Method and system for determining the risk of transporting undisturbed samples based on shock attenuation theory

Abstract

This invention relates to the field of undisturbed sample transportation. To achieve hazardous transportation assessment of undisturbed samples, this application provides a method and system for assessing the transportation risks of undisturbed samples based on vibration reduction theory. The method involves acquiring the vibration of the transport container, the force and displacement of the vibration-damping support, and the constraint changes of the encapsulated soil sample cylinder to form a transportation response segment sequence. This sequence is then input into MTS-JEPA, where joint embedding mapping is performed on short-term impact and long-term cumulative scales to obtain a transportation potential state sequence. An orthogonal subspace state codeword is constructed using AMP in a soft codebook to form an orthogonal subspace codebook. The transportation potential state sequences are then categorized to form hazardous state groups and hazardous state change sequences. Finally, a hazard assessment result is generated under the structural instability threshold. This method achieves hazardous transportation assessment of undisturbed samples with high accuracy.

Description

Technical Field

[0001] This invention relates to the field of undisturbed sample transportation, specifically a method and system for risk assessment of undisturbed sample transportation based on shock absorption theory. Background Technology

[0002] When undisturbed soil samples are used for laboratory tests such as strength, deformation, and seepage, it is necessary to maintain the on-site structure, moisture state, porosity, and boundary constraints as much as possible. Domestic and international geotechnical engineering data generally indicate that the transportation stage is a crucial factor affecting sample disturbance throughout the entire process from sampling, packaging, and handling to laboratory testing. Especially under long-distance road transport conditions, continuous vibration, accidental impacts, changes in orientation, and fluctuations in the external environment can all alter the in-situ state of the sample. Therefore, specialized technical designs focusing on the protection of undisturbed samples during transportation, the identification of transportation disturbances, and the assessment of transportation risks have become a practical requirement in engineering surveying and testing.

[0003] Existing technologies regarding undisturbed sample transportation have developed into two main approaches: device protection and condition assessment. Regarding device protection, CN208217327U discloses a soil sample transport shock-absorbing box, which incorporates springs and support plates within the box to cushion and support the soil sample container; CN112660553A discloses an undisturbed soil sample transport device, which incorporates a triaxial stabilization system between the box and the soil sample container to reduce transport disturbances under complex road conditions; CN110834793A discloses an undisturbed soil sample transport box, which incorporates a honeycomb shock-absorbing plate between the outer shell and inner box, and works with a placement rack and sampler to complete the sealed transport; CN207374913U discloses a lightweight undisturbed soil sample transport box, which uses multi-layered sample chambers and chamber racks for layered storage; CN215099424U discloses an undisturbed soil sample transport box with good seismic resistance, which incorporates a sealed box, airtight rubber, and cushioning air pads inside the protective box to enhance transport protection capabilities. In terms of literature and standards, relevant manuals from the U.S. Bureau of Reclamation, ASTM D4220, and ISSMGE documents all regard vibration protection, shock protection, and maintaining a stable posture as basic requirements for the preservation and transportation of undisturbed samples. Regarding the evaluation of sample disturbance, international research often measures the degree of sample disturbance through comparisons of field and laboratory tests, shear wave velocities, and consolidation-compression property analysis. Meanwhile, in the field of continuous monitoring signal processing, international papers have employed methods such as masked autoencoding, self-supervised channel modeling, multi-resolution joint embedding prediction, vector quantization codebooks, and subspace clustering to encode, cluster, or identify the state of multivariate time series, providing a data processing foundation for signal analysis during complex transportation processes.

[0004] Based on the aforementioned patents and literature, it can be seen that existing original sample transportation solutions focus more on container cushioning, attitude maintenance, sealed storage, or post-transport disturbance evaluation, and lack a comprehensive connection between device protection and risk assessment for the same transportation process. Summary of the Invention

[0005] To achieve the determination of hazardous transportation of undisturbed samples, this invention provides a method and system for determining the transportation risk of undisturbed samples based on vibration reduction theory.

[0006] The technical solution adopted by the present invention to solve the above problems is:

[0007] Methods for assessing the transportation risks of undisturbed samples based on vibration reduction theory include:

[0008] Step 1: Obtain the vibration of the transport container, the force and displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube, and form a transport response segment sequence according to the continuous transport time period;

[0009] Step 2: Input the transport response fragment sequence into MTS-JEPA, perform joint embedding mapping on the short-term impact scale and the long-term cumulative scale to obtain the transport potential state sequence, and input the transport potential state sequence into the soft codebook.

[0010] Step 3: In the soft codebook, AMP is used to construct each state codeword into an orthogonal subspace state codeword to form an orthogonal subspace codebook. The orthogonal subspace state codewords correspond to similar transmission patterns caused by the vibration of the transport box, the force displacement of the shock-absorbing support, and the change of the contact boundary of the sealed soil sample tube.

[0011] Step 4: Classify the transportation potential state sequence based on the orthogonal subspace codebook to obtain the transportation state number sequence, and determine the transportation potential states that are continuously classified into the same orthogonal subspace state codeword as the same dangerous state group.

[0012] Step 5: Track the continuous distribution of hazardous status groups based on the transport status number sequence to form a hazardous status change sequence;

[0013] Step 6: Determine whether the continuous occurrence condition is met based on the sequence of dangerous state changes below the structural instability threshold, and form a dangerous judgment mark;

[0014] Step 7: When the hazard assessment indicator indicates that the continuous occurrence condition is met, output the hazard transportation assessment result of the original sample; when the hazard assessment indicator indicates that the continuous occurrence condition is not met, output the non-hazard transportation assessment result of the original sample.

[0015] Furthermore, step 1 specifically includes:

[0016] At the same sampling time, the vibration of the transport box, the force displacement of the shock-absorbing support, and the constraint change of the sealed soil sample tube are obtained respectively, and combined according to the positional order of the transport box, the shock-absorbing support, and the sealed soil sample tube in the transport transmission chain to form a transport state vector corresponding to the sampling time.

[0017] The transportation state vectors corresponding to consecutive sampling times are arranged in chronological order to form a temporal state chain within a continuous transportation time period;

[0018] A continuous transport state vector is extracted from the temporal state chain with a preset segment length. Each continuous transport state vector is spliced ​​together in chronological order, while keeping the transport chain order within each continuous transport state vector unchanged, to form a candidate transport segment.

[0019] Based on the complete coverage relationship of continuous sampling times in the candidate transportation segments and the synchronous correspondence between the vibration of the transportation box, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample tube, the candidate transportation segments are retained to form a transportation response segment sequence.

[0020] Furthermore, MTS-JEPA includes:

[0021] The input mapping layer is used to map the original input features to the basic temporal features through one-dimensional convolution;

[0022] The local temporal extraction layer is used to extract local state features by concatenating one-dimensional convolution and ReLU;

[0023] A dual-scale state generation layer is used to perform short-term impulsive scale convolution and long-term cumulative scale dilation convolution on local state features to generate short-term state features and long-term state features respectively.

[0024] The transportation latent state generation layer is used to concatenate short-term state features and long-term state features at the same time location and then pass them through a fully connected mapping to obtain a transportation latent state sequence.

[0025] Furthermore, step 3 includes:

[0026] Each state codeword in the soft codebook is determined as a state codeword to be constructed, and each state codeword to be constructed corresponds to a potential transportation state distribution range to be carried.

[0027] Based on the temporal adjacency relationship, amplitude drift relationship, and linkage relationship of transportation box vibration, shock absorber support displacement and encapsulated soil sample tube constraint changes in the continuous transportation time period, similar transmission patterns caused by transportation box vibration, shock absorber support displacement and encapsulated soil sample tube contact boundary changes are extracted to form subspace construction constraints corresponding to each codeword to be constructed.

[0028] Under the action of AMP, orthogonal basis construction is performed on each state codeword to be constructed according to the subspace construction constraints. Each state codeword to be constructed is updated from a single vector state codeword to a candidate orthogonal subspace state codeword composed of multiple mutually orthogonal basis vectors, so that each candidate orthogonal subspace state codeword carries the potential transport states with different amplitudes and different local drifts under the same contact boundary change mechanism.

[0029] The potential state sequence of transportation is projected onto the state codewords of each candidate orthogonal subspace, and the projection coefficients and subspace residuals corresponding to each candidate orthogonal subspace state codeword are calculated to obtain the basis for classifying the state codewords that distinguish between similar and dissimilar transmission patterns.

[0030] Based on the state codeword attribution criteria, the orthogonal basis direction and subspace coverage of each candidate orthogonal subspace state codeword are adjusted, and the candidate orthogonal subspace state codewords that meet the preset attribution conditions are determined as orthogonal subspace state codewords, so that the transport potential states belonging to the same orthogonal subspace state codeword correspond to the same contact boundary change mechanism.

[0031] Write the state codewords of each orthogonal subspace into the soft codebook to form the orthogonal subspace codebook.

[0032] Furthermore, in step 3, the preset attribution condition is that for the same group of potential transport states within the same local continuous window, the attribution criteria of the state codeword corresponding to a certain candidate orthogonal subspace state codeword continuously obtains the maximum value, and the group of potential transport states satisfies the time adjacency relationship, amplitude drift relationship and linkage change relationship under the same contact boundary change mechanism.

