Anomaly Identification Method for Impact Drilling Based on the Coordination of Cutting Speed ​​and Mud Parameters

By using a multi-source data collaborative discrimination method, the problems of fault misjudgment and monitoring lag in impact drills in complex strata were solved, realizing automated closed-loop control and improving the efficiency and safety of pile foundation drilling.

CN123064458APending Publication Date: 2026-11-10GUANGDONG POWER TRANSMISSION & TRANSFORMATION ENG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611360731.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-03
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

The existing automatic control system for impact drilling cannot adapt to geological changes, resulting in misjudgments, missed judgments, and monitoring delays. It cannot achieve automated closed-loop handling of fault types, resulting in low efficiency in pile foundation drilling and potential safety hazards.

Method used

By employing synchronous acquisition of multi-source heterogeneous sensing data, adaptive preprocessing, dynamic updating of formation benchmark features, and an improved DS evidence theory collaborative discrimination method, combined with drilling speed and mud parameters, we can achieve accurate identification of fault types and automated closed-loop control.

Benefits of technology

It improves the accuracy of fault identification, reduces missed detections and false alarms, achieves millisecond-level real-time monitoring, improves drilling efficiency and construction safety, and reduces the investment and energy consumption of on-site technical personnel.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123064458A_ABST
    Figure CN123064458A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of intelligent construction monitoring technology for pile foundations, specifically a shock drilling anomaly identification method based on the coordination of drilling speed and mud parameters. The method includes the following steps: S1, synchronous acquisition of multi-source heterogeneous sensing data; S2, adaptive preprocessing of multi-source data; S3, dynamic updating of formation benchmark features; S4, confidence calculation of three-modal feature fusion; S5, quantitative judgment of multi-parameter coupled anomalies; S6, hierarchical early warning and intelligent closed-loop control. This invention eliminates the need for manual fixed-point measurement and monitoring, achieving real-time millisecond-level anomaly identification, significantly reducing the probability of false alarms and missed alarms in basalt hard rock pile foundation construction. It can be integrated with an energy storage-type diesel-storage power supply system for coordinated operation, possessing multiple technical advantages such as reduced manpower, increased efficiency, safe construction, and green energy saving. It is suitable for intelligent construction scenarios of hard rock perforated cast-in-place piles in Southern Power Grid transmission and transformation projects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of intelligent construction monitoring technology for pile foundations, specifically a method for identifying abnormalities in impact drilling based on the synergy of drilling speed and mud parameters. Background Technology

[0002] Drilled pile foundations rely on the weight of the impact drill to break through rock strata and complete the drilling operation. Basalt strata are widely distributed in the Hainan region, characterized by high rock hardness, well-developed fissures, and shallow groundwater levels during the high-water season. This makes the construction process highly susceptible to problems such as borehole deviation, borehole collapse, grout leakage, and slow progress in hard rock. Existing automatic control and operational monitoring solutions for impact drills have the following shortcomings: Single-parameter static threshold discrimination suffers from prominent issues of misjudgment and omission. Traditional automatic impact equipment relies solely on a fixed threshold for advance speed to determine construction anomalies, without considering the borehole wall protection status reflected by mud density and mud level. It cannot distinguish between "normal low-speed drilling in basalt hard rock" and "slow advance caused by mud failure, borehole collapse and instability." On-site monitoring relies on manual plumb bob measurement of borehole position deviation and visual observation of mud level, which has a lag of several minutes. Small cracks and early mud leakage cannot be detected manually.

[0003] The fixed stratum discrimination threshold cannot adapt to geological changes. The existing control system has a fixed parameter threshold built in by the factory. It does not automatically update the benchmark range with drilling depth, rock weathering degree, and alternation of soft and hard interlayers. When the stratum changes, the threshold becomes invalid, resulting in frequent meaningless shutdowns and a significant reduction in the efficiency of pile foundation drilling.

[0004] Fault identification and equipment control are disconnected, and there is no automated closed-loop handling. The existing monitoring system can only provide alarm prompts and cannot automatically link the PLC to adjust the hammer stroke and impact frequency according to the fault type. After a fault occurs, technicians still need to be on-site to handle it manually. The entire process of "abnormal identification - fault judgment - automatic control" has not been automated.

[0005] Therefore, there is an urgent need to develop an impact drill anomaly identification method that integrates three modal data of drilling footage time sequence, mud multidimensional parameters, and visual drill bit offset, can adaptively update the formation benchmark, and collaboratively identify fault types based on improved evidence theory, in order to solve the industry pain points of false alarms, monitoring lag, and lack of automatic control in complex hard rock pile foundation construction. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies and solve at least one of the technical problems mentioned in the background art, this invention proposes an impact drilling anomaly identification method based on the coordination of drilling speed and mud parameters.

