A method and device for sampling undisturbed soil in deep overburden

Abstract

This invention primarily relates to the field of undisturbed soil sampling control technology, aiming to address the problem of difficulty in identifying lateral constraint continuity during sampling of undisturbed soil in deep overburden layers, which easily leads to core instability and disturbance of the undisturbed structure. This invention provides a method and apparatus for sampling undisturbed soil in deep overburden layers, constructing a closed-loop technical path from signal acquisition, feature separation, state determination to execution control around the tube exit critical section. By using the advance displacement as a recursive reference for position-by-position encoding and pre-setting dual potential spaces, combined with the calculation of the projected density ratio using a reference response group, effective separation of lateral support changes and axial load changes is achieved. Based on this, the constraint interruption interval is accurately identified according to the dual conditions of continuous attenuation gradient and duration, automatically generating exit control commands and encapsulation trigger commands, effectively improving the control of soil sample instability disturbances near the tube exit.

Description

Technical Field

[0001] This invention relates to the field of undisturbed soil sampling control technology, and in particular to a method and apparatus for sampling undisturbed soil in deep overburden layers. Background Technology

[0002] Undiscovered soil samples, which can preserve the natural structure, water content, and stratigraphic relationships relatively intact, have always been an important foundation for geotechnical engineering investigation, deep overburden testing, and laboratory mechanical analysis. In deep overburden scenarios, deep soil layers are typically loose, water-sensitive, and have poor self-stability, which places higher demands on disturbance control during the sampling process.

[0003] Existing undisturbed soil sampling techniques mainly follow two routes: structural sampling and process optimization. One type uses direct-push or percussion-direct-push sampling drills, employing an inner percussion tube, an outer percussion tube, a probe, and a drill bit to press soil samples into the sampling channel for analysis. Another type uses a semi-closed or semi-closed tube structure, consisting of a percussion shaft, an outer percussion tube, a semi-closed tube, and a sampling drill bit, emphasizing minimizing disturbance during the penetration stage. Other schemes incorporate a semi-closed tube, swivel, blades, and a tube shoe at the bottom of the sampling tube, with a base support or opening / closing component to receive the core sample. For deep overburden scenarios, some schemes propose staged drilling with accompanying undisturbed sample collection drills for obtaining large-diameter undisturbed samples from ultra-deep overburden layers. In terms of international research, existing results mainly focus on the geometric parameters of the sampling tube, cutting angle, wall thickness, piston configuration, diameter, and the stress and deformation during sampling and ejection. They point out that the sampler penetration and core ejection stages are prone to water loss and structural changes, and thin-walled sampling and ejection control have become common directions for improvement.

[0004] While existing technologies can reduce overall disturbance by focusing on drill string structure, cutting edge geometry, bottom sealing, and deep drilling processes, their emphasis is primarily on the sampling entry into the soil layer and the overall extraction stage. For the critical process of sample removal—where the soil sample tip approaches the outlet and continuous pipe wall support is about to end and the sample is transferred to short-distance support—existing solutions typically lack a continuous judgment mechanism for advancing displacement. This makes it difficult to distinguish between changes in axial thrust and local support changes at the outlet, and also lacks a unified processing link that directly links the advancement mechanism's extraction and sealing triggers based on the critical interval. Therefore, in scenarios with deep overburden, the area near the outlet remains a concentrated location for core bulging, fracture, sequence disturbances, and amplified instability.

[0005] Therefore, how to identify the lateral constraint continuity interruption interval in the critical scenario of soil sample transfer from continuous pipe wall support to short-range support after passing through the outlet, and accordingly generate propulsion and sealing control actions to avoid core instability and destruction of the original structure near the outlet, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This invention provides a method and apparatus for sampling undisturbed soil in deep overburden layers, aiming to solve the problem that during the sampling process of undisturbed soil in deep overburden layers, the lateral constraint continuity is difficult to identify in the critical stage of soil sample transfer from continuous pipe wall support to short-range support after passing through the pipe outlet, which easily leads to core instability and disturbance of the undisturbed structure.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems

[0008] On one hand, the present invention provides a method for sampling undisturbed soil in deep overburden layers, applicable to a sampling device comprising a semi-compound pipe, a screw-driven mechanism, and a short-range support component at the pipe outlet, the method comprising:

[0009] S1. Obtain the propulsion displacement, axial thrust, and outlet contact response of the sampling device during the sampling process, and divide the outlet critical response sequence according to the propulsion displacement;

[0010] S2. Based on the set coding structure, the position of the critical response sequence of the tube exit is encoded according to the propulsion displacement to construct a candidate latent variable sequence; wherein, the coding structure includes the target potential space for bearing lateral support changes and the interference potential space for bearing axial load changes;

[0011] S3. Construct a reference response group using the stable response segment corresponding to continuous pipe wall support and the critical response segment corresponding to support transition;

[0012] S4. Calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group, assign the candidate latent variables to the target latent space and the interference latent space, and use the latent variables assigned to the target latent space as the target latent variable sequence.

[0013] S5. Recover the lateral constraint continuity state quantity along the advance displacement based on the target latent variable sequence, and generate the lateral constraint continuity determination quantity corresponding to the advance displacement.

[0014] S6. Calculate the continuous attenuation gradient and duration of the lateral constraint continuity determination quantity. When the continuous attenuation gradient and duration meet the preset interruption conditions, determine the corresponding propulsion displacement interval as the constraint interruption interval.

[0015] S7. Generate the ejection control command for the screw drive mechanism and the encapsulation trigger command for the short-range support component at the outlet pipe based on the determined constraint interruption interval.

[0016] Furthermore, step S1 includes:

[0017] Align the propulsion displacement, axial thrust, and outlet contact response time to form the original corresponding set;

[0018] Based on the displacement difference between adjacent advance displacements, a continuous screening is performed, and the corresponding groups that meet the preset continuous sampling conditions are taken as continuous corresponding groups.

[0019] The range near the outlet is determined based on the changes in the propulsion displacement position and the contact response at the outlet. The continuous corresponding groups within the range near the outlet are taken as the critical segment corresponding groups.

[0020] Arrange the propulsion displacement, axial thrust, and outlet contact response in the corresponding group of the critical section in ascending order of propulsion displacement to form the outlet critical response sequence.

[0021] Furthermore, the encoding structure includes a first recursive encoding layer, a second recursive encoding layer, a compression layer, a mean layer, a variance layer, and an allocation framework; step S2, which involves encoding the position of the exit pipe critical response sequence according to the propulsion displacement, specifically includes:

[0022] The propulsion displacement, axial thrust, and outlet contact response corresponding to each propulsion position are combined into a recursive input unit to form a position input sequence.

[0023] According to the displacement recursion benchmark, the position input sequence is sequentially input into the first recursive coding layer and the second recursive coding layer to form the coding result;

[0024] The compression layer performs compression mapping on the encoded results, and the mean and variance layers calculate the mean and variance to form latent variables to be assigned.

[0025] The latent variables to be assigned are input into the assignment framework consisting of the target latent space and the disturbance latent space to form intermediate latent variables;

[0026] The intermediate latent variables are arranged in the order of their advance displacement to form a sequence of candidate latent variables.

[0027] Furthermore, step S4, calculating the projection density ratio of the candidate latent variable sequence relative to the reference response group, specifically includes:

[0028] The reference response group is input into the coding structure, and the similarity between the candidate latent variable sequence and the critical response segment and the stable response segment is calculated respectively. The ratio of the similarity is used as the projection density ratio.

[0029] Candidate latent variables are assigned to the target latent space and the interference latent space based on the projection density ratio and a preset threshold.

[0030] Furthermore, the method for calculating the projection density ratio is as follows:

[0031] ;

[0032] Indicates the first The projection density ratio corresponding to each propulsion position; This represents the total number of propulsion positions in the critical response sequence of the outlet tube; This represents the index of the advance position in the candidate latent variable sequence; Indicates the advance position index in the reference response group; Indicates the first 8D projection results of each propulsion position; Indicates the first The 8-dimensional projection results corresponding to the critical reference candidate latent variables; Indicates the first The 8-dimensional projection results corresponding to the stable reference candidate latent variables; and These represent the squared Euclidean distances between the 8-dimensional projection results; This represents the squared distance scale parameter. This represents a positive zero constant;

[0033] Furthermore, step S4, which involves assigning candidate latent variables to the target latent space and the interference latent space, includes:

[0034] If the projection density ratio of the current advancement position is greater than a preset threshold, and all advancement positions within the preset advancement displacement range satisfy the condition that the projection density ratio is greater than the threshold, the candidate latent variable corresponding to the current advancement position is assigned to the target latent space.