[0033] Furthermore, step 4 includes:

[0034] The transportation potential state sequence is input into the orthogonal subspace codebook in chronological order, and each transportation potential state in the transportation potential state sequence is taken as a state to be classified.

[0035] Each state to be classified is projected onto the orthogonal subspace state codewords in the orthogonal subspace codebook. The projection coefficients and subspace residuals of each state to be classified relative to each orthogonal subspace state codeword are calculated to form the codeword matching results corresponding to each state to be classified.

[0036] Based on the codeword matching results corresponding to each state to be classified, the attribution relationship between each state to be classified and each orthogonal subspace state codeword is compared to determine the target orthogonal subspace state codeword corresponding to each state to be classified, thus forming a state classification result corresponding to the time position.

[0037] Based on the preset number of the state codeword of each target orthogonal subspace in the orthogonal subspace codebook, the state classification results are numbered and mapped to obtain a transportation state number sequence that is consistent with the time order of the transportation potential state sequence.

[0038] The continuity of the numbering consistency of adjacent time positions in the transportation status numbering sequence is determined, and the potential transportation states that are continuously assigned to the same orthogonal subspace status codeword are extracted to form a continuous assignment result.

[0039] Based on the continuous inclusion results, the potential transport states that are continuously included in the same orthogonal subspace state codeword are identified as the same dangerous state group.

[0040] Furthermore, step 5 includes:

[0041] Based on the attribution relationship of the hazard group corresponding to each time position in the transport status number sequence, the transport status number sequence is scanned sequentially to extract the occurrence position of each hazard group in the continuous transport time period, forming a hazard group distribution sequence.

[0042] Based on the interval relationship between adjacent time positions in the distribution sequence of hazardous state groups, consecutive occurrence segments belonging to the same hazardous state group are connected to form a continuous distribution segment of hazardous state groups.

[0043] Based on the start and end positions, duration, and order of distribution of continuous distribution segments of hazardous state groups within a continuous transportation period, the continuous distribution of each hazardous state group is time-series tracked to form a hazardous state change sequence.

[0044] Furthermore, step 6 includes:

[0045] Based on the start and end positions and adjacent intervals of the continuous distribution segments of each hazard state group in the hazard state change sequence, the hazard state change sequence is continuously merged to form candidate continuous occurrence segments;

[0046] Calculate the consecutive occurrence parameter based on the duration, frequency, and distribution location of the candidate consecutive occurrence segment;

[0047] The continuous occurrence parameter is compared with the structural instability threshold: the case where the continuous occurrence parameter meets the structural instability threshold is determined to meet the continuous occurrence condition; the case where the continuous occurrence parameter does not meet the structural instability threshold is determined to not meet the continuous occurrence condition; thus forming the continuous occurrence determination result;

[0048] A hazard assessment flag is generated based on the consecutive occurrence of assessment results.

[0049] Further, step 7 includes:

[0050] Receive the danger judgment flag, and construct the result branch flag according to the judgment value corresponding to the danger judgment flag, so that the result branch flag corresponds to the condition of continuous occurrence and the condition of non-continuous occurrence;

[0051] The results are mapped to the continuous transportation time periods corresponding to the undisturbed sample. If the results indicate that the continuous occurrence condition is met, a dangerous transportation judgment result for the undisturbed sample is generated; if the results indicate that the continuous occurrence condition is not met, a non-dangerous transportation judgment result for the undisturbed sample is generated.

[0052] A risk assessment system for undisturbed sample transportation based on vibration reduction theory includes:

[0053] The detection unit is used to acquire the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube, and to form a transport response segment sequence according to the continuous transport time period;

[0054] The computational unit is used to input the transport response fragment sequence into MTS-JEPA, perform joint embedding mapping on short-term impact scales and long-term cumulative scales to form a transport latent state sequence, and input the transport latent state sequence into a soft codebook.

[0055] The computing unit is also used to construct orthogonal subspace state codewords from each state codeword in the soft codebook using AMP to form an orthogonal subspace codebook, and classify the potential state sequence of transportation based on the orthogonal subspace codebook to obtain the transportation state number sequence, determine the same dangerous state group based on the transportation state number sequence to form a dangerous state change sequence, and form a danger judgment mark under the structural instability threshold according to the dangerous state change sequence.

[0056] The output unit is used to output the result of the dangerous transport judgment of the original sample based on the hazard judgment mark.

[0057] The advantages of this invention compared to the prior art are:

[0058] (1) This invention focuses on three types of monitoring quantities: vibration of the transport container, displacement of the shock-absorbing support, and constraint changes of the sealed soil sample tube. A six-dimensional state vector is formed according to the transport transmission chain, and transport response segments are constructed at 128 consecutive sampling times. Short-term impact information and long-term cumulative information are then compressed into a 32-dimensional transport potential state. The above structure enables this invention to identify the combined effect of instantaneous impact and gradual boundary changes within the same continuous transport time period. It does not simply equate the risk of transporting undisturbed samples with a single excessive vibration, and its scenario adaptability is closer to the actual instability formation process in the long-distance transport of undisturbed samples.

[0059] (2) The single-vector state codewords in the soft codebook are transformed into orthogonal subspace state codewords to avoid the discrete fragmentation of continuous nearest neighbor states under the same contact boundary change mechanism. With eight state codewords, the dominant change direction that each codeword can carry is expanded from one to four, and the classification is completed by combining the projection coefficient and the subspace residual. Then, the continuous occurrence of dangerous state groups is tracked using sixteen sampling times as the judgment window. The above design is conducive to merging states with different amplitudes and different local drifts under the same dangerous mechanism into the same dangerous state group, thereby providing a more coherent and interpretable intermediate basis for structural instability judgment.

[0060] (3) This invention does not stop at single-moment alarms or sporadic anomaly identification, but further forms dangerous state groups, dangerous state change sequences, and continuous occurrence judgment results along the transportation state number sequence. This expands the dangerous identification object from single-moment anomalies to the continuous distribution of dangerous state groups, improves the ability to identify the continuous occurrence process of contact boundary changes, and makes dangerous transportation conclusions based on continuous evolution processes. Therefore, this invention is more suitable for completing scenario-specific dangerous judgment and result delivery under the condition of long-distance transportation of undisturbed samples. Attached Figure Description

[0061] Figure 1 The flowchart shows the method for determining the transportation risk of undisturbed samples based on vibration reduction theory.

[0062] Figure 2 Flowchart for obtaining the transportation response segment sequence;

[0063] Figure 3 This is a diagram showing the correspondence between the three-layer structure and the detection quantity transfer link;

[0064] Figure 4 Flowchart for generating a potential sequence of transportation states;

[0065] Figure 5 Flowchart for forming the orthogonal subspace codebook;

[0066] Figure 6 To create a flowchart for hazardous status groups;

[0067] Figure 7 To create a flowchart of the sequence of dangerous state changes;

[0068] Figure 8 To create a flowchart for hazard assessment signs. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0070] While general time-series coding, codebook discretization, and subspace clustering methods for monitoring data processing can characterize multivariate signal changes, when directly applied to long-distance transportation scenarios of undisturbed samples, they often fail to incorporate the interconnected responses of the transport container, shock absorber support, and encapsulated soil sample cylinder along the transmission chain into the same state semantics, nor do they stably group consecutive similar states around the contact boundary change mechanism. Therefore, existing technologies struggle to continuously track the dangerous evolution process caused by both instantaneous impact and slow accumulation to the structural instability determination stage. This invention obtains transport response segment sequences by acquiring the vibration of the transport container, the force and displacement of the shock absorber support, and the constraint changes of the encapsulated soil sample cylinder; inputs these transport response segment sequences into MTS-JEPA, performing joint embedding mapping on short-term impact and long-term accumulation scales to obtain a transport potential state sequence; constructs orthogonal subspace state codewords using AMP in the soft codebook, forming an orthogonal subspace codebook; classifies the transport potential state sequences to form dangerous state groups and dangerous state change sequences; and generates a danger determination result under the structural instability threshold. It enables a unified state characterization of the linkage response of the transport container, shock-absorbing support and encapsulated soil sample cylinder in the transport transmission chain, and identifies the dangerous transport state caused by the change of contact boundary under the combined action of short-term impact and long-term accumulation, thereby completing the dangerous transport judgment of the original sample.

[0071] like Figure 1 As shown, the method for determining the transportation risk of undisturbed samples based on vibration reduction theory includes:

[0072] Step 1: Obtain the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube, and form a transport response segment sequence according to the continuous transport time period.

[0073] like Figure 2 As shown, at the same sampling time, the vibration of the transport box, the force displacement of the shock-absorbing support, and the constraint change of the sealed soil sample tube are obtained respectively, and combined according to the positional order of the transport box, the shock-absorbing support, and the sealed soil sample tube in the transport transmission chain to form a transport state vector corresponding to the sampling time.

[0074] Vibration of the transport container, displacement of the shock-absorbing support, and constraint changes of the sealed soil sample tube were collected synchronously under the same sampling clock, and the unified time axis was written as follows: ,in , Indicates the starting sampling time of a continuous transportation period. Indicates the sampling interval. This represents the total number of sampling points within a continuous transportation period. In each Vibration of the transport container was obtained from the above. Displacement of shock-absorbing support and the constraint variation of the encapsulated soil sample tube The vibration of the transport container is written as... , Indicates longitudinal vibration, Indicates lateral vibration. Indicates vertical vibration; the force displacement of the damping support is written as... , Indicates supporting force. Represents relative displacement; changes in the constraints of the encapsulated soil sample tube are written as... , This represents the change in packaging constraints. Changes in the transport container vibration, the force displacement of the shock-absorbing support, and the constraints of the packaging soil sample cylinder are all expressed as the same... Establish a correspondence for the indexes. In each The vibration of the transport container, the force and displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube are combined according to their positional order in the transport chain. The positional order in the transport chain is fixed as transport container, shock-absorbing support, and sealed soil sample tube, as follows: Figure 3 As shown, this forms the transportation state vector. , .