[0007] The technical solution adopted by this invention to solve its technical problem is: the percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters, as described in this invention, includes the following steps: S1. Synchronous acquisition of multi-source heterogeneous sensing data: Establish a synchronous sensing acquisition link for three types of time series, fluid and vision, and synchronously acquire impact drilling footage time series dataset, mud multidimensional physical property parameter set, and drill wire rope visual offset image stream. Add a unified timestamp to all acquired data to complete time series alignment. S2. Adaptive preprocessing of multi-source data: Temporal noise reduction, fluid parameter normalization, and image enhancement and correction are performed on three types of sensing data respectively to remove outlier abnormal data caused by environmental interference and construct standardized fusion feature vectors. S3. Dynamic update of stratigraphic benchmark features: Based on the sliding time series window and combined with prior stratigraphic exploration data, the online incremental clustering algorithm is used to iteratively update the current stratigraphic normal working condition benchmark feature library and dynamically correct the normal operating range boundaries of each parameter. S4. Calculation of confidence in three-modal feature fusion: Input the preprocessed footage time sequence features, mud physical property features, and visual deviation features into the improved DS evidence theory collaborative discrimination model to generate independent evidence bodies corresponding to various potential faults, and calculate the joint confidence of various faults. S5. Quantitative Judgment of Multi-parameter Coupled Anomalies: Construct a quantitative evaluation system for coupled anomalies of footage-mud-drill string deviation, and divide the working condition range into three levels: normal, minor hidden danger, and severe fault, based on the joint confidence of each fault. S6. Graded early warning and intelligent closed-loop control: Based on the working condition judgment result, output local audible and visual graded alarms and push alarm information remotely to the cloud. Simultaneously send stroke and impact frequency adaptive adjustment instructions to the impact drill PLC control system. After the abnormal working condition continues to exceed the limit, lock the equipment and push manual intervention prompt.

[0008] Furthermore, the synchronous acquisition of multi-source heterogeneous sensing data in step S1 specifically includes the following sub-steps: S11. Advancement timing data acquisition: The timing sequence of the lifting displacement of the punch is acquired by a rotary encoder installed on the pulley block. The timing sequence differential operation of the displacement timing is used to generate the timing characteristics of the advance speed, advance acceleration, and advance fluctuation variance within a unit timing window. S12. Multidimensional physical property parameter acquisition of mud: The three core mud parameters, namely mud density, real-time mud level and level change rate, are collected synchronously through the in-hole tuning fork density sensor and the wireless liquid level sensor in the mud tank. S13. Drill tool visual offset image acquisition: An industrial camera equipped with an adaptive LED supplementary lighting module is mounted on the drill rig mast to acquire color image streams of the pile hole and wire rope at a fixed sampling frame rate, and the on-site light intensity is acquired simultaneously to assist in image pixel scale correction.

[0009] Furthermore, step S2, multi-source data adaptive preprocessing, includes the following sub-steps: S21. Preprocessing of advance timing: Kalman timing filtering algorithm is used to eliminate noise caused by encoder vibration. A fixed-length sliding timing window is set to extract four-dimensional timing features of advance mean, fluctuation variance, change slope and sudden change amplitude within the window. S22. Mud parameter preprocessing: The Min-Max interval normalization algorithm is used to map mud density, liquid level and liquid level change rate to a unified evidence domain, and a first-order difference operator is introduced to eliminate the baseline drift problem caused by slow mud seepage. S23. Visual image preprocessing: Adaptive gamma correction combined with bilateral filtering is used to complete image noise reduction. The RGB image is converted to the HSV color space and the pixel contours of the manually marked pile hole area and wire rope area are extracted by threshold segmentation. The pixel-level offset distance is output and the physical offset of the drill bit is calculated by combining the camera calibration ratio.

[0010] Furthermore, step S3, which dynamically updates the stratigraphic benchmark features, employs an online incremental K-Means clustering algorithm. After each set of sliding time-series window samples is collected, the cluster center is iteratively updated, and the upper and lower limits of the normal benchmark interval for each parameter of the current stratigraphy are dynamically output.

[0011] Furthermore, step S4, the construction of the improved DS evidence theory collaborative discriminant model, includes the following sub-steps: S41. Generation of independent evidence: Generate footage evidence based on footage feature deviation function, mud evidence based on mud parameter deviation function, and drill string deviation evidence based on visual offset deviation function. The evidence domain is uniformly set to include five types of identification targets: normal working conditions, slow footage in hard rock, mud failure and hole collapse, drill string deviation and hammer jamming, and formation slurry leakage. S42. Adaptive Conflict Weight Allocation: With the dual optimization objectives of maximizing fault identification accuracy and minimizing false alarm rate, an improved particle swarm optimization algorithm is used to solve for the optimal fusion weight of the three types of evidence. S43. Weighted Evidence Fusion Calculation: Introducing a weighted modified conflict allocation rule to complete the multi-source evidence fusion calculation and outputting the joint confidence level corresponding to various faults.

[0012] Furthermore, in step S23, the visual image is used to extract the drill bit offset using an attention-enhanced lightweight YOLOv8 network. The YOLOv8 network embeds an EMA multi-scale attention module and uses the WIoU weighted intersection and union loss function to complete the model training. The network outputs the center pixel coordinates of the pile hole and the wire rope contour fitting straight line, and solves the actual physical offset of the drill bit simultaneously.