[0035] If the projection density ratio of the current advancing position is not greater than a preset threshold, and all advancing positions within the preset advancing displacement range satisfy the condition that the projection density ratio is not greater than the threshold, the candidate latent variable corresponding to the current advancing position is assigned to the interference potential space.

[0036] When there are projection density ratios that are both greater than and less than the preset threshold within the preset propulsion displacement range, the candidate latent variable allocation direction is determined based on the majority voting principle.

[0037] Furthermore, step S5 includes:

[0038] The target latent variable sequence is expanded according to the advance displacement order, and the correlation between adjacent advance positions is established to form the state recovery input sequence;

[0039] The position state result is obtained by performing a progressive displacement recursive calculation on the state recovery input sequence. The position state result that meets the preset continuous recovery condition is determined as the continuously changing result. The continuously changing result is cumulatively mapped along the progressive displacement to form the lateral constraint continuous state quantity.

[0040] The state difference and state continuation are calculated based on the lateral constraint continuous state quantities, and then combined to generate the lateral constraint continuity determination quantity.

[0041] Furthermore, determining the constraint interruption interval in step S6 specifically includes:

[0042] The ratio of the difference between adjacent lateral constraint continuity determination quantities to the advance displacement interval is calculated according to the advance displacement sequence to form a continuous decay gradient sequence;

[0043] Extract the range of continuous propulsion displacement that meets the preset gradient conditions, and calculate the displacement span of the corresponding range as the duration.

[0044] When the continuous decay gradient is not greater than the set gradient threshold and the duration is not less than the set length threshold, the corresponding propulsion displacement start and end points are determined to form a constraint interruption interval.

[0045] Furthermore, step S7, which generates the ejection control instruction and the encapsulation trigger instruction, specifically includes:

[0046] Extract the initial thrust displacement, the final thrust displacement, and the interval length from the constrained interruption interval to form the push-out control interval;

[0047] The ejection range of the lead screw propulsion mechanism is determined based on the ejection control range, and an ejection control command containing the initial propulsion displacement, the final propulsion displacement, the range length, and the ejection enable flag is generated.

[0048] The encapsulation trigger position is determined based on the termination propulsion displacement and interval length, and compared with the preset encapsulation conditions to generate an encapsulation trigger command containing the encapsulation trigger position, interval length, and encapsulation enable flag.

[0049] On the other hand, the present invention also provides a sampling device for undisturbed soil in deep overburden layers, comprising:

[0050] A sequence forming unit is used to form a critical response sequence for tube exit based on propulsion displacement, axial thrust, and tube exit contact response.

[0051] The encoding unit is used to encode the critical response sequence of the tube outlet position by position based on the advance displacement to form a candidate latent variable sequence; the encoding unit presets a target potential space for bearing lateral support changes and an interference potential space for bearing axial load changes;

[0052] The allocation unit is used to construct a reference response group with stable response segments and critical response segments, calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group, allocate the candidate latent variables to the target latent space or the interference latent space according to the calculation results, and extract the latent variables allocated to the target latent space to form the target latent variable sequence.

[0053] The state recovery unit is used to recover the lateral constraint continuous state quantity along the propulsion displacement according to the target latent variable sequence, and generate the lateral constraint continuity determination quantity.

[0054] The interval determination unit is used to determine the continuous decay gradient and duration of the lateral constraint continuity determination quantity, and to determine the constraint interruption interval when the preset interruption condition is met.

[0055] The control unit is used to generate the ejection control command and the encapsulation trigger command of the lead screw propulsion mechanism according to the constraint interruption interval.

[0056] Beneficial effects of the present invention

[0057] (1) Accurate identification of lateral constraint continuity in the critical section of pipe detachment is achieved. This invention determines the continuity of lateral constraint through a dual condition of continuous attenuation gradient and duration. Only when the rate of decrease in support strength meets the preset gradient condition and the duration of decrease reaches the preset threshold is it confirmed as a constraint interruption interval. Compared with the existing technology that relies on single-point signal mutation or empirical threshold for judgment, by effectively identifying the constraint interruption interval, the misjudgment rate and missed judgment rate of core bulging, fracture, and stratigraphic disturbance amplification near the pipe outlet are significantly reduced. This significantly improves the accuracy and reliability of identifying the lateral constraint interruption interval in the critical section of pipe detachment, thereby improving the success rate and sample integrity of undisturbed soil sampling in deep overburden layers, and providing quantitative data support for sampling process analysis, system parameter optimization, and quality traceability.

[0058] (2) Effective separation of lateral support variation and axial load variation is achieved. In the coding structure, the present invention pre-defines the target potential space for bearing lateral support variation and the interference potential space for bearing axial load variation. By calculating the projection density ratio of the candidate latent variable sequence relative to the reference response group, the two types of variations from different physical sources are separated. This enables the lateral constraint transfer process near the outlet to obtain a more stable intermediate characterization, and establishes a latent variable basis that can be directly accepted for subsequent determination of constraint interruption interval.

[0059] (3) A closed-loop linkage from state recognition to execution control is realized. After determining the constraint interruption interval, the present invention automatically generates the push-out control command and the sealing trigger command of the screw propulsion mechanism, directly converting the judgment result into mechanical execution action. Compared with the existing technology where the recognition and control actions are disconnected, this invention can automatically complete the push-out range limitation and sealing trigger at the critical moment when the soil sample is about to lose lateral constraint, effectively reducing the probability of soil sample bulging, fracture, and stratification disturbance near the outlet.

[0060] (4) Using the advancing displacement as the recursive benchmark ensures consistency with the physical process. This invention uses the advancing displacement as the recursive benchmark for coding and judgment, anchoring the coding process to the actual position transfer process of the soil core relative to the outlet. Compared with the existing technology that organizes the sequence according to the natural sampling time, it avoids the position drift problem caused by uneven advancing speed, ensures that the judgment result strictly corresponds to the actual spatial position of the soil sample, and improves the consistency and reproducibility of the judgment.

[0061] (5) Adaptive judgment is achieved using a reference response group, without relying on a fixed threshold. This invention constructs a reference response group by extracting the stable response segment corresponding to continuous pipe wall support and the critical response segment corresponding to support transition during the sampling process. The projection density ratio (the ratio of the similarity between the candidate latent variable and the critical response segment to the similarity between the candidate latent variable and the stable response segment) is used as the allocation basis. Compared to the prior art's reliance on preset fixed thresholds, this invention does not rely on absolute value judgments and has good adaptive capabilities for different soil layers, depths, and device conditions, exhibiting stronger generalization performance.

[0062] (6) Improved allocation stability through continuity comparison rules. In the process of allocating the projection density ratio, this invention not only compares the projection density ratio of the current advancing position with the threshold, but also combines the continuity comparison of adjacent advancing positions and the majority voting principle for comprehensive judgment. Compared with the point-by-point independent allocation method, this invention effectively suppresses the interference of single-point noise on the allocation result and improves the robustness and stability of the allocation process. Attached Figure Description

[0063] Figure 1 This is a flowchart of a method for sampling undisturbed soil from a deep overburden layer as described in this invention.

[0064] Figure 2 A flowchart for generating the critical response sequence of the outlet tube;

[0065] Figure 3 Flowchart for generating candidate latent variable sequences;

[0066] Figure 4 Create a flowchart for the target latent variable sequence;

[0067] Figure 5 A flowchart for determining the continuity of lateral constraints;

[0068] Figure 6 Flowchart for determining the constraint interruption interval.

[0069] Figure 7 Flowchart for issuing control instructions and encapsulating trigger instructions;

[0070] Figure 8 Thermodynamic diagram of the lateral support strength field;

[0071] Figure 9 A schematic diagram of the three-dimensional surface and shadow of the perturbation energy field;

[0072] Figure 10 Contour map of key thresholds for lateral support instability index. Detailed Implementation

[0073] Sampling scenarios involving thick overburden undisturbed soil passing through the outlet and undergoing support transfer directly impact the critical stage of detachment from the pipe. Lateral constraints are difficult to identify stably, and core samples are prone to bulging, fracture, and sequence disturbance.