[0075] The transportation state vectors corresponding to consecutive sampling times are arranged in chronological order, and the correspondence between the vibration of the transportation box, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample tube is maintained at the same sampling time, forming a temporal state chain within the continuous transportation time period.

[0076] All transport state vectors are arranged according to Arranged in ascending order, we obtain a time-series state chain. The temporal state chain is a continuous sequence of states, specifically defined as follows: the sampling times corresponding to any adjacent index positions satisfy a fixed sampling interval. Furthermore, all six components at the same index position originate from the same sampling time. If a certain If any of the above components is missing, the sampling location does not constitute a valid transportation state vector, and the interval containing the sampling location is not included in the continuous transportation time period for interception.

[0077] A continuous transport state vector is extracted from the temporal state chain with a preset segment length. Each continuous transport state vector is spliced ​​together in chronological order, while keeping the transport chain order within each continuous transport state vector unchanged, to form a candidate transport segment.

[0078] With preset segment length In the time-series state chain The slide cut is performed on the top, where This indicates the number of consecutive sampling moments contained in a single candidate transport segment. For each starting index... Construct candidate transport segments Each candidate transport segment is composed of... A series of continuous transport state vectors are concatenated in chronological order, and the segment contains the first... The row corresponds to the first The transportation state vector at the nth sampling time, the th The six components in the row maintain the same order of transport container, shock-absorbing support, and sealed soil sample cylinder in the transport chain.

[0079] Based on the complete coverage relationship of continuous sampling times in the candidate transportation segments and the synchronous correspondence between the vibration of the transportation box, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample tube, the candidate transportation segments are retained to form a transportation response segment sequence.

[0080] Among them, complete coverage relationship refers to the coverage relationship between candidate transport segments and consecutive sampling times corresponding to a preset segment length within a continuous transport time period, without missing or jumping. Index interval Each sampling location within the range has a valid transport state vector, and the sampling time interval between any two adjacent sampling locations is 0. There are no missing sampling positions or jump sampling positions within the index interval.

[0081] The synchronization relationship is specifically limited to: within the index range within, no. The transport state vector corresponding to each sampling location must be written in full. ,in, The three vibration components of the transport container, the two force-displacement components of the shock-absorbing support, and the one constraint change component of the encapsulated soil sample tube in the transport state vector all originate from the same sampling time. Furthermore, within the candidate transport segment, a fixed arrangement order is maintained for "transport container vibration, shock absorber support force displacement, and changes in the constraints of the encapsulated soil sample cylinder." If the index interval... If any sampling position contains a missing component, or if the sampling time interval between any two adjacent sampling positions is not equal to the fixed sampling interval. Or, if the six components in any transport state vector do not correspond to the same sampling time, then the candidate transport segment... Not to be reserved.

[0082] Step 2: Input the transport response fragment sequence into MTS-JEPA, perform joint embedding mapping on the short-term impact scale and the long-term cumulative scale to obtain the transport potential state sequence, and input the transport potential state sequence into the soft codebook.

[0083] like Figure 4As shown, the transport response fragment sequence is input into the input mapping layer of MTS-JEPA. The input mapping layer uses a one-dimensional convolution layer to locally connect and map the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample tube at adjacent sampling times, forming basic temporal features that maintain the time sequence and the correspondence between the transport transmission chain.

[0084] Suppose that the transport response fragment sequence contains A transport response fragment, with the fragment index written as... . No. A transportation response fragment is written as ,in Indicates the sampling time index within the segment. Each Given a 6-dimensional input vector, written as . , , These represent the longitudinal vibration, lateral vibration, and vertical vibration of the transport container, respectively. , These represent the supporting force and relative displacement in the stress-displacement process of the damping support, respectively. This represents the change in encapsulation constraints during the variation of the encapsulated soil sample tube constraints. Therefore, the input dimension for each transport response segment is... The transport response segment sequence enters the MTS-JEPA sequentially according to the order of continuous transport time periods, without changing the arrangement of transport container vibration, shock absorber support force displacement and encapsulated soil sample cylinder constraint changes in the transport transmission chain.

[0085] The first part of MTS-JEPA is the input mapping layer. The input mapping layer maps each transport response segment. For separate processing, the input mapping layer uses a single one-dimensional convolution with 48 kernels and a kernel length of 3. The convolution direction is along the sampling time index. The input mapping layer expands in an increasing direction. Each convolutional neuron receives a 6-dimensional input vector from three adjacent sampling times. It performs local connection mapping on the vibration of the transport container, the force and displacement of the shock absorber support, and the constraint changes of the encapsulated soil sample cylinder within the local time neighborhood. Each convolutional neuron contains convolutional weights, biases, and ReLU nonlinear units. The convolutional parameters of the input mapping layer are set to keep the time length constant, so that the time positions output by the input mapping layer are still 128. The input mapping layer forms the basic temporal features. Each It is a 48-dimensional vector, and the overall dimension of the basic time series features is... Each time position in the basic temporal features corresponds to the corresponding sampling time in the transport response segment. The basic temporal features are used to transform the original 6-dimensional input into a feature representation that preserves the temporal order and the correspondence between the transport delivery chain.

[0086] The basic temporal features are input into the local temporal extraction layer. The local temporal extraction layer uses two layers of cascaded one-dimensional convolution to progressively extract the basic temporal features of continuous time locations, so that the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube are uniformly correlated within the same continuous transport time period, thus obtaining local state features.

[0087] The second part of MTS-JEPA is the local timing extraction layer. The local timing extraction layer receives the basic timing features. The local temporal extraction layer employs two concatenated one-dimensional convolutions. The first layer has 64 kernels with a kernel length of 5, and the second layer has 64 kernels with a kernel length of 3. Both one-dimensional convolutions perform progressive extraction along the time axis. Each convolutional neuron in both layers contains convolutional weights, biases, and ReLU nonlinear units. The parameters of both convolutions are set to a time length of 128. The local temporal extraction layer forms local state features. Each It is a 64-dimensional vector, and the overall dimension of the local state features is... The local state feature organizes local fluctuations at continuous time locations into a unified and correlated state representation. This means that the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube are no longer used as discrete inputs for judgment, but are instead used as locally correlated states within the same continuous transport time period for subsequent processing.

[0088] The local state features are input into the short-term impact-scale convolutional layer in the dual-scale state generation layer to perform convolution calculations on the local fluctuations caused by instantaneous impacts, forming short-term state features that reflect the synchronous response of sudden vibration changes in the transport container, instantaneous displacement of the shock-absorbing support, and changes in the constraints of the encapsulated soil sample tube. The local state features are then input into the long-term cumulative-scale dilated convolutional layer in the dual-scale state generation layer to perform dilated convolution calculations on the gradually changing transmission relationship caused by the gradual changes in the contact boundaries of the transport container, shock-absorbing support, and encapsulated soil sample tube, forming long-term state features.

[0089] The third part of MTS-JEPA is the dual-scale state generation layer. The dual-scale state generation layer receives local state features. Furthermore, short-time impact-scale convolutional layers and long-time cumulative-scale dilated convolutional layers are applied in parallel on the same input object. The short-time impact-scale convolutional layer performs one-dimensional convolution calculations on the instantaneous fluctuations in the local state features, with the convolution parameters set to keep the time length constant, thus forming short-time state features. Each It is a 64-dimensional vector, with an overall dimension of . . The short-term state features are used to characterize the synchronous response of the sudden change of the vibration of the transportation box, the instantaneous displacement of the force on the shock-absorbing support seat, and the change of the constraint of the encapsulated soil sample cylinder. The long-term cumulative scale dilation convolutional layer performs dilation convolution calculation on the slow-changing relationship in the local state features, and the dilation parameter and the convolution parameter are set to keep the time length unchanged, forming the long-term state features , each is a 64-dimensional vector, and the overall dimension is . The long-term state features are used to characterize the slow-changing transfer relationship caused by the gradual change of the contact boundary between the transportation box, the shock-absorbing support seat and the encapsulated soil sample cylinder. MTS-JEPA does not adopt a recursive state memory structure in the dual-scale state generation layer. The state prediction of MTS-JEPA for each time position depends on the neighborhood context extracted by the input mapping layer, the local time series extraction layer, the short-term shock scale convolutional layer and the long-term cumulative scale dilation convolutional layer along the time axis, thereby realizing the joint embedding prediction on the short-term shock scale and the long-term cumulative scale.

[0090] Align the short-term state features and the long-term state features at the same time position, and construct a joint state feature that simultaneously carries local shock information and cumulative change information according to the alignment result. Input the joint state feature into the transportation latent state generation layer, perform splicing mapping on the joint state feature, and do not form independent determination results for the short-term state feature and the long-term state feature respectively, but form a transportation latent state vector corresponding to the same time position. Arrange the transportation latent state vectors in the chronological order of consecutive transportation time periods to form a transportation latent state sequence.