[0013] Furthermore, the closed-loop control logic in step S6 includes: when the working condition is determined to be normal slow advance in hard rock, the PLC control system automatically increases the stroke and impact frequency of the hammer; when the condition is determined to be a minor potential failure of the mud, the impact parameters are slightly reduced and a mud preparation prompt is sent; when the condition is determined to be a severe fault such as severe deviation of the drill string, hole collapse, or severe leakage of the formation, the equipment is immediately shut down and a dual alarm is triggered simultaneously on the local and cloud sides.

[0014] Furthermore, an intelligent anomaly identification system for impact drills includes: a multi-source synchronous sensing module, an edge preprocessing calculation unit, a formation benchmark dynamic update unit, an improved DS fusion discrimination unit, a graded early warning output unit, a PLC collaborative control unit, and a 4G cloud transmission unit. The multi-source synchronous sensing module includes a rotary encoder, a tuning fork density sensor, a wireless liquid level sensor, and an industrial camera with LED supplementary lighting. The edge computing unit deploys a lightweight YOLOv8 inference model, an incremental clustering algorithm, and an improved DS evidence fusion algorithm. The incremental clustering algorithm adopts the online incremental K-Means clustering algorithm. The edge computing unit receives the raw data output by the multi-source synchronous sensing module, completes the entire process of data cleaning, noise reduction, standardization, feature extraction, and image target inference, and outputs standardized fusion feature vectors to supply the downstream stratigraphic benchmark dynamic update unit and the improved DS fusion discrimination unit. The 4G cloud transmission unit synchronously uploads the original sensing data, anomaly judgment results, and device energy consumption operation data to the remote management mini-program / computer client.

[0015] Furthermore, the formation benchmark dynamic update unit relies on the standardized fusion feature vector output by the edge preprocessing calculation unit to continuously iteratively update the normal working condition benchmark feature library of the current drilling formation, and dynamically correct the normal operating range boundaries of the three types of parameters: footage, mud, and drill bit offset. The improved DS fusion discrimination unit is based on the stratigraphic normal interval output by the stratigraphic benchmark dynamic update unit and the standardized features output by the edge preprocessing calculation unit. It completes multimodal evidence construction, weight adaptive optimization, evidence fusion, fault confidence calculation, comprehensive anomaly score quantification, and outputs fault type and fault severity level.

[0016] Furthermore, the hierarchical early warning output unit receives the operating condition classification result and fault type output by the improved DS fusion discrimination unit, outputs hierarchical local early warning prompts, and synchronously packages and transmits the early warning data to the 4G cloud transmission unit. The PLC collaborative control unit receives the working condition judgment instructions issued by the hierarchical early warning output unit, and adjusts the impact drilling operation parameters in a differentiated manner. In case of severe fault conditions, the equipment is directly locked and shut down.