[0074] The core of this invention lies in the following: under the condition that only three field data—propulsion displacement, axial thrust, and outlet contact response—can be stably acquired, a critical response sequence for outlet is first constructed around the vicinity of the outlet, with propulsion displacement set as the recursive main line, so that the encoding process is consistent with the position transfer process of the soil core relative to the outlet. Subsequently, the mixed response is encoded segment by segment using the encoding structure, while retaining the lateral support change and axial load change in the unified latent variable expression. Then, a latent variable mechanism for separating the support source is established through the target latent space and the interference latent space. On this basis, the stable response segment and the critical response segment constitute a reference response group, and the projection density ratio is calculated in the allocation stage, so that the change component that can characterize the support transfer enters the target latent variable sequence, further restoring the lateral constraint continuous state quantity and the lateral constraint continuity judgment quantity. Finally, the constraint interruption interval is identified and the screw extension control command and the encapsulation trigger command are generated in linkage, thus forming a closed-loop technical path from response acquisition, mechanism separation, state recovery to execution control.

[0075] Compared to general sampling schemes that rely on structural support, empirical thresholds, or sudden changes in single-channel signals for judgment, this invention firstly adopts a recursive encoding method based on the advancing displacement in its algorithm structure and judgment chain. It no longer organizes response sequences along natural sampling times, thus making the model closer to the actual process of the transition from continuous pipe wall support to short-range support, reducing the risk of mixing local fluctuations at different advancing positions. Secondly, this invention limits the dual potential space to lateral support change channels and axial load change channels, and introduces reference response groups and projection density ratio constraints when candidate latent variables enter the allocation framework. This transforms the potential allocation basis from a simple encoding result to a structured basis combining support state differences, enhancing the directionality of the target latent variable sequence towards the source of lateral constraint changes. Thirdly, this invention further restores the target latent variable sequence to a continuous state trajectory and determines the constraint interruption interval by combining continuous decay gradients and duration lengths. This allows the judgment results to directly correspond to the advancing displacement range and subsequent control actions, avoiding judgment remaining at the offline analysis level. Through these improvements, a judgmental, controllable, and encapsulated continuous processing mechanism for the risk of pipe derailment in deep overburden undisturbed soil is achieved.

[0076] The technical solutions in this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0077] The method for sampling undisturbed soil with deep overburden described in this invention is applicable to sampling devices including a semi-compound pipe, a screw propulsion mechanism, and a short-range support component at the outlet.

[0078] like Figure 1 As shown, the method for sampling undisturbed soil from deep overburden layers according to the present invention includes the following steps:

[0079] Step S1: Divide the critical response sequence of the tube.

[0080] like Figure 2 As shown, the acquisition of the critical response sequence for pipe exit includes: acquiring the original sampling sequences of propulsion displacement, axial thrust, and pipe outlet contact response, and aligning them in time to form an original corresponding group; performing continuity screening based on the displacement difference between adjacent propulsion displacements, and selecting corresponding groups that meet the preset continuous sampling conditions as continuous corresponding groups; determining the range adjacent to the pipe outlet based on the propulsion displacement position and the change in pipe outlet contact response, and selecting continuous corresponding groups located within the range adjacent to the pipe outlet as critical segment corresponding groups; arranging the propulsion displacement, axial thrust, and pipe outlet contact response in the critical segment corresponding groups in ascending order of propulsion displacement to form the critical response sequence for pipe exit.

[0081] Specifically, the original sampling sequence of the propulsion displacement is denoted as ,in Indicates the first Each propulsion displacement sampling time, Indicates the first One propulsion displacement value, Indicates the total number of displacement samples taken;

[0082] The original sampling sequence of axial thrust is denoted as ,in Indicates the first Each axial thrust sampling time, Indicates the first One axial thrust value, Indicates the total number of axial thrust samples;

[0083] The original sampling sequence of the outlet contact response is denoted as ,in Indicates the first Each outlet contact response sampling time, Indicates the first The contact response value at the outlet pipe. This indicates the total number of contact response samples taken at the outlet. The short-range support component at the outlet is located near the outlet to provide short-distance transition support after the soil sample tip detaches from the continuous support of the semi-closed pipe wall. It is used to characterize the changes in local support conditions when a soil sample passes near the outlet.

[0084] When establishing the correspondence between the same acquisition time, the original sampling sequence of the propulsion displacement is used as the alignment reference. For any propulsion displacement sampling time... Searching for the satisfying in the original axial thrust sampling sequence sampling points, The preset time matching window is defined; among all sampling points that meet the conditions, the sampling point with the smallest time difference is selected as the matching result, and the corresponding index is denoted as [index missing]. Using the same rule, the corresponding index is determined in the original sampling sequence of the outlet contact response. .when and When both exist simultaneously, the first... Original corresponding record When any index does not exist, the current propulsion displacement sampling point does not participate in the corresponding construction. Arrange all the original corresponding records in natural order to obtain the original corresponding group.

[0085] When performing continuous filtering on the original corresponding groups, the displacement difference between adjacent records is calculated according to the order of the advance displacement in the original corresponding records. The advance displacement difference between two adjacent original corresponding records is then compared with the threshold corresponding to the preset continuous sampling condition. Comparison, This indicates the maximum allowable thrust displacement interval within the same continuous segment; it also checks if the subsequent thrust displacement value is greater than the previous thrust displacement value. The displacement difference must not exceed [a certain value]. Furthermore, adjacent records with increasing displacement are retained as consecutive corresponding records. After filtering, a sequence of consecutive corresponding records is obtained. ,in Indicates the number of consecutive corresponding records. , Indicates the first The propulsion displacement of a series of consecutive propulsion positions, Indicates the first Axial thrust at each successive advance position, Indicates the first Outlet contact response at each successive advancement position .

[0086] When determining the vicinity of the outlet, directly use the corresponding records in the sequence. and First, take a length of... The front stabilization window, This indicates the number of consecutive propulsion positions used to establish the contact response baseline; the median value of the outlet contact response within the pre-stabilization window is taken to obtain the baseline contact response. Then scan point by point along the direction of increasing displacement, until continuous Each propulsion position satisfies When the conditions are first met, the corresponding thrust displacement is determined as the starting point of the range near the outlet, whereby... Indicates the length of the contact response change confirmation window. This indicates the threshold for contact response change. Continue scanning along the direction of increasing propulsion displacement; when continuous... The advancement position was satisfied again. When entering this state, the advance displacement corresponding to the previous advance position is determined as the end point of the range near the outlet. The advance displacement interval defined by the starting advance displacement and the ending advance displacement is the range near the outlet, which represents the advance displacement interval where the leading edge of the soil sample approaches the outlet and the contact response at the outlet begins to change.

[0087] Extract all records whose propulsion displacement is within the vicinity of the outlet from the continuous corresponding record sequence to form the critical segment corresponding record sequence. ,in This indicates the number of records corresponding to the critical section. Then according to The record sequences corresponding to the critical segments are rearranged in ascending order to form the exit critical response sequence. This is to ensure consistency with the fixed input specifications of the encoding structure. In this embodiment, the total number of propulsion positions in the exit critical response sequence is... The critical response sequence of the outlet pipe is organized as follows: ,in , For the first The propulsion displacement at each propulsion position, For the first Axial thrust at each propulsion position, For the first The outlet contact response at each propulsion position, therefore the input dimension of a single outlet critical response sequence is... When the number of records corresponding to a critical segment is greater than 32, they are extracted at equal intervals according to the increasing order of the advance displacement; when the number of records corresponding to a critical segment is less than 32, linear interpolation is performed based on adjacent advance displacement positions to supplement the number of advance positions to 32. After this processing, Each of them All correspond to the same propulsion position, propulsion displacement, axial thrust, and outlet contact response, and all Arranged in ascending order of displacement, they form the direct input of the coding structure.