[0091] The fourth part of MTS-JEPA is the alignment and splicing organization process before the transportation latent state generation layer. For each and each , align with at the same time position. After completing the alignment, splice them in a fixed order, so that the first part of the joint state feature corresponds to the short-term state feature, and the second part corresponds to the long-term state feature. The joint state feature is written as , each has a dimension of 128, where the first 64 dimensions come from the short-term state feature, and the last 64 dimensions come from the long-term state feature. All constitute the joint state feature sequence , and the overall dimension is . This fixed-order splicing method enables the local shock information and the cumulative change information at the same time position to enter the unified state layer, without forming an independent determination result for the short-term state feature and without forming an independent determination result for the long-term state feature.

[0092] The transportation latent state generation layer receives the joint state feature sequence The transportation potential state generation layer generates each... Performing the same concatenation mapping, the transport latent state generation layer consists of 128 fully connected neurons and 32 output neurons. The 128 fully connected neurons receive 128 dimensions of joint state features at the same time point, performing a unified mapping on the first 64 dimensions of short-term state features and the latter 64 dimensions of long-term state features; the 32 output neurons compress the mapping result into a transport latent state vector. The potential state vector for transportation is written as... The dimension is 32, of which to The part that corresponds to the short-term impact information. to This corresponds to the part that carries over long-term accumulated information. Each fully connected neuron in the transport latent state generation layer receives all 128 input components at the same time position, does not receive joint state features from other time positions, and does not generate the transport latent state vector at the current time position based on the transport latent state vector at the previous time position. Therefore, the transport latent state vector is formed by direct mapping prediction at the same time position, rather than recursive prediction across time positions.

[0093] For the After mapping all time positions of each transportation response segment using the transportation potential state generation layer, a transportation potential state sequence is formed. The overall dimension is All transport response segments in the transport response segment sequence are sorted by segment index. The process involves repeatedly executing the input mapping layer, local temporal extraction layer, dual-scale state generation layer, alignment and splicing organization process, and transport potential state generation layer in sequence to obtain a transport potential state sequence arranged in fragment order. Each time position in the transport potential state sequence is a 32-dimensional transport potential state vector, and each transport potential state vector simultaneously carries local impact information and cumulative change information. The transport potential state sequence serves as the input object for the soft codebook in the construction and classification of state codewords.

[0094] Although the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample cylinder all originate from the same transport process, the three types of responses exhibit alternating localized impacts and gradually accumulating boundary changes over time. Hazardous transport conditions are not always manifested as a single excessively large response, but often as short-term impacts and long-term accumulations occurring simultaneously within the same transport time period. This invention defines the transport response segment sequence as the state input on the transport transmission chain, and then uses MTS-JEPA to perform joint embedding prediction on the same transport time period at both the short-term impact scale and the long-term accumulation scale. This transforms each transport response segment from the original measurement layer into a potential transport state sequence that can simultaneously carry impact and accumulation information, thereby improving the accuracy of subsequent judgments.

[0095] Step 3: In the soft codebook, AMP is used to construct each state codeword into an orthogonal subspace state codeword, forming an orthogonal subspace codebook. The orthogonal subspace state codewords correspond to similar transmission patterns caused by the vibration of the transport box, the force displacement of the shock-absorbing support, and the change of the contact boundary of the sealed soil sample tube.

[0096] like Figure 5 As shown, the sequence of potential transportation states is input into a soft codebook, and each state codeword in the soft codebook is determined as a state codeword to be constructed, so that each state codeword to be constructed corresponds to a distribution range of potential transportation states to be carried.

[0097] The software codebook has 8 storage locations for status codewords, and the status codeword index is written as... Before AMP intervention, each state codeword storage location corresponds to a 32-dimensional single-vector state codeword, written as... . to Corresponding to the 16 potential state dimensions where short-term shock information is located, to The 16 potential state dimensions correspond to the long-term accumulated information. Eight single-vector state codewords are used as the state codewords to be constructed, each corresponding to one of the eight potential transportation state distribution ranges to be carried.

[0098] Based on the temporal adjacency relationship, amplitude drift relationship, and linkage relationship of the transportation potential state sequence during continuous transportation time period, as well as the changes in the vibration of the transportation box, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample tube, similar transmission patterns caused by the vibration of the transportation box, the force displacement of the shock-absorbing support, and the contact boundary changes of the encapsulated soil sample tube are extracted to form subspace construction constraints corresponding to each codeword to be constructed.

[0099] For each state codeword to be constructed The subspace construction constraints are extracted from the potential transport state sequence. The extraction of subspace construction constraints is performed on a unit of locally continuous windows, denoted as [equation missing]. ,in This represents the length of a local continuous window. Temporal adjacency is defined as the transportation potential state vectors within the same local continuous window being continuously arranged at the sampling time; amplitude drift is defined as the 32-dimensional transportation potential state vectors within the same local continuous window gradually changing along similar directions; and linkage is defined as the first 16 dimensions of short-term impact information and the last 16 dimensions of long-term cumulative information in the 32-dimensional transportation potential state vectors changing together within the same local continuous window, jointly corresponding to the unified state evolution of the transportation box vibration, the force displacement of the damping support, and the constraint changes of the encapsulated soil sample cylinder along the contact boundary. Based on temporal adjacency, amplitude drift, and linkage, similar transmission patterns carried by each state codeword to be constructed are aggregated, thereby forming subspace construction constraints corresponding to eight state codewords to be constructed.

[0100] Under the action of AMP, orthogonal basis construction is performed on each state codeword to be constructed according to the subspace construction constraints. Each state codeword to be constructed is updated from a single vector state codeword to a candidate orthogonal subspace state codeword composed of multiple mutually orthogonal basis vectors, so that each candidate orthogonal subspace state codeword carries the potential transport states with different amplitudes and different local drifts under the same contact boundary change mechanism.

[0101] Within the soft codebook, AMP performs orthogonal basis construction on each state codeword to be constructed based on subspace construction constraints. For the... Given a set of state codewords to be constructed, AMP generates four unit orthogonal basis vectors, denoted as... , , and Each unit orthogonal basis vector is a 32-dimensional vector, written as . to Corresponding to the 16 potential state dimensions where short-term shock information is located, to This corresponds to the 16 latent state dimensions where long-term accumulated information resides. The four unit orthogonal basis vectors are combined column-wise as follows: , for The orthogonal basis structure, composed of The status code representing the state is written as , This represents the candidate orthogonal subspace state codeword. Four unit orthogonal basis vectors jointly carry the transport potential states with different amplitudes and local drifts under the same contact boundary change mechanism. In one embodiment, Characterizes the dominant direction of change in similar transmission patterns. and Characterizing the local drift direction under the same contact boundary change mechanism. Characterizes the cumulative offset direction under the same contact boundary change mechanism.

[0102] The potential transport state sequence is projected onto each candidate orthogonal subspace state codeword, and the projection coefficient and subspace residual corresponding to each candidate orthogonal subspace state codeword are calculated to form the basis for classifying state codewords that distinguish between similar and dissimilar transport patterns.

[0103] Each transportation potential state vector The state codewords are projected onto eight candidate orthogonal subspaces respectively. Because... , , and The transport potential state vector is a unit orthogonal basis vector. The projection coefficients in the four orthogonal basis directions are directly obtained by the inner product of the 32-dimensional transport latent state vector and the corresponding unit orthogonal basis vector, forming a 4-dimensional projection coefficient vector. A subspace reconstruction vector is constructed based on four projection coefficients and four orthogonal basis vectors, and the length of the difference between the transportation potential state vector and the subspace reconstruction vector is used as the subspace residual. The transport response segment in the first The sampling time relative to the first sampling time The criteria for assigning a state codeword to a candidate orthogonal subspace state codeword are written as follows:

[0104] ,

[0105] In the formula, Indicates the basis for the status code word's attribution; Indicates the index of the transport response fragment; Indicates the sampling time index; Indicates the state codeword index of the candidate orthogonal subspace; Indicates the first Indices of orthogonal basis vectors in candidate orthogonal subspace state codewords; Represents the potential state vector of transportation In the unit orthogonal basis vectors Projection coefficient in the direction; Represents a 32-dimensional transportation potential state vector; Indicates the first The th candidate orthogonal subspace state codeword A 32-dimensional identity orthogonal basis vector; Represents the squared Euclidean norm, which corresponds to the transport potential state vector relative to the first... Subspace residuals of candidate orthogonal subspace state codewords; This represents a pre-defined positive number with the same scale as the square Euclidean norm. Since the transport latent state vector, unit orthogonal basis vector, projection coefficients, and subspace residuals all reside in the same 32-dimensional latent state space, the numerator and denominator of the state codeword attribution are both composed of square quantities within the latent state space.

[0106] Based on the state codeword attribution criteria, the orthogonal base direction and subspace coverage of each candidate orthogonal subspace state codeword are adjusted, and the candidate orthogonal subspace state codewords that meet the preset attribution conditions are determined as orthogonal subspace state codewords, so that the potential transport states belonging to the same orthogonal subspace state codeword correspond to the same contact boundary change mechanism; each orthogonal subspace state codeword is written into a soft codebook to form an orthogonal subspace codebook.