[0017] The beneficial effects of this invention are as follows: This invention significantly improves fault identification accuracy; the three-dimensional parameter collaborative verification can accurately distinguish between low-speed normal advance in hard rock and slow advance due to hole collapse and instability, reducing the failure rate of fault detection in complex basalt formations and effectively reducing invalid false alarms; the visual offset recognition is fully automated and not limited by the operator's technical level, achieving millisecond-level real-time monitoring. It has strong formation self-adaptation capability, reducing on-site manual debugging; the incremental clustering algorithm dynamically updates the formation benchmark, and automatically adapts to the current rock formation parameters after each drilling section is completed, eliminating the need for technicians to repeatedly debug the control system threshold on-site, thus reducing the input of on-site technicians; Automated closed-loop control improves the efficiency of pile foundation drilling; after identifying the working conditions, the system automatically issues stroke and impact frequency adjustment commands: in hard rock conditions, the impact parameters are automatically strengthened to increase the advance; in case of mud hazards, the impact load is reduced in advance to prevent hole collapse; in case of severe deviation or hole collapse, the machine is automatically stopped, which greatly reduces the time spent on fault handling and significantly improves the overall drilling efficiency. The equipment structure is simplified, enabling green and energy-saving construction. This invention relies on a servo motor-driven PLC control system, eliminating the need for a matching air compressor and complex pneumatic pipelines, thus eliminating the additional energy consumption of pneumatic equipment. It can be linked and optimized with the energy management strategy of energy storage diesel generators, resulting in a comprehensive reduction in fuel consumption and adapting to the requirements of green and low-carbon construction of the power grid. Construction safety performance is significantly improved; major safety hazards such as hole collapse, grout leakage, and hammer jamming can be warned in advance, effectively avoiding serious construction accidents such as large-scale instability of the borehole wall and hammer burial, and ensuring the long-term stability of the borehole quality of hard rock power transmission and transformation pile foundations. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a schematic diagram of the method flow in this invention; Figure 2 This is a schematic diagram of the system flow in this invention. Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0021] Example 1 like Figure 1 As shown in the embodiment of the present invention, the percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters includes the following steps: S1. Synchronous acquisition of multi-source heterogeneous sensing data: Establish a synchronous sensing acquisition link for three types of time series, fluid and vision, and synchronously acquire impact drilling footage time series dataset, mud multidimensional physical property parameter set, and drill wire rope visual offset image stream. Add a unified timestamp to all acquired data to complete time series alignment. Step S1, which involves the synchronous acquisition of multi-source heterogeneous sensing data, specifically includes the following sub-steps: S11. Data acquisition of advance timing: The timing sequence of the lifting and lowering displacement of the hammer is acquired by the rotary encoder installed on the pulley block. The timing sequence differential operation of the displacement timing is used to generate the timing characteristics of the advance speed, advance acceleration, and advance fluctuation variance within a unit timing window, which reflects the difficulty of rock strata breaking. S12. Multi-dimensional physical property parameter acquisition of mud: The three core mud parameters, namely mud density, real-time mud level and level change rate, are collected synchronously by the in-hole tuning fork density sensor and the wireless liquid level sensor in the mud tank. These parameters directly characterize the mud wall protection effect and whether there is mud leakage in the formation. S13. Drill tool visual offset image acquisition: An industrial camera equipped with an adaptive LED supplementary lighting module is mounted on the drill rod to acquire color image streams of the pile hole and wire rope at a fixed sampling frame rate, and the on-site light intensity is acquired simultaneously to assist in image pixel scale correction. S2. Adaptive preprocessing of multi-source data: Temporal noise reduction, fluid parameter normalization, and image enhancement and correction are performed on three types of sensing data respectively to remove outlier abnormal data caused by environmental interference and construct standardized fusion feature vectors. Step S2, multi-source data adaptive preprocessing, includes the following sub-steps: S21. Preprocessing of advance timing: Kalman timing filtering algorithm is used to eliminate noise caused by encoder vibration. A fixed-length sliding timing window is set to extract four-dimensional timing features of advance mean, fluctuation variance, change slope and sudden change amplitude within the window. S22. Mud parameter preprocessing: The Min-Max interval normalization algorithm is used to map mud density, liquid level and liquid level change rate to a unified evidence domain, and a first-order difference operator is introduced to eliminate the baseline drift problem caused by slow mud seepage. S23. Visual image preprocessing: Adaptive gamma correction combined with bilateral filtering is used to complete image noise reduction. The RGB image is converted to the HSV color space and the pixel contours of the manually marked pile hole area and wire rope area are extracted by threshold segmentation. The pixel-level offset distance is output and the physical offset of the drill bit is calculated by combining the camera calibration ratio. In step S23, the visual image is used to extract the drill bit offset using an attention-enhanced lightweight YOLOv8 network. The YOLOv8 network embeds an EMA multi-scale attention module and uses the WIoU weighted intersection and union loss function to complete the model training. The network outputs the center pixel coordinates of the pile hole and the wire rope contour fitting straight line, and solves the actual physical offset of the drill bit simultaneously. A lightweight YOLOv8 network with attention enhancement is used to identify the contours of pile holes and wire ropes. An EMA multi-scale attention module is embedded to enhance the extraction of features of small offset targets at long distances and in low light conditions. The model training loss function adopts the WIoU weighted intersection and union loss function, which reduces the target false negative rate in complex construction site scenarios compared with the general CIoU loss. The model inference outputs the contours of pile holes and segmented contours of wire ropes, and the center coordinates of the pile hole and the straight line of the wire rope are fitted to obtain the physical offset of the drill bit. The physical offset of the drill bit is calculated by combining the camera calibration ratio. The final output is the pixel-level offset distance, which is then combined with the camera calibration ratio to calculate the physical offset of the drill bit. S3. Dynamic update of stratigraphic benchmark features: Based on the sliding time series window and combined with prior stratigraphic exploration data, the online incremental clustering algorithm is used to iteratively update the current stratigraphic normal working condition benchmark feature library and dynamically correct the normal operating range boundaries of each parameter. Step S3, dynamic updating of stratigraphic benchmark features, adopts an online incremental K-Means clustering algorithm. After each set of sliding time window samples is collected, the cluster center is iteratively updated, and the upper and lower limits of the normal benchmark interval of each parameter of the current stratigraphy are dynamically output. Instead of the traditional fixed threshold, it updates the characteristic boundary of the current formation under normal working conditions in real time as drilling progresses, and automatically adjusts the discrimination criteria when the geology changes to avoid misjudgment; The specific strategy for using the online incremental K-Means clustering algorithm is as follows: The complete formula for the loss function is as follows: ,in, This represents the total number of samples within the current sliding time window. The time-series decay weight for a single sample is used, with newer samples having higher weights. This is a standardized single multimodal fusion feature vector; For the first The coordinates of the cluster center of each sample; Historical baseline constraint coefficients are used to balance the weights of old and new stratigraphic data. This represents the total number of cluster centers in the previous iteration. , This refers to the m-th stratigraphic reference cluster center in the current iteration and the previous iteration; For each complete sliding window sample collected, the cluster center is iteratively optimized by substituting it into the loss function. The historical benchmark is constrained to not be covered by short-term abnormal samples. After the iteration is completed, the upper and lower limits of the normal benchmark intervals for three types of parameters, namely footage, mud, and offset, are output. S4. Calculation of confidence in three-modal feature fusion: Input the preprocessed footage time-series features, mud physical property features, and visual offset features into the improved DS evidence theory collaborative discrimination model to generate independent evidence bodies corresponding to various potential faults. Calculate the joint confidence of each type of fault and construct independent fault evidence for the three types of data: footage, mud, and visual offset. This solves the problem of one-sided identification by a single sensor. Calculate the confidence of each type of fault occurrence through weighted fusion to quantify the probability of fault occurrence. Step S4, the construction of the improved DS evidence theory collaborative discriminant model, includes the following sub-steps: S41. Generation of Independent Evidence: Generate advance evidence based on the advance feature deviation function. Mud parameter deviation function generates mud evidence. Visual offset deviation function generates evidence of drill string misalignment. The evidence domain is uniformly set to include five categories of identification targets: normal working conditions, slow progress in hard rock, mud failure and borehole collapse, drill string deviation and hammer jamming, and formation slurry leakage. ,in Indicates normal operating conditions Indicates slow advance in hard rock. Indicates mud failure and borehole collapse. Indicates drill bit deviation and hammer clamping. This indicates grout leakage from the formation; S42. Adaptive Conflict Weight Allocation: With the dual optimization objectives of maximizing fault identification accuracy and minimizing false alarm rate, an improved particle swarm optimization algorithm is used to solve for the optimal fusion weights of the three types of evidence. , , ; The formula for calculating the weighted improved DS evidence fusion is as follows: ;in, The overall conflict coefficient of the three sources of evidence characterizes the degree of difference in the judgment results of multi-sensor data; Assign coefficients to fault category conflicts to control the allocation of conflict information to the corresponding fault type; It is any subset of faults within the domain of evidence; Fault subset after fusion The corresponding basic confidence assignment value; S43. Weighted Evidence Fusion Calculation: Introducing a weighted modified conflict allocation rule to complete the multi-source evidence fusion calculation, outputting the joint confidence scores corresponding to various types of faults, and calculating the joint confidence score for each type of fault after fusion. , representing the The likelihood of such failures occurring.