[0088] Step S2: Construct candidate latent variable sequences

[0089] Existing critical response algorithms for pipe exit often model directly based on natural sampling time sequences or directly determine thresholds based on fluctuations in the original response. This easily leads to the misinterpretation of local disturbances at different advancement positions as similar changes, causing the lateral support changes with real mechanistic significance during the critical pipe exit stage to be masked by axial load redistribution. This invention establishes an encoding sequence around the spatial transfer process of the soil core approaching the pipe exit, passing through the pipe exit, and leaving the continuous pipe wall support range. It preserves the competitive relationship between the two types of changes in a unified latent variable expression, completing the scenario-based transformation from both the input organization method and the latent expression method. This enables the lateral constraint transfer process near the pipe exit to obtain a more stable intermediate representation, establishing a directly inheritable latent variable foundation for subsequent determination of constraint interruption intervals.

[0090] Specifically, the critical response sequence at the pipe exit is divided into several consecutive segments according to the advance displacement order, and the advance displacement, axial thrust, and outlet contact response in each segment are arranged in a corresponding order. A recursive relationship is established between the starting advance position of each segment and the ending advance position of the previous segment, and the recursive state of the starting advance position of the current segment is determined based on the recursive state of the ending advance position of the previous segment. Each segment is sequentially input into the encoding structure for segment-by-segment encoding, and after each segment is encoded, the corresponding latent variables to be assigned are connected according to the advance displacement order. A candidate latent variable sequence is formed based on the latent variables to be assigned after connecting each segment. For example... Figure 3 As shown, step S2 specifically includes:

[0091] The critical response sequence of the outlet pipe is denoted as... , . , Indicates the first The propulsion displacement at each propulsion position, Indicates the first Axial thrust at each propulsion position, Indicates the first The outlet contact response at each propulsion position, therefore the input dimension of a single outlet critical response sequence is... The three-dimensional combination result corresponding to each propulsion position is defined as a recursive input unit, denoted as . The position input sequence is composed of all recursive input units arranged in ascending order of propulsion displacement. The displacement recursion datum is directly determined by the sequential relationship of the advancing displacements in the position input sequence, i.e., according to... Establish a recursive order based on the increasing order of the elements, so that the first element... The advancement position and the first The advancement positions form a recursive relationship, thereby anchoring the coding process to the positional transfer of the soil sample relative to the outlet, rather than to the natural sampling moment.

[0092] To maintain state continuity during long sequence encoding, the position input sequence The segments are divided into several consecutive segments according to the order of advancement displacement, and the number of segments is denoted as . . No. Duan Jiwei , Indicates the first Segment start index, Indicates the first Segment termination index, Adjacent segments satisfy Therefore, all segments are kept contiguous at the index. When the first segment enters the coding structure from its starting position, the recursive state of the previous position in the first recursive coding layer is initialized as a 32-dimensional zero vector, and the recursive state of the previous position in the second recursive coding layer is initialized as a 64-dimensional zero vector. After the segment is encoded, the first segment will be... The 32-dimensional and 64-dimensional recursive states formed at the end of the segment are respectively passed to the first... The starting position of the segment, as the first The recursive state of the preceding thrust position at the starting thrust position of a segment ensures that segmented encoding does not break the state update chain in the thrust displacement direction. Within each segment, the correspondence between thrust displacement, axial thrust, and outlet contact response is maintained without introducing new sorting rules.

[0093] The first recursive coding layer of the coding structure has 32 neurons. The first recursive coding layer is in the... Each advancement position receives the current recursive input unit. Given the 32-dimensional recursive state at the previous advancement position, a fully connected computation is performed on the current 3D input and the 32-dimensional recursive state at the previous advancement position, and the result is obtained through a ReLU nonlinear unit. The 32-dimensional primary coding result of each advancement position is denoted as... The primary coding sequence is composed of the primary coding results corresponding to all advancement positions. The dimension of the primary coding sequence is The first recursive coding layer only processes the current propulsion displacement, axial thrust, outlet contact response, and the 32-dimensional recursive state of the previous propulsion position, without introducing other data into the first recursive coding layer.

[0094] Primary coding sequence Continue with the second recursive coding layer of the input coding structure. The second recursive coding layer has 64 neurons. The second recursive coding layer is in the... Each propulsion position receives the 32-dimensional primary encoding result of the current propulsion position. Given the 64-dimensional recursive state of the previous advancement position, a fully connected computation is performed on the 32-dimensional primary encoding result and the 64-dimensional recursive state, and the result is obtained through a ReLU nonlinear unit. The 64-dimensional deep coding result of each propulsion position is denoted as... The deep coding sequence is composed of the deep coding results corresponding to all propulsion positions. The dimension of the deep coding sequence is The 64-dimensional recursive state of the previous advance position in the second recursive coding layer is directly derived from the output of the previous advance position within the same segment, or directly from the 64-dimensional recursive state passed from the last advance position of the previous segment. Therefore, the displacement recursion reference remains consistent across all advance positions.

[0095] Deep coding sequence Input compression mapping stage. The compression layer has 24 fully connected neurons, and the compression result of each 24-dimensional layer is denoted as... , Indicates the first A 24-dimensional compressed representation of each propulsion position. The compressed sequence consists of the 24-dimensional compressed representations of all propulsion positions. The dimension of the compressed sequence is 24-dimensional mean layer reception It outputs a 24-dimensional mean vector, denoted as... ; 24-dimensional variance layer receiver It outputs a 24-dimensional variance vector, denoted as... The 24-dimensional variance layer employs Softplus nonlinear units to ensure that each dimension of the 24-dimensional variance vector is positive. To obtain the latent variables to be assigned, a 24-dimensional auxiliary random vector is generated at each propulsion position. , Each dimension is independently taken from the standard normal distribution; then according to Generate 24-dimensional latent variables to be assigned. This indicates that corresponding dimensions are multiplied one by one. The sequence of latent variables to be assigned is composed of 24 dimensions of latent variables to be assigned from all advancement positions. , dimension In this stage, changes in lateral support and axial load are both retained within the same... Therefore, the two types of changes in the same advancement position still maintain a unified latent variable expression before entering the potential allocation framework.

[0096] A latent allocation framework consisting of a target latent space and a disturbance latent space is set after the 24-dimensional latent variables to be allocated. The target latent space is used to carry the latent variable components corresponding to lateral support changes, and the disturbance latent space is used to carry the latent variable components corresponding to axial load changes. Two existing allocation mapping layers are set within the latent allocation framework, each composed of 12 fully connected neurons, corresponding to the target latent space and the disturbance latent space respectively. Each 24-dimensional latent variable to be allocated... Simultaneously, two existing assignment mapping layers are fed in, enabling them to establish unassigned connections with the target potential space and the interference potential space at the same advancement position. However, at this stage, no assignment determination is performed based on the internal calculation results of the two existing assignment mapping layers. The 24-dimensional unassigned latent variables in the unassigned connection state are denoted as intermediate latent variables. All intermediate latent variables are organized according to the order of their advance displacements to obtain a sequence of candidate latent variables. Candidate latent variable sequence The dimension is Each Each of these corresponds to a 24-dimensional intermediate latent variable at the same propulsion position, maintaining its unassigned state before entering the target latent space and the interference latent space. When using segmented encoding, the intermediate latent variables formed by each segment are first arranged according to the propulsion displacement order within the segment, and then concatenated according to the order of each segment in the position input sequence, ultimately obtaining a complete sequence of candidate latent variables. Candidate latent variable sequence It is directly used as the input object when PP-DRE enters the allocation stage of the target potential space and the interference potential space.

[0097] S3. Construct a reference response group

[0098] Specifically, such as Figure 4 As shown, step S3 includes:

[0099] A reference response set is constructed using the stable response segment corresponding to continuous pipe wall support and the critical response segment corresponding to support transition. The reference response set is formed by extracting local segments from the exit critical response sequence. A length of [missing information] is used along the increasing direction of the thrust displacement. Sliding window Window-by-window scanning. For any adjacent advance positions within a window, the absolute values ​​of the differences in nozzle contact response and axial thrust are extracted as the criteria for variation within the window. The change is determined when all adjacent advance positions within the window satisfy the condition that the difference in nozzle contact response is no greater than a threshold. Furthermore, the difference between adjacent axial thrust values ​​is not greater than the threshold value. When the current window is defined as a stable response segment, the window is considered a continuous response segment; when the window contains consecutive... Each adjacent propulsion position satisfies the condition that the difference between adjacent outlet contact responses is greater than a threshold. When the corresponding advance displacement continues to increase, the current window is determined as the critical response segment. The windows determined as stable response segments are rearranged according to the advance displacement order and resampled into 32 advance positions, denoted as... The windows identified as critical response segments were rearranged according to their advance displacement order and resampled into 32 advance positions, denoted as... .Depend on and A reference response group is formed by combining the corresponding advance displacements. .