[0107] For each transport potential state vector After calculating the eight state codeword attribution criteria, the orthogonal basis directions and subspace coverage of each candidate orthogonal subspace state codeword are adjusted based on these criteria. If the same group of potential transport state vectors continuously achieves the maximum value relative to the state codeword attribution criteria of a candidate orthogonal subspace state codeword within the same local continuous window, and the same group of potential transport state vectors satisfies the temporal adjacency, amplitude drift, and linkage change relationships under the same contact boundary change mechanism, then the four unit orthogonal basis vectors of the corresponding candidate orthogonal subspace state codeword are retained, and the subspace coverage is adjusted according to the concentration range of the state codeword attribution criteria in the 32-dimensional potential state space. If the same group of potential transport state vectors does not continuously achieve the maximum value relative to the state codeword attribution criteria of a candidate orthogonal subspace state codeword within the same local continuous window, or if the same group of potential transport state vectors corresponds to different contact boundary change mechanisms, then the orthogonal basis directions and subspace coverage of the corresponding candidate orthogonal subspace state codewords are adjusted so that the corresponding candidate orthogonal subspace state codewords only carry the potential transport states under the same contact boundary change mechanism. Candidate orthogonal subspace state codewords that meet the preset attribution conditions are determined as orthogonal subspace state codewords.

[0108] The eight determined orthogonal subspace state codewords are written into the software codebook to form the orthogonal subspace codebook, which is written as follows: Each orthogonal subspace state codeword in the orthogonal subspace codebook is composed of a... The orthogonal basis structure representation no longer uses a single 32-dimensional state codeword representation. When any 32-dimensional transportation potential state vector is input into the orthogonal subspace codebook, the orthogonal subspace codebook can generate 8 state codeword classification criteria corresponding to 8 orthogonal subspace state codewords, and the 8 state codeword classification criteria are used as the unified judgment basis for the classification of transportation potential states.

[0109] The hazardous states in undisturbed sample transportation are not fixed single-point states, but rather a set of similar transmission patterns that unfold along local directions as the contact boundary loosens, local slippage occurs, and constraints change. The potential transportation states corresponding to the same hazardous mechanism are often distributed as a cluster of neighboring states, rather than concentrated at a single location. This invention transforms the state codeword from a single-vector prototype into an orthogonal subspace state codeword, so that the soft codebook no longer stores potential transportation states as single points, but rather as a set of similar transmission patterns with a common contact boundary change mechanism. The orthogonal subspace codebook formed by this processing directly determines the classification basis in step 4. If the orthogonal subspace codebook is replaced with ordinary state codewords, the hazardous state group will be broken up, the hazardous state change sequence will not reflect the continuous occurrence of the same hazardous mechanism in the time direction, and the hazard judgment indicator under the structural instability threshold will also lose reliable input.

[0110] Step 4: Classify the transportation potential state sequence based on the orthogonal subspace codebook to obtain the transportation state number sequence, and determine the transportation potential states that are continuously classified into the same orthogonal subspace state codeword as the same dangerous state group.

[0111] like Figure 6 As shown, the sequence of potential transportation states is input into the orthogonal subspace codebook in chronological order, and each potential transportation state in the sequence is taken as a state to be classified.

[0112] Each state to be classified is projected onto the orthogonal subspace state codewords in the orthogonal subspace codebook. The projection coefficients and subspace residuals of each state to be classified relative to each orthogonal subspace state codeword are calculated to form the codeword matching results corresponding to each state to be classified.

[0113] Based on the codeword matching results corresponding to each state to be classified, the attribution relationship between each state to be classified and each orthogonal subspace state codeword is compared to determine the target orthogonal subspace state codeword corresponding to each state to be classified, thus forming a state classification result corresponding to the time position.

[0114] Based on the preset number of each target orthogonal subspace state codeword in the orthogonal subspace codebook, the state classification result is numbered and mapped to obtain a transportation state number sequence that is consistent with the time order of the transportation potential state sequence.

[0115] The continuity of the numbering consistency of adjacent time positions in the transportation state numbering sequence is determined, and the potential transportation states that are continuously assigned to the same orthogonal subspace state codeword are extracted to form a continuous assignment result.

[0116] Based on the continuous inclusion results, the potential transport states that are continuously included in the same orthogonal subspace state codeword are identified as the same dangerous state group.

[0117] Step 4 directly uses the transportation potential state sequence and orthogonal subspace codebook as input objects, performs codeword matching on the transportation potential state vector at each time position one by one, and then performs continuity determination on the codeword matching results in chronological order.

[0118] Transport potential state vector As states to be classified, they are projected onto the state codewords of the eight orthogonal subspaces, respectively. to For the first State codewords of orthogonal subspaces ,Will and , , and Perform inner product calculations separately to obtain four projection coefficients, forming a 4-dimensional projection coefficient vector. The four projection coefficients are all directly calculable values, representing the characteristics of the potential transport state vector along the four unit orthogonal basis directions. The four projection coefficients are then multiplied and summed term by term by the four unit orthogonal basis vectors to obtain the... A pair of orthogonal subspace state codewords The subspace reconstruction vector. The transportation potential state vector. The squared Euclidean distance between the vectors and the subspace reconstruction vectors is denoted as . , That is, the first The transport response segment in the first The time position relative to the first The subspace residuals of the codewords in each orthogonal subspace state. Thus, each state to be classified corresponds to 8 sets of projection coefficients and 8 subspace residuals. The 8 sets of projection coefficients and 8 subspace residuals together constitute the codeword matching result of the current state to be classified.

[0119] For each state to be classified, the projection coefficients first reflect the degree of fit between the state and the orthogonal subspace state codewords. Then, the subspace residuals reflect the unexplained differences between the orthogonal subspace state codewords and the state to be classified. To ensure that the decision order of "first comparing projection coefficients, then comparing the subspace residuals corresponding to the projection coefficients" is implemented as a single calculation rule, the... The transport response segment in the first The target orthogonal subspace state codeword index corresponding to each time position is denoted as . And determine it according to the following formula:

[0120] ,

[0121] In the formula, Indicates the first The transport response segment in the first The target orthogonal subspace state codeword index corresponding to each time position; Indicates the index of the transport response fragment; Indicates the time location index within the transport response segment; This represents the orthogonal subspace state codeword index that maximizes the ratio within the parentheses; The preset number representing the state codeword of the orthogonal subspace; Indicates the first Numbering of the unit orthogonal basis vectors in the state codeword of each orthogonal subspace; Indicates the first The transport response segment in the first The potential transportation state vector corresponding to the nth time position follows the i-th time position along the i-th time position. In the state codeword of the orthogonal subspace, the first Projection coefficients of the directions of the unit orthogonal basis vectors; Indicates the first The total projection intensity of the orthogonal subspace state codewords onto the current state to be classified; Indicates the first The transport response segment in the first The potential transportation state vector corresponding to the nth time position is relative to the nth time position. Subspace residuals of orthogonal subspace state codewords; Indicates a preset positive number. subspace residual Numerical scales in the same potential state space are used.

[0122] In the above formulas, all objects involved in the calculation are directly obtainable underlying values. The four projection coefficients are obtained by directly performing the inner product of the 32-dimensional transportation latent state vector and the 32-dimensional unit orthogonal basis vector; the subspace residual is obtained directly by the squared Euclidean distance between the 32-dimensional transportation latent state vector and the 32-dimensional subspace reconstruction vector; a preset positive number is used. It is preset by the system. Since the sum of squares of the projection coefficients and the square Euclidean distance are both on the same square numerical scale of the 32-dimensional potential state space, there is no dimension mismatch between the numerator and denominator.

[0123] For each state to be classified, eight sets of projection coefficients and eight subspace residuals are substituted to calculate eight comparable codeword matching values. The orthogonal subspace state codeword with the largest codeword matching value is selected as the target orthogonal subspace state codeword. If multiple orthogonal subspace state codewords have the same codeword matching value, the target orthogonal subspace state codeword is determined according to the preset number from smallest to largest.

[0124] After determining the target orthogonal subspace state codeword at each time position, the time position is... State codewords in the orthogonal subspace of the target Establish the corresponding relationships to generate the state classification results. The state classification results are written as follows: . Each item in the table represents the classification correspondence between a time position and a target orthogonal subspace state codeword. Since the preset numbers of each orthogonal subspace state codeword in the orthogonal subspace codebook are fixed at 1 to 8, no additional conversion rules are introduced during number mapping; the codewords are directly mapped... As the first Transport status number at each time location A transport status number sequence is composed of transport status numbers at all time locations. The transport status number sequence is: The discrete numbering sequence of the transportation state numbering sequence, where each number corresponds to a potential transportation state vector at a given time location.

[0125] When performing continuity determination on a transport status number sequence, it is based on the time position index. Scan sequentially from 0 to 127. Continuity is determined by comparing the consistency of numbers at adjacent time positions. If Then the first The time position and the first The potential transportation state vectors corresponding to each time location are retained within the same continuous inclusion segment; if Then in the first The current continuous segment ends at the [time position], and from the [time position]... A new consecutive inclusion segment begins at each time position. All consecutive inclusion segments obtained through sequential scanning are written as follows: .in, Indicates the first The number of consecutively included segments in a transport response segment Indicates the first The start and end time intervals of a consecutive segment. Indicates the first Numbering of the orthogonal subspace state codewords corresponding to consecutive incorporation segments, and satisfying any All The continuous induction result consists of all continuous induction segments, each of which has a clear start time position, end time position, and a unique orthogonal subspace state codeword number.