[0022] S5. Quantitative Judgment of Multi-Parameter Coupling Anomalies: Construct a quantitative evaluation system for coupled anomalies of footage, mud, and drill string deviation. Combine the joint confidence of each fault to divide the working condition into three levels: normal, minor hidden danger, and severe fault. Integrate the deviation of each parameter with the fault confidence to uniformly quantify the severity of the fault, divide the working condition into three levels, and achieve refined fault classification. Complete formula for coupling anomaly quantization function: ,in, The overall anomaly score is used to quantify the fault; the higher the score, the more severe the fault. , , Fixed weights were assigned to the following sub-items: footage, mud, and visual offset. The deviation mapping function of the advance rate relative to the formation reference; This is a joint deviation mapping function for mud density and liquid level. This is a mapping function for the physical offset of the drill string; This represents the actual physical offset distance of the drill string; The joint confidence level of the fault type with the highest current confidence level; The system according to The numerical values ​​are divided into three levels: normal range, minor hidden danger warning range, and severe fault shutdown range, and the fault type with the highest confidence level is output simultaneously. S6, graded early warning and intelligent closed-loop control: Based on the working condition judgment result, output local audible and visual graded alarms and push alarm information remotely to the cloud. Simultaneously send stroke and impact frequency adaptive adjustment instructions to the impact drill PLC control system. After the abnormal working condition continues to exceed the limit, lock the equipment and push manual intervention prompt. Step S6 closed-loop control logic includes: when the working condition is determined to be normal slow advance in hard rock, the PLC control system automatically increases the hammer stroke and impact frequency; when the condition is determined to be a minor mud failure risk, the impact parameters are slightly reduced and a mud preparation prompt is pushed; when the condition is determined to be a severe drill bit deviation, hole collapse, or severe formation leakage, the equipment is immediately shut down and a dual alarm is triggered simultaneously on the local and cloud sides.

[0023] Example 2 like Figure 2 As shown, an intelligent anomaly identification system for impact drills includes: a multi-source synchronous sensing module, an edge preprocessing calculation unit, a formation benchmark dynamic update unit, an improved DS fusion discrimination unit, a hierarchical early warning output unit, a PLC collaborative control unit, and a 4G cloud transmission unit. The multi-source synchronous sensing module includes a rotary encoder, a tuning fork density sensor, a wireless liquid level sensor, and an industrial camera with LED supplementary lighting. The edge computing unit deploys a lightweight YOLOv8 inference model, an incremental clustering algorithm, and an improved DS evidence fusion algorithm. The incremental clustering algorithm adopts the online incremental K-Means clustering algorithm. The edge computing unit receives the raw data output by the multi-source synchronous sensing module, completes the entire process of data cleaning, noise reduction, standardization, feature extraction, and image target inference, and outputs standardized fusion feature vectors to supply the downstream stratigraphic benchmark dynamic update unit and the improved DS fusion discrimination unit. The 4G cloud transmission unit synchronously uploads the original sensing data, anomaly judgment results, and device energy consumption operation data to the remote management mini-program / computer client.