[0100] Step S4: Construct the target latent variable sequence

[0101] Calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group, assign the candidate latent variables to the target latent space and the interference latent space, and use the latent variables assigned to the target latent space as the target latent variable sequence; for example... Figure 4 As shown, step S4 includes:

[0102] The input object is the candidate latent variable sequence formed by S2. and the critical response sequence of the tube outlet . Indicates the first Each advancement position corresponds to a 24-dimensional intermediate latent variable, and each of the 24 components is a candidate latent variable component output by the encoding structure at the same advancement position. Indicates the first The three-dimensional response vector corresponding to each propulsion position, where Indicates the forward displacement. Indicates axial thrust. The input dimension of the outlet contact response and the outlet critical response sequence is: Candidate latent variable sequence It has been organized according to the order of advance displacement, therefore and Maintain the corresponding relationship.

[0103] Will and The input uses the same encoding structure as S2, and performs progressive position recursive encoding along the displacement recursive reference already determined in S2. The first recursive encoding layer receives the 3D input of each advancement position and the 32D recursive state of the previous advancement position, and outputs the 32D position encoding result; the second recursive encoding layer receives the 32D position encoding result of the current position and the 64D recursive state of the previous advancement position, and outputs the 64D deep encoding result; the 24D compression layer maps the 64D deep encoding result to a 24D compressed representation; the 24D mean layer and the 24D variance layer generate a 24D mean vector and a 24D variance vector; after sampling, a stable reference candidate latent variable sequence is obtained. and critical reference candidate latent variable sequence . , The 24 components represent the 24-dimensional reference intermediate latent variable components at the same propulsion position in both the stable and critical response segments. Because... , and All three sequences are generated by the same coding structure and are arranged in the order of advancement displacement, so they are directly comparable at each advancement position.

[0104] Calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group. The projection density ratio is calculated using the following formula:

[0105] ;

[0106] In the formula, Indicates the first The projection density ratio corresponding to each propulsion position; This represents the index of the advance position in the candidate latent variable sequence; Indicates the advance position index in the reference response group. The range of values ​​is to ; Indicates the first 8D projection results of each propulsion position; Indicates the first The 8-dimensional projection results corresponding to the critical reference candidate latent variables; Indicates the first The 8-dimensional projection results corresponding to the stable reference candidate latent variables; and These represent the squared Euclidean distances between the 8-dimensional projection results; This represents the squared distance scale parameter. It is a positive value, and its dimensions are consistent with the square of the Euclidean distance; This represents the natural index mapping, used to convert the squared distance into the corresponding density contribution value; Represents a positive zero constant. The value is dimensionless to avoid a denominator of zero. According to the calculation method described above, the numerator corresponds to the cumulative projected density of the candidate latent variable relative to the critical reference candidate latent variable distribution, and the denominator corresponds to the cumulative projected density of the candidate latent variable relative to the stable reference candidate latent variable distribution. Therefore... It directly constitutes the basis for potential allocation.

[0107] The target latent space and the interference latent space are each connected to a 12-neuron existing assignment mapping layer. The existing assignment mapping layer corresponding to the target latent space will... Mapping to a 12-dimensional target mapping result The existing allocation mapping layer corresponding to the interference potential space will... Mapped to 12-dimensional interference mapping result .Will With preset threshold Compare and in the advance position index window Perform consecutive comparisons within the scope. This indicates the length of the index window corresponding to the preset propulsion displacement range. and All internal propulsion positions meet the requirements At that time, Assigned to the target potential space; when and All internal propulsion positions meet the requirements At that time, Assigned to the potential interference space. When There exists simultaneously greater than and not greater than When the projected density ratio is , statistical Internal satisfaction The number of propulsion positions and satisfying The number of propulsion positions; satisfying When there are more propulsion positions, Assigned to the target potential space, satisfying When there are more propulsion positions, Assigned to potential interference space; when the two types have the same quantity, proceed according to the central advancement position. corresponding and The comparison results determine the allocation direction. Following the above rules, the potential allocation results for all advancement positions can be obtained.

[0108] Extract the propulsion position indices assigned to the target potential space from the potential allocation results, and arrange the indices assigned to the target potential space in ascending order of propulsion displacement. ,in This indicates the number of propulsion positions allocated to the target potential space, and For each ,Pick The target latent variable sequence is obtained. . This represents the 12-dimensional latent variables of the target extracted in the order of advancement displacement. Each of the 12 components corresponds to a latent variable component in the target's potential space. Target latent variable sequence. The dimension is and with index sequence Maintaining the corresponding order provides a direct input for restoring the continuous state variables of lateral constraints according to the order of advancing displacement.

[0109] like Figure 8 As shown, this invention utilizes a gridded grayscale field to represent the changes in lateral support strength corresponding to the advance displacement and burial depth. The color transitions from dark to light, reflecting the evolution of support capacity from strong to weak. Critical displacement locations are used to identify key sections where the soil core transitions from stable to weak constraints, and dense grids are used to refine local variation trends. This facilitates a visual representation of the lateral support attenuation distribution, providing a basis for identifying the critical section for tube removal, strength field analysis, and determining the encapsulation trigger interval.

[0110] Step S5: Generate the lateral constraint continuity determination quantity corresponding to the propulsion displacement.

[0111] Based on the positional state results between adjacent advancing positions, determine whether the direction of change is consistent, and based on whether the direction of change is consistent, determine whether the current position and the previous advancing position belong to the same continuous change process; when the direction of change meets the preset continuous recovery condition, merge the positional state result of the current position into the lateral constraint continuous state quantity corresponding to the previous advancing position for continuous accumulation; when the direction of change does not meet the preset continuous recovery condition, re-form the lateral constraint continuous state quantity according to the positional state result of the current position; calculate the state difference and state continuation amount based on the continuously accumulated lateral constraint continuous state quantity or the re-formed lateral constraint continuous state quantity, and generate the lateral constraint continuity determination quantity based on the state difference and state continuation amount. Specifically, as shown in Figure 5, step S5 includes:

[0112] The input object is a sequence of target latent variables formed by projection density ratio calculation and latent assignment. and the propulsion position index sequence corresponding to the target latent variable sequence . This indicates the number of propulsion positions allocated to the target potential space. Indicates the first The original propulsion position indices corresponding to each target latent variable, and satisfying the following conditions: . Indicates the first Each propulsion position corresponds to a 12-dimensional latent variable of the target, and each of the 12 components is a latent variable component corresponding to the target's latent space. Based on the propulsion position index sequence... The corresponding thrust displacement is retrieved from the critical response sequence of the tube exit to obtain the thrust displacement sequence. The propulsion displacement and 12-dimensional target latent variables at each propulsion position are expanded in the same order to form the state recovery input sequence. ,in State recovery input sequence The dimension is In the state recovery input sequence In the middle, the first The advancement position and the first Each propulsion position is linked to the previous and subsequent positions, and the propulsion displacement interval between adjacent propulsion positions is denoted as . , express and The difference, It is directly used to characterize the position span in the direction of propulsion displacement.

[0113] The state recovery layer uses a recursive approach to process the state recovery input sequence. The state recovery layer consists of a 16-neuron fully connected layer and a 1-neuron output layer. The 16-neuron fully connected layer uses ReLU nonlinear units to generate the positional state result; the 1-neuron output layer maps the positional state result to a single-point state value. For the... At the next propulsion position, the 16-dimensional state recovery result from the previous propulsion position is initialized as a 16-dimensional zero vector. The current 12-dimensional target latent variable is then concatenated with the 16-dimensional zero vector in a fixed order to form a 28-dimensional recursive input vector. This 28-dimensional recursive input vector enters a 16-neuron fully connected layer to obtain the [missing information - likely a specific neuron's input vector]. 16-dimensional position state results for each propulsion position Then... Input the output layer of neuron 1 to obtain the first neuron. 1D single-point state value at each propulsion position For any The 12-dimensional latent variables of the current advancement position 16-dimensional position state results compared to the previous advance position Assemble them in a fixed order to form the first... The 28-dimensional recursive input vector corresponds to the propulsion position. After the 28-dimensional recursive input vector enters the same 16-neuron fully connected layer, the result is obtained. 16-dimensional position state results for each propulsion position Then... Inputting the same 1-neuron output layer yields the 1st... 1D single-point state value at each propulsion position From all The position state results are sequenced according to the order of advance displacement. Position state result sequence The dimension is ; by all A sequence of single-point state values ​​is formed according to the order of advance displacement. Single-point state value sequence The dimension is .