[0126] When determining a hazard state group based on consecutive inclusion results, all potential transport state vectors within the same consecutive inclusion segment are uniformly assigned to the same hazard state group. Each dangerous state group is written as If two time positions are adjacent and correspond to the same target orthogonal subspace state codeword, then the potential transport state vectors at the two time positions are assigned to the same hazardous state group. If two time positions are not adjacent, or correspond to different target orthogonal subspace state codewords, then the potential transport state vectors at the two time positions are not assigned to the same hazardous state group. Thus, hazardous state groups maintain a correspondence with consecutively assigned segments. Transport State Number Sequence Record the numbering changes at 128 time points, and establish a set of hazardous state groups. The set of potential transport states that are continuously classified into the same orthogonal subspace state codewords is recorded. Together, these two forms the basis for state classification within a continuous transport time period.

[0127] Step 5: Track the continuous distribution of hazardous status groups based on the transport status number sequence to form a hazardous status change sequence.

[0128] like Figure 7 As shown, based on the attribution relationship of the danger state group corresponding to each time position in the transportation status number sequence, the transportation status number sequence is sequentially scanned to extract the occurrence position of each danger state group in the continuous transportation time period, forming a danger state group distribution sequence.

[0129] Based on the interval relationship between adjacent time positions in the distribution sequence of the dangerous state groups, the consecutive occurrence segments belonging to the same dangerous state group are connected to form a continuous distribution segment of dangerous state groups.

[0130] Based on the start and end positions, duration, and distribution sequence of the continuous distribution segments of the hazardous state groups within a continuous transportation time period, the continuous distribution of each hazardous state group is time-series tracked to form a hazardous state change sequence.

[0131] The transport status number sequence obtained in step 4 is used as the sole input for step 5. Since each transport status number in step 4 has been directly mapped from the preset number of the target orthogonal subspace state codeword, and each preset number corresponds to a hazardous state group category, therefore... Directly as the first The hazard group attribution identifier for each time location is denoted as: The hazardous status group affiliation identifier does not require additional calculation; it is directly obtained from the transport status number at the same time and location, satisfying the correspondence of "one time and location corresponds to one hazardous status group affiliation identifier".

[0132] All in ascending order of time position. Perform a sequential scan and pair each time location with the corresponding hazard group affiliation identifier to form a hazard group distribution sequence. Each distribution record is written as The former Indicates the position of occurrence, the next item This indicates the hazard group affiliation identifier corresponding to the location where the hazard occurs. Therefore, the hazard group distribution sequence is as follows: The sequence of records is as follows: the first column is the time position, and the second column is the hazard group affiliation identifier. The hazard group distribution sequence completely preserves the distribution of hazard groups at all 128 time positions within the continuous transportation period, without deleting any time positions or changing their original order.

[0133] After the distribution sequence of hazard state groups is formed, all distribution records are merged and organized according to the hazard state group affiliation identifier. For each hazard state group affiliation identifier... Extract all that meet the requirements The time location forms a location set. Location set The various time positions are still arranged in the original scanning order. Then they are grouped at the same location. Within the process, the intervals between adjacent time positions are compared one by one. If the difference between two adjacent time positions is equal to 1, the two adjacent time positions are connected to form the same continuous occurrence segment; if the difference between two adjacent time positions is greater than 1, the current continuous occurrence segment ends at the previous time position, and a new continuous occurrence segment begins at the next time position. Through this processing method, all continuous occurrence segments of the same hazard group within a continuous transportation time period are separated segment by segment, forming continuous distribution segments of the hazard group.

[0134] Each extracted continuous distribution segment of the hazardous state group is written as: ,in, Indicates the first The total number of consecutively distributed hazardous state groups formed within a single transport response segment. Indicates the index of a continuous distribution segment of a hazardous state group. Indicates the group affiliation identifier for hazardous conditions. Indicates the start time position. Indicates the end time position. Indicates the duration. The duration is obtained directly by counting from the time position. When the starting time position is... The end time position is At that time, the duration corresponds to from arrive The number of consecutive sampling points covered. A sequence of consecutive distribution segments of all hazard state groups. The continuous distribution sequence of the hazardous state group is as follows: The record sequence has four fields, representing the danger group category, start time position, end time position, and duration, respectively.

[0135] After the continuous distribution segment sequence of hazard state groups is formed, all continuous distribution segments of hazard state groups are reordered according to their start time positions in ascending order; if the start time positions are the same, they are sorted according to their end time positions in ascending order. After sorting, each continuous distribution segment of a hazard state group is read sequentially, and the hazard state group identifier, start time position, end time position, and duration are written into the time series tracking record to form a hazard state change sequence. Each time-tracing record is written as Each record in the hazard status change sequence corresponds to a complete continuous distribution of a hazard status group within a continuous transportation time period. The four fields within each record specify the hazard status group category, the start point, the end point, and the duration. The order of the records directly reflects the chronological order of the hazard status groups' distribution within the continuous transportation time period. Therefore, the hazard status change sequence is no longer a single numbered stream, but rather... The time-series tracking record sequence can be directly used in step 6 to perform continuity merging and continuous occurrence determination according to the start and end positions, adjacent interval relationships, duration and occurrence frequency.

[0136] Step 6: Determine whether the continuous occurrence condition is met based on the sequence of dangerous state changes below the structural instability threshold, and form a dangerous judgment mark.

[0137] like Figure 8 As shown, based on the start and end positions and adjacent intervals of the continuous distribution segments of each dangerous state group in the dangerous state change sequence, the dangerous state change sequence is continuously merged to form candidate continuous occurrence segments;

[0138] The consecutive occurrence parameter is calculated based on the duration, frequency, and distribution location of the candidate consecutive occurrence segments;

[0139] The continuous occurrence parameter is compared with the structural instability threshold: if the continuous occurrence parameter meets the structural instability threshold, it is determined that the continuous occurrence condition is met; if the continuous occurrence parameter does not meet the structural instability threshold, it is determined that the continuous occurrence condition is not met; thus forming a continuous occurrence determination result;

[0140] A danger assessment flag is generated based on the consecutive occurrence of the assessment results.

[0141] This invention makes a comprehensive judgment by statistically analyzing the continuous occurrence of dangerous state groups on the time axis and combining the duration of the groups with the intensity of the danger. When the continuous dangerous state groups reach a preset threshold, structural instability is triggered, thereby improving the stability of the judgment and the reliability of the engineering.

[0142] When continuously merging sequences of hazardous state changes, the continuous distribution segments of two adjacent hazardous state groups are compared sequentially. The first... The continuous distribution segment of the first danger state group and the first The adjacent interval between consecutive distribution segments of a hazard state group is defined as the number of sampling points obtained by subtracting the start time position of the subsequent segment from the end time position of the previous segment and then subtracting 1. The adjacent interval is used to represent the number of missing sampling points between two consecutive distribution segments of hazard state groups that are not covered by the current hazard state group. A preset interval threshold is set. , This indicates the maximum number of vacant sampling points allowed when the same hazardous condition group is allowed to maintain continuous occurrence of attributes within a continuous transportation period. A positive integer pre-written into the decision configuration table. If and Same, and adjacent intervals are no greater than Then the first The continuous distribution segment of the first danger state group and the first Continuous distribution segments of several dangerous state groups are merged into the same candidate consecutive occurrence segment; if and Different, or adjacent intervals greater than Then in the first The current candidate consecutive occurrence segment ends at the continuous distribution segment of the dangerous state group, and starts from the first... A new candidate consecutive occurrence segment begins from a continuously distributed segment of a hazardous state group. After completing the entire segment scan, a sequence of candidate consecutive occurrence segments is formed. ,in, Indicates the first The number of candidate consecutive occurrence segments formed within a transport response segment This indicates a candidate segment index that appears consecutively.

[0143] Each consecutively occurring candidate segment is written as ,in, This indicates the group affiliation identifier for the dangerous state corresponding to the consecutively occurring candidate segments. Indicates the starting time position of the consecutively occurring candidate segment. This indicates the end time position of the consecutively occurring candidate segment. This indicates the number of consecutive distribution segments of the dangerous state group that are merged into the current candidate consecutive occurrence segment. This represents the maximum value among all adjacent intervals within the currently consecutive candidate segment. Take the starting time position of the first continuous distribution segment of the danger state group that constitutes the current candidate continuous occurrence segment. Take the termination time position of the last continuous distribution segment of the dangerous state group that constitutes the current candidate continuous occurrence segment. The results are obtained directly by comparing adjacent intervals one by one during the merging process. Therefore, candidate consecutive occurrence segments not only retain the merging results of the same hazard group within consecutive transport periods, but also retain all the direct values ​​required to determine the consecutive occurrence condition.

[0144] Calculate the consecutive occurrence parameter based on each candidate consecutive occurrence segment. Then, calculate the consecutive occurrence parameter for the first consecutive occurrence segment. The consecutive occurrence parameter corresponding to each candidate consecutive occurrence segment is written as: ,in, Indicates the duration. Indicates the number of occurrences. This represents the location parameter of the distribution. Depend on and Obtained by direct counting, taken from arrive The number of sampling points covered; Take directly ; Take directly Therefore, all three components of the consecutive occurrence parameter are obtained by directly reading or counting the underlying values ​​in the candidate consecutive occurrence segment.