[0024] The formation benchmark dynamic update unit relies on the standardized fusion feature vector output by the edge preprocessing calculation unit to continuously iterate and update the normal working condition benchmark feature library of the current drilling formation, and dynamically correct the normal operating range boundaries of the three types of parameters: footage, mud, and drill string offset. The improved DS fusion discrimination unit is based on the stratigraphic normal interval output by the stratigraphic benchmark dynamic update unit and the standardized features output by the edge preprocessing calculation unit. It completes multimodal evidence construction, weight adaptive optimization, evidence fusion, fault confidence calculation, comprehensive anomaly score quantification, and outputs fault type and fault severity level.

[0025] The hierarchical early warning output unit receives the operating condition classification results and fault types output by the improved DS fusion discrimination unit, outputs hierarchical local early warning prompts, and simultaneously packages and transmits the early warning data to the 4G cloud transmission unit. The PLC collaborative control unit receives the working condition judgment instructions issued by the hierarchical early warning output unit, and adjusts the impact drilling operation parameters in a differentiated manner. In case of severe fault conditions, the equipment is directly locked and shut down.

[0026] Example 3 To better illustrate the present invention, this embodiment uses the complete on-site implementation process of Hainan basalt power transmission and transformation pile foundation as an example for detailed explanation; Step 1: Synchronous acquisition of data using multi-source heterogeneous sensing; A rotary encoder is installed on the impact drill pulley block to collect the real-time displacement sequence of the impact hammer; a tuning fork density sensor is embedded in the hole and a wireless liquid level sensor is deployed in the mud pit to synchronously transmit mud density, real-time liquid level, and liquid level change rate; an industrial camera with adaptive LED supplementary lighting is installed on the drill rig upright to collect image streams of the pile hole and wire rope at a fixed frame rate. All sensor data are added with a unified millisecond-level timestamp through the local edge gateway to eliminate the sampling time difference of multiple devices and ensure the timing alignment of multimodal data.

[0027] Step 2: Adaptive preprocessing of multi-source data; Footage advance timing: Kalman filter is used to filter out the mechanical vibration noise of the drilling rig. A fixed-length sliding timing window is set to extract four-dimensional timing features of the footage mean, fluctuation variance, change slope, and abrupt change amplitude within the window, which are used as the input for footage advance evidence. Mud parameters: Min-Max normalization is used to map density, liquid level, and liquid level change rate to a 0-1 unified evidence domain. A first-order difference operator is introduced to eliminate baseline drift caused by slow mud seepage and output mud parameters deviating from the baseline characteristics. Visual imagery: Adaptive gamma correction balances day and night light intensity, and bilateral filtering removes on-site dust, water mist, and noise. RGB images are converted to HSV color space, and the red pile hole area and green wire rope segment area are segmented according to preset color thresholds. A lightweight YOLOv8 network with attention enhancement is used to identify the pile hole and wire rope contours, embedding an EMA multi-scale attention module to strengthen the extraction of small-shift target features under long-distance and low-light conditions. The model training loss function adopts the WIoU weighted intersection and union loss function, which reduces the target false detection rate in complex construction site scenarios compared to the general CIoU loss. The model inference outputs the pile hole contour and wire rope segment contour, and the center coordinates of the pile hole and the straight line of the wire rope are fitted to obtain the physical offset of the drill bit by combining the camera calibration ratio. .

[0028] Step 3: Dynamic incremental update of stratigraphic benchmark features. The edge computing unit runs the online incremental K-Means clustering algorithm, substitutes the loss function described in this manual, and iteratively updates the stratigraphic feature cluster center after each set of sliding window sample collection is completed; the historical benchmark constraint coefficient balances the weight of new and old stratigraphic data to avoid short-term local abnormal samples from polluting the benchmark library; after the iteration is completed, the normal benchmark interval of the current stratigraphic footage, mud, and offset parameters is output, and the original static discrimination threshold of the system is replaced in real time.

[0029] Step 4: Improve the calculation of collaborative confidence level in DS evidence theory; Independent evidence of advance was constructed based on advance characteristics and the difference between the baseline intervals. Constructing mud evidence using mud parameter differences Drill string offset to construct visual evidence The domain of evidence is uniformly divided into five categories of identification targets: normal, slow progress in hard rock, mud collapse, drill bit deviation, and formation leakage. With the dual objectives of maximizing fault identification accuracy and minimizing false alarm rate, an adaptive particle swarm optimization algorithm is used to iteratively solve for the optimal fusion weights of the three types of evidence. ; Calculate the conflict coefficient of the three types of evidence Substitute the values ​​into the weighted fusion formula described in this manual to complete the evidence synthesis and output the joint confidence level corresponding to each type of fault. .