[0114] Based on the sequence of location status results Calculate the direction and magnitude of change between adjacent advance positions. For any Calculate component by component and The 16 component differences are used, and the algebraic sum of the 16 component differences is denoted as the direction discrimination value. The sum of the absolute values ​​of the differences among the 16 components is recorded as the magnitude of change. .when When it is greater than zero, the first The propulsion position is relative to the first The direction of change in the propulsion position is determined to be a positive change; when When the value is less than zero, the direction of change is defined as negative; when... When the value is zero, the direction of change is determined to be zero-towards. Preset continuous recovery conditions include a consistent direction condition and an amplitude threshold. and displacement interval threshold When the first The direction of change of the propulsion position is related to the first The direction of change of each propulsion position is consistent, and Not greater than ,and Not greater than At that time, the first The position status result of the first propulsion position is determined as a continuously changing result; if the above conditions are not met, the position status result of the first propulsion position is determined as a continuously changing result. The positional state results at each propulsion position are determined to be discontinuous changes. Sequences are labeled with continuous changes. This indicates the above judgment result. Indicates the first The advancement position is a result of continuous change. Indicates the first Each advancement position is a result of discontinuous change, and it is specified that... .

[0115] Lateral constraint continuous state variables are composed of single-point state value sequences. Continuous change marker sequence and propulsion displacement interval Determined jointly. Regarding the first... One advancement position, directly to Determined as the number Lateral constraint continuous state variables at each propulsion position For any Calculate the first according to the following formula Lateral constraint continuous state variables at each propulsion position :

[0116] ;

[0117] In the formula, Indicates the first The lateral constraint continuous state variables for each propulsion position; Indicates the first A continuous change marker for each propulsion position, The value can be 0 or 1; Indicates the first The lateral constraint continuous state variables for each propulsion position; Indicates the first One-dimensional single-point state value at each propulsion position; Indicates the first One-dimensional single-point state value at each propulsion position; Indicates the first The advancement position and the first The propulsion displacement interval between each propulsion position; This represents the displacement interval threshold in the preset continuous recovery conditions. It is a positive value, and its dimensions are the same as those of the previous value. Consistent; This represents the position constraint coefficient corresponding to the propulsion displacement interval; Indicates the first The reconstruction selection factor when the advancement position does not belong to the continuously changing results. According to the above calculation method, when the... When the current propulsion position is a continuously changing result, the single-point state change at the current position is constrained according to the propulsion displacement interval and accumulated onto the lateral constraint continuous state quantity of the previous propulsion position; when the... When a propulsion position does not belong to the continuously changing result, the lateral constraint continuous state quantity is directly reconstructed based on the single-point state value of the current position. (From all) A sequence of lateral constraint continuous state variables is formed according to the order of propulsion displacement. Lateral constraint continuous state quantity sequence The dimension is .

[0118] Based on the lateral constraint continuous state quantity sequence Calculate the state difference and state duration. For the ... At each advancement position, initialize the state difference to 0 and the state continuation value to 0. For any... ,Will and The difference is determined as the first State difference at each propulsion position The state duration is denoted as... .when At that time, the first The advancement position and the first The propulsion displacement interval between each propulsion position The state duration of the current advancement position is obtained by adding it to the state duration of the previous advancement position. ;when At that time, the state of the current advancement position will be extended by a certain amount. Reset the values ​​to 0. Combine the state difference and state continuation value corresponding to each advancement position in a fixed order to form the judgment input result. From all The input sequence is determined according to the order of the advancing displacement. Determine the input sequence The dimension is .

[0119] The decision layer consists of a single fully connected output neuron. The decision layer receives 2D decision inputs at each progressively advancing position. and for each Output 1D lateral constraint continuity decision quantity From all A sequence of lateral constraint continuity determination quantities is formed according to the order of advance displacement. Lateral constraint continuity decision sequence The dimension is At the same time, maintain and The correspondence ensures that the lateral constraint continuity judgment quantity for each propulsion position corresponds to the actual propulsion displacement position, forming a sequence of judgment results arranged in the order of propulsion displacement. .

[0120] like Figure 9 In the 3D surface image shown, the raised areas of the shaded 3D surface of the disturbed energy field correspond to the stages of weakened lateral constraint and increased local damage accumulation in the soil core. The vertical projection of the critical displacement is used to identify the location of rapid risk increase. The image elevates the planar intensity change to a three-dimensional energy evolution expression, providing an intuitive basis for identifying peak areas, locking high-risk segments, and extracting threshold boundaries.

[0121] Step S6: Determine the corresponding propulsion displacement interval as the constraint interruption interval.

[0122] The lateral constraint continuity determination quantity is calculated by determining the difference between adjacent positions according to the advance displacement sequence, and a continuous attenuation gradient is calculated based on the difference between adjacent positions and the corresponding advance displacement interval to form a gradient sequence. Based on the range of advance displacements that continuously satisfy the preset gradient conditions in the gradient sequence, the span of the advance displacements that continuously satisfy the preset gradient conditions is calculated to form a duration. The continuous attenuation gradient and duration are compared with preset interruption conditions. If both the continuous attenuation gradient and duration simultaneously satisfy the preset interruption conditions, the corresponding advance displacement start point and end point are determined. An advance displacement interval is formed based on the advance displacement start point and end point, and this advance displacement interval is determined as the constraint interruption interval. Specifically, as follows... Figure 6 As shown, step S6 includes:

[0123] The continuity determination quantity of lateral constraint is determined by the continuous decay gradient and duration. When the continuous decay gradient and duration meet the preset interruption conditions, the corresponding propulsion displacement interval is determined as the constraint interruption interval.

[0124] The input object is the decision result sequence formed by S5. . This indicates the number of propulsion positions involved in the interruption determination. Indicates the first The index of each propulsion position in the original propulsion position sequence. Indicates the first The propulsion displacement corresponding to each propulsion position Indicates and The corresponding lateral constraint continuity determination quantity, and satisfies According to the increasing order of propulsion displacement, for any... Calculate the difference in lateral constraint continuity determination between the current advancement position and the previous advancement position, denoted as ,in Depend on minus We obtain; simultaneously calculate the propulsion displacement interval between the current propulsion position and the previous propulsion position, denoted as . ,in Depend on minus Obtain. Only when Only when the gradient is greater than zero is a continuously decaying gradient calculated for the current adjacent advance positions. Divide by Get the first The continuously decaying gradient of a pair of adjacent advance positions, denoted as . This represents the rate of change of the difference in lateral constraint continuity determination value between adjacent advance positions relative to the advance displacement interval. (From all) A gradient sequence is formed according to the order of propulsion displacement. .

[0125] For gradient sequences Perform continuous segment scanning to determine the range of propulsion displacements that continuously satisfy the preset gradient conditions. Let the gradient threshold be... .when At that time, the judgment of the first A pair of adjacent advancement positions satisfies the preset gradient condition; when At that time, the judgment of the first A pair of adjacent advance positions does not satisfy the preset gradient condition. Adjacent advance position pairs with consecutive indices in the gradient sequence that continuously satisfy the preset gradient condition are merged into candidate decay segments. The candidate attenuation segments are denoted as , Indicates the candidate decay segment index. Indicates the first The starting gradient index of each candidate decay segment. Indicates the first The termination gradient index of each candidate decay segment, and satisfying Due to gradient indexing Corresponding to the advance position With the position of advancement The changes between them, therefore the first The candidate initial propulsion displacement of each candidate attenuation segment is taken as... , No. The candidate termination propulsion displacement of each candidate attenuation segment is taken as Subtracting the candidate initial propulsion displacement from the candidate termination propulsion displacement yields the first... The duration of each candidate decay segment is denoted as . . This represents the span of propulsion displacement that continuously satisfies the preset gradient conditions.