[0145] The structural instability threshold is written as ,in, Indicates the duration threshold. Indicates the threshold for the number of occurrences. This represents the distribution location threshold. The duration threshold, occurrence frequency threshold, and distribution location threshold are all integer constants pre-written into the decision configuration table. Each consecutive occurrence parameter... With structural instability threshold Item-by-item comparison: when Not less than , Not less than ,and Not greater than At that time, determine the first The candidate consecutive occurrence segments satisfy the consecutive occurrence condition; when any one of the three comparisons does not satisfy the corresponding threshold condition, the first segment is determined. The candidate consecutive occurrence segments do not meet the consecutive occurrence condition. The judgment results corresponding to all candidate consecutive occurrence segments are written as a consecutive occurrence judgment result sequence. ,in, , This indicates that the condition of consecutive occurrence is met. This indicates that the condition for consecutive occurrences is not met.

[0146] A hazard assessment flag is generated based on a sequence of consecutive assessment results. If a sequence of consecutive assessment results appears... There exists at least one Then the first The hazard assessment flag corresponding to each transportation response segment is denoted as follows: If a sequence of judgment results appears consecutively All judgment results were Then the first The hazard assessment flag corresponding to each transportation response segment is denoted as: Hazard assessment signs It is a one-dimensional binary label that directly corresponds to whether the current continuous transportation period has reached the structural instability condition.

[0147] Step 7: When the hazard assessment indicator indicates that the continuous occurrence condition is met, output the hazard transportation assessment result of the original sample; when the hazard assessment indicator indicates that the continuous occurrence condition is not met, output the non-hazard transportation assessment result of the original sample.

[0148] Receive the danger judgment flag, and construct the result branch flag according to the judgment value corresponding to the danger judgment flag, so that the result branch flag corresponds to the condition of continuous occurrence and the condition of non-continuous occurrence;

[0149] The results are mapped to the continuous transportation time periods corresponding to the undisturbed sample. If the results indicate that the continuous occurrence condition is met, a dangerous transportation judgment result for the undisturbed sample is generated; if the results indicate that the continuous occurrence condition is not met, a non-dangerous transportation judgment result for the undisturbed sample is generated.

[0150] Use the hazard assessment flag generated in step 6 as the input for assessment. Step 7 uses only the hazard assessment flag. and the The continuous transportation time period corresponding to each transportation response segment Perform result branch construction and result mapping.

[0151] The continuous transportation time period corresponding to the original sample is written as... ,in, Indicates the first The starting time position of each transport response segment on a unified time index axis. Indicates the first The end time position of each transport response segment on a unified time index axis. Obtained without recalculation Directly from the first The time location range corresponding to each transport response segment is obtained when it is retained in step 1, and remains unchanged along the processing chain from step 2 to step 6. This is due to the hazard assessment flag. With the Each transport response fragment maintains the same index Therefore, danger assessment signs With continuous transportation time period They have a natural correspondence and do not require an additional matching process.

[0152] Receive hazard assessment sign Then, construct the result branch markers based on the judgment values ​​corresponding to the hazard judgment markers. The first... The result branch label corresponding to each transportation response segment is written as: ,in, .when At that time, Construct result branch labels to satisfy the condition of consecutive occurrence; when At that time, Construct result branch tags corresponding to those that do not meet the consecutive occurrence condition. Result branch tags It is only used to distinguish between two mutually exclusive decision branches, without introducing new thresholds, changing the original meaning of the hazard decision flag, or performing numerical conversion on the hazard decision flag. Therefore, the result branch flag and the hazard decision flag maintain the same index, the same value logic, and the same decision boundary.

[0153] After the result branch tags are constructed, the result branch tags will be... With continuous transportation time period Establish result mapping relationships. To ensure that the result mapping relationships have a field structure that can be directly recorded and retrieved, the first... The result mapping record corresponding to each transportation response segment is written as follows: .in, Indicates the result mapping record, the first field The second field indicates the start time and location of a continuous transportation period. The third field indicates the end time location of a continuous transportation period. Indicates the result branch marker. Result mapping record. for The judgment record has three fields, all of which are directly readable data fields. The result is mapped to the record. This allows the determination of branch categories to be stably bound to the continuous transportation time period corresponding to the original sample, so that the determination result of each transportation response segment has a clear time boundary.

[0154] When the result branch is marked This indicates that when the condition of consecutive occurrence is met, the result is mapped to a record. Generate the hazardous transport determination result for the original sample. (The first...) The hazardous transport determination result for the original sample corresponding to each transport response segment is written as follows: .in, This indicates the result of the hazardous transport judgment for the original sample. The first field indicates the start time and location, the second field indicates the end time and location, and the third field, with a judgment value of "1", indicates that the first field is the original sample. A continuous transportation time period corresponding to a transportation response segment is identified as hazardous transportation. The hazardous transportation identification result of the original sample directly inherits the time boundary in the result mapping record, without further segmenting the continuous transportation time period.

[0155] When the result branch is marked When the condition of consecutive occurrence is not met, the result is mapped to a record. Generate the non-hazardous transport determination result for the original sample. The non-hazardous transport determination result of the original sample corresponding to each transport response segment is written as follows: .in, This indicates the result of the non-hazardous transport determination for the original sample. The first field indicates the start time and location, the second field indicates the end time and location, and the third field, with a determination value of "0", indicates that the original sample was transported in its original state. The continuous transportation time period corresponding to each transportation response segment is determined to be non-hazardous transportation. The non-hazardous transportation determination result of the undisturbed sample adopts the same field structure as the hazardous transportation determination result of the undisturbed sample, with only the determination value of the third field being different.

[0156] For the same transport response fragment index Only according to generate or One of the judgment results will not generate two conflicting judgment results simultaneously. Therefore, step 7 will set the danger judgment flag. Convert the original sample transportation judgment record into a record with a clear boundary of continuous transportation time period, and keep the judgment record consistent in field structure.

[0157] This invention establishes a unified potential state generation mechanism for the transportation transmission chain, focusing on three types of realistically obtainable monitoring quantities in long-distance transportation scenarios: vibration of the transport container, force displacement of the shock-absorbing support, and constraint changes of the encapsulated soil sample tube. By mapping short-term impact information and long-term cumulative information to a transportation potential state vector at the same time location, it departs from the single-time-scale encoding, statistical compression, or single-channel anomaly extraction approaches used in existing time-series algorithms. This allows for the expression of instantaneous impacts and slow changes in contact boundaries within the same state semantics. Based on this structure, local mutations, gradual relaxation, and linked drift during transportation are no longer scattered and remain at the original measurement layer but can be transformed into continuously trackable intermediate state objects, thus establishing a state foundation that fits the scenario mechanism for hazardous transportation judgment.

[0158] By introducing AMP within the soft codebook, the state codewords are transformed from single-vector prototypes into orthogonal subspace state codewords. This allows for the unified inclusion of similar transmission patterns with different amplitudes and local drifts under the same contact boundary change mechanism. The attribution criteria are established through projection coefficients and subspace residuals, enabling a stable correspondence between potential transport state sequences and contact boundary change mechanisms during the classification process. The resulting state numbering results no longer merely reflect numerical similarity but also reflect similar transmission patterns under the coordinated action of the transport container, shock absorber support, and encapsulated soil sample tube. This is beneficial for maintaining consistency in hazardous state identification under complex road conditions and multiple sources of disturbance.

[0159] A hazardous state group is constructed around potential transport states that are continuously classified into the same orthogonal subspace state codeword. The continuous distribution, merging relationships, and persistent segments of this hazardous state group are further tracked, ultimately determining hazardous transport conditions at the structural instability threshold. By unifying state representation, consistent mechanism classification, and continuous occurrence determination along the same technical chain, the observable response to hazardous mechanisms during long-distance transport of undisturbed samples is gradually transformed into discriminable results. This facilitates the reliable identification of continuous hazardous states caused by changes in contact boundaries, even when relying solely on transport process monitoring data, and provides clear scenario-specificity and engineering implementation pathways for undisturbed sample transport risk determination.

[0160] Correspondingly, the present invention also provides a risk assessment system for undisturbed sample transportation based on vibration reduction theory, comprising:

[0161] The detection unit is used to acquire the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube, and to form a transport response segment sequence according to the continuous transport time period;

[0162] The computational unit is used to input the transport response fragment sequence into MTS-JEPA, perform joint embedding mapping on short-term impact scales and long-term cumulative scales to form a transport latent state sequence, and input the transport latent state sequence into a soft codebook.

[0163] The computing unit is also used to construct orthogonal subspace state codewords from each state codeword in the soft codebook using AMP to form an orthogonal subspace codebook, and classify the potential state sequence of transportation based on the orthogonal subspace codebook to obtain the transportation state number sequence, determine the same dangerous state group based on the transportation state number sequence to form a dangerous state change sequence, and form a danger judgment mark under the structural instability threshold according to the dangerous state change sequence.

[0164] The output unit is used to output the result of the dangerous transport judgment of the original sample based on the hazard judgment mark.