[0030] Step 5: Multi-parameter Coupled Anomaly Quantization Determination. Substitute the sub-item deviation features and the joint confidence of the fault into the coupled anomaly quantization function in this manual to calculate the comprehensive fault score. The system has three preset score ranges, corresponding to normal operating conditions, minor hidden dangers, and serious faults, and simultaneously determines the fault type with the highest confidence level.

[0031] Step 6: Graded early warning and PLC closed-loop control; If the rock is judged to be a normal slow advance in hard rock: the edge unit sends a command to the PLC control system of the impact drill to increase the stroke and impact frequency, which automatically enhances the rock breaking efficiency. The problem was identified as a minor potential failure of the mud: the impact parameters were slightly reduced, local audio and visual prompts were given to staff to prepare mud materials, and the potential failure record was simultaneously recorded and pushed to the cloud. If the fault is determined to be severe drill bit deviation, severe borehole collapse, or formation grout leakage: immediately control the servo winch to stop, trigger a local high-decibel audible and visual alarm, and simultaneously push the fault type, on-site image, and equipment number to the management personnel's mini-program via the 4G network for remote alarm. The equipment will be locked until manual reset is performed.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A percussion drilling anomaly identification method based on the synergy of drilling speed and mud parameters, characterized in that: Includes the following steps: S1. Synchronous acquisition of multi-source heterogeneous sensing data: Establish a synchronous sensing acquisition link for three types of time series, fluid and vision, and synchronously acquire impact drilling footage time series dataset, mud multidimensional physical property parameter set, and drill wire rope visual offset image stream. Add a unified timestamp to all acquired data to complete time series alignment. S2. Adaptive preprocessing of multi-source data: Temporal noise reduction, fluid parameter normalization, and image enhancement and correction are performed on three types of sensing data respectively to remove outlier abnormal data caused by environmental interference and construct standardized fusion feature vectors. S3. Dynamic update of stratigraphic benchmark features: Based on the sliding time series window and combined with prior stratigraphic exploration data, the online incremental clustering algorithm is used to iteratively update the current stratigraphic normal working condition benchmark feature library and dynamically correct the normal operating range boundaries of each parameter. S4. Calculation of confidence in three-modal feature fusion: Input the preprocessed footage time sequence features, mud physical property features, and visual deviation features into the improved DS evidence theory collaborative discrimination model to generate independent evidence bodies corresponding to various potential faults, and calculate the joint confidence of various faults. S5. Quantitative Judgment of Multi-parameter Coupled Anomalies: Construct a quantitative evaluation system for coupled anomalies of footage-mud-drill string deviation, and divide the working condition range into three levels: normal, minor hidden danger, and severe fault, based on the joint confidence of each fault. S6. Graded early warning and intelligent closed-loop control: Based on the working condition judgment result, output local audible and visual graded alarms and push alarm information remotely to the cloud. Simultaneously send stroke and impact frequency adaptive adjustment instructions to the impact drill PLC control system. After the abnormal working condition continues to exceed the limit, lock the equipment and push manual intervention prompt.

2. The percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters according to claim 1, characterized in that: Step S1, which involves the synchronous acquisition of multi-source heterogeneous sensing data, specifically includes the following sub-steps: S11. Advancement timing data acquisition: The timing sequence of the lifting displacement of the punch is acquired by a rotary encoder installed on the pulley block. The timing sequence differential operation of the displacement timing is used to generate the timing characteristics of the advance speed, advance acceleration, and advance fluctuation variance within a unit timing window. S12. Multidimensional physical property parameter acquisition of mud: The three core mud parameters, namely mud density, real-time mud level and level change rate, are collected synchronously through the in-hole tuning fork density sensor and the wireless liquid level sensor in the mud tank. S13. Drill tool visual offset image acquisition: An industrial camera equipped with an adaptive LED supplementary lighting module is mounted on the drill rig mast to acquire color image streams of the pile hole and wire rope at a fixed sampling frame rate, and the on-site light intensity is acquired simultaneously to assist in image pixel scale correction.

3. The percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters according to claim 1, characterized in that: Step S2, multi-source data adaptive preprocessing, includes the following sub-steps: S21. Preprocessing of advance timing: Kalman timing filtering algorithm is used to eliminate noise caused by encoder vibration. A fixed-length sliding timing window is set to extract four-dimensional timing features of advance mean, fluctuation variance, change slope and sudden change amplitude within the window. S22. Mud parameter preprocessing: The Min-Max interval normalization algorithm is used to map mud density, liquid level and liquid level change rate to a unified evidence domain, and a first-order difference operator is introduced to eliminate the baseline drift problem caused by slow mud seepage. S23. Visual image preprocessing: Adaptive gamma correction combined with bilateral filtering is used to complete image noise reduction. The RGB image is converted to the HSV color space and the pixel contours of the manually marked pile hole area and wire rope area are extracted by threshold segmentation. The pixel-level offset distance is output and the physical offset of the drill bit is calculated by combining the camera calibration ratio.