[0126] The continuous decay gradient and duration corresponding to each candidate decay segment are compared with the preset interruption condition. The preset interruption condition consists of a gradient condition and a duration condition. The gradient condition requires the... All gradient values ​​within each candidate decay segment Continuous satisfaction ; Length requirement Duration of each candidate decay segment Not less than the length threshold , This represents the minimum propulsion displacement span required for the constraint interruption determination. When the... When the nth candidate decay segment simultaneously satisfies both the gradient condition and the length condition, the nth... The candidate attenuation segments were identified as valid interrupt segments, and Determined as the starting point of the propulsion displacement, This is determined as the endpoint of the advance displacement. If the advance displacement interval between two adjacent interrupted valid segments is not greater than the inter-segment merging threshold... If two adjacent valid interrupted segments are merged, the starting point of the merged advance displacement is taken from the starting point of the previous valid interrupted segment, and the ending point of the merged advance displacement is taken from the ending point of the next valid interrupted segment. Inter-segment merging threshold. This indicates the maximum inter-segment interval that can be considered as the interruption process of the same constraint.

[0127] After comparing and merging all candidate attenuation segments as necessary, one or more thrust displacement intervals are obtained. All thrust displacement intervals are arranged in ascending order of their thrust displacement starting points, denoted as... . Indicates the number of constraint interruption intervals. Indicates the first The starting point of the propulsion displacement in a constrained interruption interval. Indicates the first The endpoint of the advance displacement of each constrained interruption interval, and satisfying Each Each interruption interval is directly formed by the starting and ending points of the corresponding effective interruption segment's advance displacement, and is directly defined as the constraint interruption interval. This constraint interruption interval maintains a correspondence with the advance displacement and can be directly used to limit the extension range and trigger position of the lead screw advance mechanism.

[0128] Figure 10 Multiple contour lines are shown to illustrate the continuous variation of the instability index in the space of thrust displacement and burial depth. This visually demonstrates the contour distribution of the lateral constraint continuity determination quantity (R) under different thrust displacements and depths, providing an intuitive visual basis for determining the starting boundary of the constraint interruption interval, the encapsulation trigger position, and the safe operation window. In this embodiment, R=1.0 is used as the critical threshold, representing the boundary condition for the lateral support to transition from a safe state to an unstable state; the region where R<1.0 corresponds to a relatively intact stable region of lateral constraint, while the region where R>1.0 corresponds to a high-risk region of lateral constraint loss.

[0129] Step S7: Generate the ejection control command for the lead screw drive mechanism and the encapsulation trigger command for the short-range support component at the outlet.

[0130] The initial propulsion displacement, final propulsion displacement, and interval length are extracted based on the constraint interruption interval to form the ejection control interval. The ejection range of the screw propulsion mechanism within the initial to final propulsion displacement is determined based on the ejection control interval, and this range is converted into ejection control commands for the screw propulsion mechanism. The sealing trigger position is determined based on the final propulsion displacement and interval length, and is compared with preset sealing conditions. If the comparison conditions are met, a sealing trigger command is generated. The constraint interruption interval, the ejection control command of the screw propulsion mechanism, and the sealing trigger command are output accordingly to form the lateral constraint continuity determination result for the detachment critical section. Specifically, such as... Figure 7 As shown, step S7 includes:

[0131] The input object is the constraint interruption interval formed by S6. And the encapsulation allows for initial displacement Encapsulation allows for termination displacement Minimum encapsulation length threshold and trigger ratio coefficient . Indicates the number of constraint interruption intervals. Indicates the first The initial advance displacement of each constrained interruption interval, Indicates the first The termination propulsion displacement of each constrained interruption interval, and satisfying . and Used to limit the propulsion displacement window that the encapsulation trigger position is allowed to fall into. Used to limit the minimum interval length that allows triggering of the encapsulation. A scaling factor with a value between 0 and 1, used to determine the forward movement of the encapsulation trigger position relative to the termination propulsion displacement based on the interval length.

[0132] For any First, start with the first Extracting the initial propulsion displacement from each constraint interruption interval and termination of propulsion displacement Then minus Obtain the interval length . Indicates the first The span of the propulsion displacement within each constrained interruption interval. , and The control range is generated by combining fixed fields in a specific order. Exit the control zone This is a 3D control vector, where the first dimension is the initial propulsion displacement, the second dimension is the final propulsion displacement, and the third dimension is the interval length. The control interval is then decoupled. Only the data actually generated in this step is retained; other fields not involved in the calculation are not included.

[0133] According to the launch control zone Determine the extension range of the lead screw drive mechanism. Since the drive process of the lead screw drive mechanism is controlled by the drive displacement, therefore the first... The launch range corresponding to each launch control interval is directly taken as the initial propulsion displacement. To terminate the propulsion displacement The displacement range. The initial propulsion displacement. Termination of propulsion displacement Interval length and launch enable flag Combined into lead screw drive mechanism ejection control commands Issue control commands. This is a 4-dimensional control vector. The first dimension is the initial thrust displacement control value, the second dimension is the final thrust displacement control value, the third dimension is the interval length control value, and the fourth dimension is the push-out enable flag. For the constraint interruption interval already determined by S6, a unified command is used. This indicates that the screw drive mechanism has reached the required displacement. The push-out control begins at that time, and the push-out displacement reaches [a certain value]. The control will be released when the time is up.

[0134] Based on the termination of propulsion displacement and interval length Determine the encapsulation trigger position. First, set the trigger ratio coefficient. With interval length Multiply to obtain the trigger shift amount. . This indicates the forward movement distance of the encapsulation trigger position relative to the termination propulsion displacement. Then, the termination propulsion displacement... Subtract the trigger shift amount Obtain the encapsulation trigger position .because The value of is between 0 and 1, therefore Always in and Between. Encapsulate the trigger position. With package allowable initial displacement Encapsulation allows for termination displacement Compare and simultaneously set the interval length With minimum package length threshold Comparison. When , and At that time, enable the package flag. If any of the above conditions are not met, enable the encapsulation flag. Then encapsulate the trigger position. Interval length and package enable flag Combined into encapsulated trigger instructions Encapsulate trigger instructions It is a 3D control vector, with the first dimension being the encapsulation trigger position, the second dimension being the interval length, and the third dimension being the encapsulation enable flag.

[0135] The first The initial propulsion displacement in each constraint interruption interval and termination of propulsion displacement With the same index Corresponding ejection control command and encapsulated trigger instructions Establish a correspondence based on the order of advancement displacement. For all After repeating the above process, each constraint interruption interval corresponds to a set of explicit control data, in which the exit control command is executed. Directly limit the extension range of the lead screw drive mechanism and encapsulate the trigger command. The encapsulation trigger position and encapsulation enable state are directly defined, thereby forming a lateral constraint continuity determination result for the de-managed critical segment that corresponds one-to-one with each constraint interruption interval.

[0136] The present invention also provides a sampling device for undisturbed soil in deep overburden layers, the device comprising a sequence forming unit, an encoding unit, an allocation unit, a state recovery unit, an interval determination unit, and a control unit.

[0137] The sequence forming unit is used to form the critical response sequence of the tube outlet based on the propulsion displacement, axial thrust, and outlet contact response.

[0138] The encoding unit is used to encode the critical response sequence of the tube exit position by position, using the advance displacement as a recursive reference, to form a candidate variable sequence. The encoding unit pre-defines the target potential space for bearing changes in lateral support and the disturbance potential space for bearing changes in axial load.

[0139] The allocation unit is used to construct a reference response group with stable response segments and critical response segments, calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group, allocate the candidate latent variables to the target latent space or the interference latent space according to the calculation results, and extract the latent variables allocated to the target latent space to form the target latent variable sequence.

[0140] The state recovery unit is used to recover the lateral constraint continuous state quantities along the advance displacement based on the target latent variable sequence, and to generate the lateral constraint continuity determination quantity.

[0141] The interval determination unit is used to determine the continuous decay gradient and duration of the lateral constraint continuity determination quantity, and to determine the constraint interruption interval when the preset interruption condition is met.

[0142] The control unit is used to generate the ejection control command and encapsulation trigger command of the lead screw propulsion mechanism according to the constraint interruption interval.

[0143] The above-mentioned units work together to realize the deep overburden undisturbed soil sampling method described in this invention.

