod lowering speed (range: 5-20 sampling points), and the singular value detection threshold is three times the local standard deviation of the clean displacement sequence.

[0051] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A laser-based intelligent detection method for the verticality of rotary drilling pile foundations, characterized in that, Include: A laser beam is emitted by a laser measuring instrument installed at the borehole of the rotary drilling pile to continuously scan the surface of the rotating and descending drill rod, thereby collecting the original pile displacement data sequence. A dynamic noise benchmark library is constructed, and the collected original pile displacement data sequence is input into the dynamic noise benchmark library to perform background noise pattern matching and adaptive subtraction to obtain a clean pile displacement data stream. Based on the pure pile displacement data stream, the real-time projection trajectory of the pile center on the horizontal plane is calculated, and the low-frequency drift component and high-frequency vibration component of the pile are separated from the real-time projection trajectory. The low-frequency drift component is input into a preset pile attitude evolution model. The pile attitude evolution model predicts the future deviation trend of the pile based on the current stratum resistance parameters and historical attitude data, and outputs the predicted pile offset direction and offset speed. Based on the predicted pile offset direction and offset speed, a set of pre-correction instruction sequences containing time tags is generated; Simultaneously, pattern recognition is performed on the high-frequency vibration components to determine whether they belong to normal vibration of construction machinery or abnormal vibration caused by hole wall collapse. If an abnormal vibration pattern is identified, an emergency state flag is triggered. By combining the pre-correction instruction sequence with the emergency status flag, the final integrated control instruction is generated and issued to the pile correction execution mechanism in real time.

2. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 1, characterized in that, A dynamic noise benchmark library is constructed, and the collected raw pile displacement data sequence is input into the dynamic noise benchmark library to perform background noise pattern matching and adaptive subtraction to obtain a clean pile displacement data stream. The specific implementation includes: The dynamic noise benchmark library was established by analyzing background noise data of laser measuring instruments acquired in an environment free from construction interference; Before the rotary drilling pile construction begins, the laser measuring instrument is activated and static measurements are performed at a fixed position. Background noise signals are continuously collected for a period of time. The collected background noise signals are analyzed in the time domain and frequency domain to extract a variety of typical background noise waveform patterns, including low-frequency fluctuation patterns caused by environmental airflow disturbances and high-frequency random patterns caused by the instrument's own electronic thermal noise. The extracted typical background noise waveform patterns are classified and stored according to their waveform characteristics, frequency range and occurrence probability to form an initial background noise pattern library. During construction, the original pile displacement data sequence is acquired in real time. At the same time, the background noise pattern library is dynamically retrieved from the background noise pattern library based on the current environmental sensor readings to find the set of background noise patterns that best match the current environmental conditions, which serves as the reference noise for the current moment. The original pile displacement data sequence is compared point by point with the retrieved reference noise at the current moment. By calculating the cross-correlation coefficient, the phase difference and amplitude ratio between the original pile displacement data sequence and the retrieved reference noise at the current moment are determined. Based on the determined phase difference and amplitude ratio, the matching reference noise is synchronously subtracted from the original pile displacement data sequence to complete the initial subtraction of background noise. After initially removing background noise, the data sequence is then subjected to local singularity detection to identify and remove abnormal data points caused by transient strong interference. Then, the data at the removed positions is repaired by using adjacent normal data points through a cubic spline interpolation algorithm, and finally, a continuous and smooth pure pile displacement data stream is output.

3. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 2, characterized in that, Based on the pure pile displacement data stream, the real-time projection trajectory of the pile center on the horizontal plane is calculated. The low-frequency drift component and high-frequency vibration component of the pile are separated from the real-time projection trajectory, specifically including: From the pure pile displacement data stream, a two-dimensional coordinate sequence of the pile center is extracted according to a fixed time window. The two-dimensional coordinate sequence is composed of displacement values ​​measured by a laser measuring instrument in two mutually perpendicular directions. The two-dimensional coordinate sequence is transformed and unified to a global coordinate system with the design center of the pile hole as the origin, resulting in a discrete set of position points of the pile center point on the horizontal plane. The discrete location points are sorted and connected based on timestamps to form a real-time projection trajectory that characterizes the change of the center position of the pile over time. The real-time projected trajectory is processed using a moving average filter with an adaptive window length. The adaptive window length is dynamically adjusted according to the current drill rod lowering speed. The faster the lowering speed, the shorter the window length. The smooth trajectory obtained after moving average filtering is the low-frequency drift component. The corresponding low-frequency drift component coordinate values ​​are subtracted point by point from the real-time projection trajectory to obtain the residual sequence, which is the high-frequency vibration component. The standard deviation of the high-frequency vibration component is calculated to obtain the vibration intensity index within the current time window.

4. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 3, characterized in that, The low-frequency drift component is input into a preset pile attitude evolution model. This model predicts the future deviation trend of the pile based on current ground resistance parameters and historical attitude data, and outputs the predicted pile offset direction and velocity. Specifically, it includes: The pile attitude evolution model is a recurrent neural network model based on a long short-term memory network. Its input layer receives the low-frequency drift component sequence arranged in time order, and at the same time receives the current formation resistance parameters reported in real time by the drilling rig sensors. The current formation resistance parameters include drilling pressure and torque. The pile attitude evolution model contains a state memory unit, which is used to store and update the historical characteristics of the pile attitude changes in the previous few time steps. The pile attitude evolution model calculates based on the input low-frequency drift component sequence and the current formation resistance parameters, combined with the historical characteristics stored in the state memory unit, through its internal multi-layer neural network. The output of the multilayer neural network is the predicted coordinates of the pile center position in the next few time steps. The predicted coordinates of the pile center position in the next few time steps are compared with the current pile center position coordinates to calculate the predicted displacement increment of the pile on the horizontal plane along the X and Y axes. Based on the predicted displacement increment, the direction angle of the composite vector is calculated, which is the predicted pile offset direction. At the same time, the ratio of the magnitude of the composite vector to the predicted time span is calculated, which is the predicted pile offset velocity.

5. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 4, characterized in that, Based on the predicted pile offset direction and offset velocity, a set of pre-correction instruction sequences containing time tags is generated, specifically including: A pre-correction command mapping table is established, which defines the basic correction action parameters corresponding to different offset direction intervals and offset speed intervals. The basic correction action parameters include the extension and retraction direction and theoretical stroke of the correction cylinder. Query the pre-correction instruction mapping table, and determine the corresponding basic correction action parameters based on the azimuth interval of the predicted pile offset direction and the velocity threshold interval of the predicted pile offset velocity. Based on the current drill pipe lowering depth and lowering speed, calculate the time delay from issuing the command to the actual action of the correction mechanism. Based on the calculated time delay, add a future time stamp to the basic correction action parameters to form an initial pre-correction command with a future execution time. A feedback correction factor is introduced, which is dynamically calculated based on the deviation between the actual correction effect and the expected effect of the recently executed correction command. The feedback correction factor is multiplied by the theoretical travel in the initial pre-correction command to correct the theoretical travel and obtain the corrected theoretical travel. The corrected theoretical stroke, the extension and retraction direction of the correction cylinder, and the time stamp of the future moment are combined to generate a complete pre-correction instruction. Multiple such instructions are arranged in the order of the time stamps to form the pre-correction instruction sequence.

6. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 5, characterized in that, Pattern recognition is performed on the high-frequency vibration components to determine whether they belong to normal vibration of construction machinery or abnormal vibration caused by hole wall collapse. Specifically, this includes: Two types of vibration samples were extracted and labeled from historical construction data. The first type is the normal vibration signal sample generated by the drill bit cutting the soil layer and the operation of the drill power head. The second type is the abnormal vibration signal sample generated by the collapse of the soil layer in the borehole wall impacting the drill rod. Feature extraction was performed on the two types of vibration signal samples. The extracted features included the energy ratio of the vibration signal in a specific frequency band, the zero-crossing rate, the steepness of the waveform envelope, and the probability distribution skewness of the signal amplitude. A support vector machine classifier was trained using the extracted features. The high-frequency vibration components within the current time window are collected in real time, and the same feature extraction operation as the training samples is performed on the collected high-frequency vibration components to obtain a set of real-time vibration feature vectors. The real-time vibration feature vector is input into a trained support vector machine classifier. The support vector machine classifier outputs a classification result label and a corresponding confidence score. If the classification result label is abnormal vibration and the confidence score exceeds the preset confidence threshold, the current vibration is determined to be an abnormal vibration mode, and the emergency state flag is triggered. If the emergency state flag is triggered, the vibration waveform characteristics and the corresponding pile depth information at the trigger time are recorded simultaneously and stored in the historical database as new abnormal samples for subsequent incremental learning of the support vector machine classifier.

7. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 6, characterized in that, Combining the pre-correction command sequence with the emergency status flag, the final integrated control command is generated and issued to the pile correction actuator in real time, specifically including: A command fusion scheduling center is established, which continuously receives a pre-correction command sequence from the pre-correction command generation module and an emergency status flag from the vibration pattern recognition module. The instruction fusion scheduling center maintains an instruction execution queue. The instruction execution queue stores pre-correction instructions to be executed in the order of the instruction timestamps. When a new pre-correction instruction sequence is received, the instructions in it are merged and inserted into the corresponding position of the instruction execution queue according to the timestamps. The instruction fusion scheduling center monitors the status of the emergency status flag in real time. When the emergency status flag is triggered to an effective state, the instruction fusion scheduling center immediately clears all pre-correction instructions in the current instruction execution queue. After clearing the instruction execution queue, the instruction fusion scheduling center generates a highest priority emergency correction instruction based on the energy magnitude of the high-frequency vibration component when the emergency state flag is triggered. The emergency correction instruction includes action parameters to quickly retract the correction cylinder to a safe position, and marks the emergency correction instruction as to be executed immediately. When the emergency status flag is invalid, the instruction fusion scheduling center retrieves the pre-correction instruction that arrives at the execution time first from the head of the instruction execution queue in chronological order, and uses it as the final integrated control instruction. The generated integrated control commands are sent to the controller of the pile body correction actuator in real time via fieldbus. The controller parses the commands and drives the correction cylinder to perform the corresponding extension and retraction actions.

8. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 7, characterized in that, After issuing the integrated control command to the pile body correction actuator, the process also includes online evaluation of the correction effect and self-calibration of model parameters. After the correction actuator starts to move, new pure pile displacement data streams fed back by the laser measuring instrument are continuously collected over a period of time. The displacement response curves corresponding to the correction action are extracted from the new pure pile displacement data streams. Calculate the actual characteristic parameters of the displacement response curve, including the response delay time, the settling time to reach steady state, and the overshoot. Compare the actual characteristic parameters with the expected ideal response characteristic parameters when issuing the integrated control command, and calculate the error between the actual characteristic parameters and the expected ideal response characteristic parameters when issuing the integrated control command. If the error exceeds the allowable range, the actual characteristic parameters, the content of the issued comprehensive control command, and the prediction result of the pile body attitude evolution model at that time will be used together as a training sample and stored in an effect evaluation sample library. Periodically call the recently accumulated sample data in the effect evaluation sample library to fine-tune the neural network connection weights in the pile posture evolution model, so as to reduce the error of the predicted pile offset direction and offset speed in the future. Meanwhile, based on the recent error trends of multiple correction actions, the correspondence between the offset speed range and the basic correction action parameters in the pre-correction command mapping table is dynamically adjusted to make the generated pre-correction command sequence more consistent with the dynamic characteristics of the current strata and equipment.

9. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 8, characterized in that, The process involves using a laser measuring instrument positioned at the borehole opening of the rotary drill pile to emit a laser beam, continuously scanning the surface of the descending drill rod to acquire the original pile displacement data sequence, specifically including: Two laser displacement sensors are installed on the drill frame crossbeam of the rotary drilling rig. The laser emitters of the two laser displacement sensors are arranged at a 90-degree angle and are vertically aligned with the outer surface of the drill rod below. Calibrate the measurement zero points of the two laser displacement sensors so that their intersection point coincides with the projection point of the designed central axis of the pile hole on the horizontal plane. Activate the laser displacement sensors so that the laser beams emitted by them continuously irradiate the same height annular band on the side of the drill rod during the uniform rotation and lowering of the drill rod. Two laser displacement sensors simultaneously measure the instantaneous distance from the center of their respective laser spots to the surface of the drill rod at a sampling frequency of no less than 100 Hz, thus obtaining two raw distance data streams in mutually perpendicular directions; The original distance data streams in both directions are timestamped and aligned. The two distance measurements at each sampling time are geometrically calculated with the installation position coordinates of the laser displacement sensor itself to obtain the two-dimensional relative displacement coordinates of the measured point on the side of the drill rod in the horizontal plane at the sampling time. Arranging the two-dimensional relative displacement coordinates calculated at all sampling times in chronological order constitutes the original pile displacement data sequence.

10. The intelligent detection method for verticality of rotary drilling pile foundations based on laser measurement according to claim 9, characterized in that, The process involves calculating the time delay from issuing the command to the actual action of the correction mechanism based on the current drill pipe lowering depth and speed. Based on this calculated time delay, a future time stamp is added to the basic correction action parameters. Specifically, this includes: It receives real-time feedback from the drilling rig control system on the current drill rod lowering depth and lowering speed. Query the preset equipment response parameter table to obtain the inherent mechanical delay time of the correction actuator from receiving the command to starting the action, as well as the stress wave propagation time required for the correction action to be transmitted to the target depth on the drill pipe; The inherent mechanical delay time is added to the stress wave propagation time to obtain a fixed delay component; The dynamic delay component is calculated by multiplying the reciprocal of the lowering speed value by an adjustment coefficient related to the change in lowering depth. The fixed delay component is added to the dynamic delay component to obtain the total time delay from the issuance of the command to the actual action of the correction mechanism. Obtain the current system timestamp, add the total time delay to the current system timestamp, and calculate the future time when the correction action is expected to take effect; The calculated future moment is used as a time tag and bound to the extension and retraction direction and theoretical stroke of the correction cylinder in the basic correction action parameters to generate an initial pre-correction command with a specific future execution time.

Images