4. The percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters according to claim 1, characterized in that: Step S3, dynamic updating of stratigraphic benchmark features, employs an online incremental K-Means clustering algorithm. After each set of sliding time-series window samples is collected, the cluster centers are iteratively updated, and the upper and lower limits of the normal benchmark interval for each parameter of the current stratigraphy are dynamically output.

5. The percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters according to claim 1, characterized in that: Step S4, the construction of the improved DS evidence theory collaborative discriminant model, includes the following sub-steps: S41. Generation of independent evidence: Generate footage evidence based on footage feature deviation function, mud evidence based on mud parameter deviation function, and drill string deviation evidence based on visual offset deviation function. The evidence domain is uniformly set to include five types of identification targets: normal working conditions, slow footage in hard rock, mud failure and hole collapse, drill string deviation and hammer jamming, and formation slurry leakage. S42. Adaptive Conflict Weight Allocation: With the dual optimization objectives of maximizing fault identification accuracy and minimizing false alarm rate, an improved particle swarm optimization algorithm is used to solve for the optimal fusion weight of the three types of evidence. S43. Weighted Evidence Fusion Calculation: Introducing a weighted modified conflict allocation rule to complete the multi-source evidence fusion calculation and outputting the joint confidence level corresponding to various faults.

6. The percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters according to claim 3, characterized in that: In step S23, the drill bit offset is extracted from the visual image using an attention-enhanced lightweight YOLOv8 network. The YOLOv8 network embeds an EMA multi-scale attention module and uses the WIoU weighted intersection and union loss function to complete the model training. The network outputs the center pixel coordinates of the pile hole and the wire rope contour fitting straight line, which are then used to solve for the actual physical offset of the drill bit.

7. The percussion drilling anomaly identification method based on the coordination of drilling speed and mud parameters according to claim 5, characterized in that: The closed-loop control logic of step S6 includes: when it is determined to be a normal slow advance condition in hard rock, the PLC control system automatically increases the stroke and impact frequency of the hammer; when it is determined to be a minor potential failure of the mud, the impact parameters are slightly reduced and a mud preparation prompt is pushed; when it is determined to be a severe fault of severe drill bit deviation, hole collapse, or formation leakage, the equipment is immediately stopped and a dual alarm is triggered simultaneously on the local and cloud sides.

8. An intelligent anomaly identification system for impact drills based on the method of any one of claims 1-7, characterized in that, include: Multi-source synchronous sensing module, edge preprocessing calculation unit, stratigraphic benchmark dynamic update unit, improved DS fusion discrimination unit, hierarchical early warning output unit, PLC collaborative control unit, 4G cloud transmission unit; The multi-source synchronous sensing module includes a rotary encoder, a tuning fork density sensor, a wireless liquid level sensor, and an industrial camera with LED supplementary lighting. The edge computing unit deploys a lightweight YOLOv8 inference model, an incremental clustering algorithm, and an improved DS evidence fusion algorithm. The incremental clustering algorithm adopts the online incremental K-Means clustering algorithm. The edge computing unit receives the raw data output by the multi-source synchronous sensing module, completes the entire process of data cleaning, noise reduction, standardization, feature extraction, and image target inference, and outputs standardized fusion feature vectors to supply the downstream stratigraphic benchmark dynamic update unit and the improved DS fusion discrimination unit. The 4G cloud transmission unit synchronously uploads the original sensing data, anomaly judgment results, and device energy consumption operation data to the remote management applet / computer client.

9. The percussion drilling anomaly identification system based on the coordination of drilling speed and mud parameters according to claim 8, characterized in that: The formation benchmark dynamic update unit relies on the standardized fusion feature vector output by the edge preprocessing calculation unit to continuously iterate and update the normal working condition benchmark feature library of the current drilling formation, and dynamically correct the normal operating range boundaries of the three types of parameters: footage, mud, and drill string offset. The improved DS fusion discrimination unit is based on the stratigraphic normal interval output by the stratigraphic benchmark dynamic update unit and the standardized features output by the edge preprocessing calculation unit. It completes multimodal evidence construction, weight adaptive optimization, evidence fusion, fault confidence calculation, comprehensive anomaly score quantification, and outputs fault type and fault severity level.

10. The percussion drilling anomaly identification system based on the coordination of drilling speed and mud parameters according to claim 9, characterized in that: The hierarchical early warning output unit receives the operating condition classification result and fault type output by the improved DS fusion discrimination unit, outputs hierarchical local early warning prompts, and synchronously packages and transmits the early warning data to the 4G cloud transmission unit. The PLC collaborative control unit receives the working condition judgment instructions issued by the hierarchical early warning output unit, and adjusts the impact drilling operation parameters in a differentiated manner. In case of severe fault conditions, the equipment is directly locked and shut down.