Claims

1. A method for sampling undisturbed soil from deep overburden layers, applied to a sampling device comprising a semi-compound pipe, a screw-driven mechanism, and a short-range support component at the pipe outlet, characterized in that, The method includes: S1. Obtain the propulsion displacement, axial thrust, and outlet contact response of the sampling device during the sampling process, and divide the outlet critical response sequence according to the propulsion displacement; S2. Based on the set coding structure, the position of the critical response sequence of the tube exit is encoded according to the propulsion displacement to construct a candidate latent variable sequence; wherein, the coding structure includes the target potential space for bearing lateral support changes and the interference potential space for bearing axial load changes; S3. Construct a reference response group using the stable response segment corresponding to continuous pipe wall support and the critical response segment corresponding to support transition; S4. Calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group, assign the candidate latent variables to the target latent space and the interference latent space, and use the latent variables assigned to the target latent space as the target latent variable sequence. S5. Recover the lateral constraint continuity state quantity along the advance displacement based on the target latent variable sequence, and generate the lateral constraint continuity determination quantity corresponding to the advance displacement. S6. Calculate the continuous attenuation gradient and duration of the lateral constraint continuity determination quantity. When the continuous attenuation gradient and duration meet the preset interruption conditions, determine the corresponding propulsion displacement interval as the constraint interruption interval. S7. Generate the ejection control command for the screw drive mechanism and the encapsulation trigger command for the short-range support component at the outlet pipe based on the determined constraint interruption interval.

2. The method for sampling undisturbed soil from a deep overburden layer according to claim 1, characterized in that, Step S1 includes: Align the propulsion displacement, axial thrust, and outlet contact response time to form the original corresponding set; Based on the displacement difference between adjacent advance displacements, a continuous screening is performed, and the corresponding groups that meet the preset continuous sampling conditions are taken as continuous corresponding groups. The range near the outlet is determined based on the changes in the propulsion displacement position and the contact response at the outlet. The continuous corresponding groups within the range near the outlet are taken as the critical segment corresponding groups. Arrange the propulsion displacement, axial thrust, and outlet contact response in the corresponding group of the critical section in ascending order of propulsion displacement to form the outlet critical response sequence.

3. The method for sampling undisturbed soil from a deep overburden layer according to claim 1, characterized in that, The encoding structure includes a first recursive encoding layer, a second recursive encoding layer, a compression layer, a mean layer, a variance layer, and an allocation framework; step S2, which involves encoding the position of the exit critical response sequence according to the propulsion displacement, specifically includes: The propulsion displacement, axial thrust, and outlet contact response corresponding to each propulsion position are combined into a recursive input unit to form a position input sequence. According to the displacement recursion benchmark, the position input sequence is sequentially input into the first recursive coding layer and the second recursive coding layer to form the coding result; The compression layer performs compression mapping on the encoded results, and the mean and variance layers calculate the mean and variance to form latent variables to be assigned. The latent variables to be assigned are input into the assignment framework consisting of the target latent space and the disturbance latent space to form intermediate latent variables; The intermediate latent variables are arranged in the order of their advance displacement to form a sequence of candidate latent variables.

4. The method for sampling undisturbed soil from a deep overburden layer according to claim 1, characterized in that, Step S4, calculating the projection density ratio of the candidate latent variable sequence relative to the reference response group, specifically includes: The reference response group is input into the coding structure, and the similarity between the candidate latent variable sequence and the critical response segment and the stable response segment is calculated respectively. The ratio of the similarity is used as the projection density ratio. Candidate latent variables are assigned to the target latent space and the interference latent space based on the projection density ratio and a preset threshold.

5. The method for sampling undisturbed soil from a deep overburden layer according to claim 4, characterized in that, The method for calculating the projection density ratio is as follows: ； Indicates the first The projection density ratio corresponding to each propulsion position; This represents the total number of propulsion positions in the critical response sequence of the outlet tube; This represents the index of the advance position in the candidate latent variable sequence; Indicates the advance position index in the reference response group; Indicates the first 8D projection results of each propulsion position; Indicates the first The 8-dimensional projection results corresponding to the critical reference candidate latent variables; Indicates the first The 8-dimensional projection results corresponding to the stable reference candidate latent variables; and These represent the squared Euclidean distances between the 8-dimensional projection results; This represents the squared distance scale parameter. This represents a positive zero constant.

6. The method for sampling undisturbed soil from a deep overburden layer according to claim 4, characterized in that, Step S4, which involves assigning candidate latent variables to the target latent space and the interference latent space, includes: If the projection density ratio of the current advancement position is greater than a preset threshold, and all advancement positions within the preset advancement displacement range satisfy the condition that the projection density ratio is greater than the threshold, the candidate latent variable corresponding to the current advancement position is assigned to the target latent space. If the projection density ratio of the current advancing position is not greater than a preset threshold, and all advancing positions within the preset advancing displacement range satisfy the condition that the projection density ratio is not greater than the threshold, the candidate latent variable corresponding to the current advancing position is assigned to the interference potential space. When there are projection density ratios that are both greater than and less than the preset threshold within the preset propulsion displacement range, the candidate latent variable allocation direction is determined based on the majority voting principle.

7. The method for sampling undisturbed soil from a deep overburden layer according to claim 1, characterized in that, Step S5 includes: The target latent variable sequence is expanded according to the advance displacement order, and the correlation between adjacent advance positions is established to form the state recovery input sequence; The position state result is obtained by performing a progressive displacement recursive calculation on the state recovery input sequence. The position state result that meets the preset continuous recovery condition is determined as the continuously changing result. The continuously changing result is cumulatively mapped along the progressive displacement to form the lateral constraint continuous state quantity. The state difference and state continuation are calculated based on the lateral constraint continuous state quantities, and then combined to generate the lateral constraint continuity determination quantity.

8. A method for sampling undisturbed soil from a deep overburden layer according to claim 1, characterized in that, Step S6, determining the constraint interruption interval, specifically includes: The ratio of the difference between adjacent lateral constraint continuity determination quantities to the advance displacement interval is calculated according to the advance displacement sequence to form a continuous decay gradient sequence; Extract the range of continuous propulsion displacement that meets the preset gradient conditions, and calculate the displacement span of the corresponding range as the duration. When the continuous decay gradient is not greater than the set gradient threshold and the duration is not less than the set length threshold, the corresponding propulsion displacement start and end points are determined to form a constraint interruption interval.

9. A method for sampling undisturbed soil from a deep overburden layer according to claim 1, characterized in that, Step S7, which generates the ejection control instruction and the encapsulation trigger instruction, specifically includes: Extract the initial thrust displacement, the final thrust displacement, and the interval length from the constrained interruption interval to form the push-out control interval; The ejection range of the lead screw propulsion mechanism is determined based on the ejection control range, and an ejection control command containing the initial propulsion displacement, the final propulsion displacement, the range length, and the ejection enable flag is generated. The encapsulation trigger position is determined based on the termination propulsion displacement and interval length, and compared with the preset encapsulation conditions to generate an encapsulation trigger command containing the encapsulation trigger position, interval length, and encapsulation enable flag.

10. A sampling device for undisturbed soil in deep overburden layers, used to implement the sampling method for undisturbed soil in deep overburden layers as described in any one of claims 1-9, characterized in that, include: A sequence forming unit is used to form a critical response sequence for tube exit based on propulsion displacement, axial thrust, and tube exit contact response. The encoding unit is used to encode the critical response sequence of the tube outlet position by position based on the advance displacement to form a candidate latent variable sequence; the encoding unit presets a target potential space for bearing lateral support changes and an interference potential space for bearing axial load changes; The allocation unit is used to construct a reference response group with stable response segments and critical response segments, calculate the projection density ratio of the candidate latent variable sequence relative to the reference response group, allocate the candidate latent variables to the target latent space or the interference latent space according to the calculation results, and extract the latent variables allocated to the target latent space to form the target latent variable sequence. The state recovery unit is used to recover the lateral constraint continuous state quantity along the propulsion displacement according to the target latent variable sequence, and generate the lateral constraint continuity determination quantity. The interval determination unit is used to determine the continuous decay gradient and duration of the lateral constraint continuity determination quantity, and to determine the constraint interruption interval when the preset interruption condition is met. The control unit is used to generate the ejection control command and the encapsulation trigger command of the lead screw propulsion mechanism according to the constraint interruption interval.

Images