Claims

1. A method for determining the risk of transporting a sample in its original state based on the theory of shock absorption, characterized by, include: Step 1: Obtain the vibration of the transport container, the force and displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube, and form a transport response segment sequence according to the continuous transport time period; Step 2: Input the transport response fragment sequence into MTS-JEPA, perform joint embedding mapping on the short-term impact scale and the long-term cumulative scale to obtain the transport potential state sequence, and input the transport potential state sequence into the soft codebook. Step 3: In the soft codebook, AMP is used to construct each state codeword into an orthogonal subspace state codeword to form an orthogonal subspace codebook. The orthogonal subspace state codewords correspond to similar transmission patterns caused by the vibration of the transport box, the force displacement of the shock-absorbing support, and the change of the contact boundary of the sealed soil sample tube. Step 4: Classify the transportation potential state sequence based on the orthogonal subspace codebook to obtain the transportation state number sequence, and determine the transportation potential states that are continuously classified into the same orthogonal subspace state codeword as the same dangerous state group. Step 5: Track the continuous distribution of hazardous status groups based on the transport status number sequence to form a hazardous status change sequence; Step 6: Determine whether the continuous occurrence condition is met based on the sequence of dangerous state changes below the structural instability threshold, and form a dangerous judgment mark; Step 7: When the hazard assessment indicator indicates that the condition for continuous occurrence is met, output the hazard transportation assessment result of the original sample; When the hazard determination flag indicates that the condition for continuous occurrence is not met, the result of determining that the original sample is not a hazard transport is output.

2. The method of claim 1, wherein the method is characterized by, Step 1 is as follows: At the same sampling time, the vibration of the transport box, the force displacement of the shock-absorbing support, and the constraint change of the sealed soil sample tube are obtained respectively, and combined according to the positional order of the transport box, the shock-absorbing support, and the sealed soil sample tube in the transport transmission chain to form a transport state vector corresponding to the sampling time. The transportation state vectors corresponding to consecutive sampling times are arranged in chronological order to form a temporal state chain within a continuous transportation time period; A continuous transport state vector is extracted from the temporal state chain with a preset segment length. Each continuous transport state vector is spliced ​​together in chronological order, while keeping the transport chain order within each continuous transport state vector unchanged, to form a candidate transport segment. Based on the complete coverage relationship of continuous sampling times in the candidate transportation segments and the synchronous correspondence between the vibration of the transportation box, the force displacement of the shock-absorbing support, and the constraint changes of the encapsulated soil sample tube, the candidate transportation segments are retained to form a transportation response segment sequence.

3. The method according to claim 1, wherein MTS-JEPA includes: The input mapping layer is used to map the original input features to the basic temporal features through one-dimensional convolution; The local temporal extraction layer is used to extract local state features by concatenating one-dimensional convolution and ReLU; A dual-scale state generation layer is used to perform short-term impulsive scale convolution and long-term cumulative scale dilation convolution on local state features to generate short-term state features and long-term state features respectively. The transportation latent state generation layer is used to concatenate short-term state features and long-term state features at the same time location and then pass them through a fully connected mapping to obtain a transportation latent state sequence.

4. The method of claim 1, wherein the method is characterized by, Step 3 includes: Each state codeword in the soft codebook is determined as a state codeword to be constructed, and each state codeword to be constructed corresponds to a potential transportation state distribution range to be carried. Based on the temporal adjacency relationship, amplitude drift relationship, and linkage relationship of transportation box vibration, shock absorber support displacement and encapsulated soil sample tube constraint changes in the continuous transportation time period, similar transmission patterns caused by transportation box vibration, shock absorber support displacement and encapsulated soil sample tube contact boundary changes are extracted to form subspace construction constraints corresponding to each codeword to be constructed. Under the action of AMP, orthogonal basis construction is performed on each state codeword to be constructed according to the subspace construction constraints. Each state codeword to be constructed is updated from a single vector state codeword to a candidate orthogonal subspace state codeword composed of multiple mutually orthogonal basis vectors, so that each candidate orthogonal subspace state codeword carries the potential transport states with different amplitudes and different local drifts under the same contact boundary change mechanism. The potential state sequence of transportation is projected onto the state codewords of each candidate orthogonal subspace, and the projection coefficients and subspace residuals corresponding to each candidate orthogonal subspace state codeword are calculated to obtain the basis for classifying the state codewords that distinguish between similar and dissimilar transmission patterns. Based on the state codeword attribution criteria, the orthogonal basis direction and subspace coverage of each candidate orthogonal subspace state codeword are adjusted, and the candidate orthogonal subspace state codewords that meet the preset attribution conditions are determined as orthogonal subspace state codewords, so that the transport potential states belonging to the same orthogonal subspace state codeword correspond to the same contact boundary change mechanism. Write the state codewords of each orthogonal subspace into the soft codebook to form the orthogonal subspace codebook.

5. The as-lifted sample transportation risk assessment method based on shock attenuation theory according to claim 4, characterized by, In step 3, the preset attribution condition is that within the same local continuous window, the state codeword corresponding to a certain candidate orthogonal subspace state codeword of the same group of transportation potential states continuously obtains the maximum value, and the group of transportation potential states satisfies the time adjacency relationship, amplitude drift relationship and linkage change relationship under the same contact boundary change mechanism.

6. The as-is sample transportation risk determination method based on shock attenuation theory according to claim 4, characterized by, Step 4 includes: The transportation potential state sequence is input into the orthogonal subspace codebook in chronological order, and each transportation potential state in the transportation potential state sequence is taken as a state to be classified. Each state to be classified is projected onto the orthogonal subspace state codewords in the orthogonal subspace codebook. The projection coefficients and subspace residuals of each state to be classified relative to each orthogonal subspace state codeword are calculated to form the codeword matching results corresponding to each state to be classified. Based on the codeword matching results corresponding to each state to be classified, the attribution relationship between each state to be classified and each orthogonal subspace state codeword is compared to determine the target orthogonal subspace state codeword corresponding to each state to be classified, thus forming a state classification result corresponding to the time position. Based on the preset number of the state codeword of each target orthogonal subspace in the orthogonal subspace codebook, the state classification results are numbered and mapped to obtain a transportation state number sequence that is consistent with the time order of the transportation potential state sequence. The continuity of the numbering consistency of adjacent time positions in the transportation status numbering sequence is determined, and the potential transportation states that are continuously assigned to the same orthogonal subspace status codeword are extracted to form a continuous assignment result. Based on the continuous inclusion results, the potential transport states that are continuously included in the same orthogonal subspace state codeword are identified as the same dangerous state group.

7. The as-lifted sample transportation risk assessment method based on shock attenuation theory according to claim 6, characterized by, Step 5 includes: Based on the attribution relationship of the hazard group corresponding to each time position in the transport status number sequence, the transport status number sequence is scanned sequentially to extract the occurrence position of each hazard group in the continuous transport time period, forming a hazard group distribution sequence. Based on the interval relationship between adjacent time positions in the distribution sequence of hazardous state groups, consecutive occurrence segments belonging to the same hazardous state group are connected to form a continuous distribution segment of hazardous state groups. Based on the start and end positions, duration, and order of distribution of continuous distribution segments of hazardous state groups within a continuous transportation period, the continuous distribution of each hazardous state group is time-series tracked to form a hazardous state change sequence.

8. The as-lifted sample transportation risk assessment method based on shock attenuation theory according to claim 7, characterized by, Step 6 includes: Based on the start and end positions and adjacent intervals of the continuous distribution segments of each hazard state group in the hazard state change sequence, the hazard state change sequence is continuously merged to form candidate continuous occurrence segments; Calculate the consecutive occurrence parameter based on the duration, frequency, and distribution location of the candidate consecutive occurrence segment; The continuous occurrence parameter is compared with the structural instability threshold: the case where the continuous occurrence parameter meets the structural instability threshold is determined to meet the continuous occurrence condition; the case where the continuous occurrence parameter does not meet the structural instability threshold is determined to not meet the continuous occurrence condition; thus forming the continuous occurrence determination result; A hazard assessment flag is generated based on the consecutive occurrence of assessment results.

9. The as-lifted sample transportation risk assessment method based on shock attenuation theory according to claim 7, characterized by, Step 7 includes: Receive the danger judgment flag, and construct the result branch flag according to the judgment value corresponding to the danger judgment flag, so that the result branch flag corresponds to the condition of continuous occurrence and the condition of non-continuous occurrence; The results are mapped to the continuous transportation time periods corresponding to the undisturbed sample. If the results indicate that the continuous occurrence condition is met, a dangerous transportation judgment result for the undisturbed sample is generated; if the results indicate that the continuous occurrence condition is not met, a non-dangerous transportation judgment result for the undisturbed sample is generated.

10. A system for determining the risk of transporting a sample in its original state based on shock attenuation theory, characterized in that, include: The detection unit is used to acquire the vibration of the transport container, the force displacement of the shock-absorbing support, and the constraint changes of the sealed soil sample tube, and to form a transport response segment sequence according to the continuous transport time period; The computational unit is used to input the transport response fragment sequence into MTS-JEPA, perform joint embedding mapping on short-term impact scales and long-term cumulative scales to form a transport latent state sequence, and input the transport latent state sequence into a soft codebook. The computing unit is also used to construct orthogonal subspace state codewords from each state codeword in the soft codebook using AMP to form an orthogonal subspace codebook, and classify the potential state sequence of transportation based on the orthogonal subspace codebook to obtain the transportation state number sequence, determine the same dangerous state group based on the transportation state number sequence to form a dangerous state change sequence, and form a danger judgment mark under the structural instability threshold according to the dangerous state change sequence. The output unit is used to output the result of the dangerous transport judgment of the original sample based on the hazard judgment mark.

Images