Vehicle disassembly process intelligent scoring method and system

Abstract

The application provides a vehicle disassembly process intelligent scoring method and system, relates to the technical field of vehicle disassembly process, and comprises the following steps: generating a vehicle structure atlas and synchronously collecting real-time data flow of a disassembly station; combining a pre-stored benchmark disassembly path to execute an ant colony optimization algorithm, outputting an optimal disassembly path and path deviation degree; generating a process characteristic vector based on the real-time data flow and the optimal disassembly path; taking the process characteristic vector as input, performing real-time risk scoring according to a preset rule, and generating a dynamic risk score; generating a weight set according to the process characteristic vector and the dynamic risk score, and generating a process comprehensive score according to the weight set; when a dynamic condition is triggered, performing path evolution based on the path deviation degree and the process comprehensive score, and updating the pre-stored benchmark disassembly path according to the evolution result until the dynamic condition is met; and the application realizes safe and efficient optimal disassembly by comprehensively scoring disassembly models and dynamically optimizing disassembly paths and process strategies.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle disassembly process, and in particular to a vehicle disassembly process intelligent scoring method and system. BACKGROUND

[0002] The standardization and intelligentization of vehicle disassembly process are key links to ensure production safety, improve operation efficiency and quality control. However, the existing technology still has the following problems for the disassembly process management of new vehicles for the first time or vehicles on the market after long-term use:

[0003] Path planning is disconnected from the actual scene: traditional disassembly paths are usually based on theoretical models or historical experience in the design stage of the vehicle model, without fully considering the path adaptability caused by the actual manufacturing tolerances of new vehicles (such as component assembly errors, material batch differences), or the decline in path feasibility caused by long-term use of vehicles on the market (such as rusted connecting parts, torque decay, structural deformation). This static planning method easily leads to low disassembly efficiency, tool damage or secondary damage to parts.

[0004] Insufficient risk perception: existing disassembly process monitoring relies on a single sensor (such as a torque gun) or manual inspection, lacking correlation analysis of multi-dimensional data (such as operation timing, gas concentration, and component risk level). For example, excessive gas concentration may indicate a battery liquid leakage risk, and timing deviation may indicate an incorrect operation sequence, but traditional methods cannot capture such complex risks in real time, resulting in delayed warning of high-risk events (such as short circuits, structural fractures), threatening the safety of operators.

[0005] Lack of dynamic response in scoring mechanism: process scoring usually uses fixed weight rules or manual experience evaluation, without dynamically adjusting the scoring weights according to actual disassembly scene changes (such as component risk level upgrade, path frequent deviation). For example, torque anomalies on high-risk components should be given higher weights than similar events on low-risk components, but existing technology cannot achieve such adaptive scoring, resulting in a large deviation between scoring results and actual process quality.

[0006] Path optimization closed loop is missing: even if the disassembly path is found to be unreasonable, traditional technology lacks an adaptive evolution mechanism based on real-time data. The update of the benchmark path library relies on manual intervention or periodic reconfiguration, and cannot continuously adapt to new vehicle manufacturing process iterations (such as the use of lightweight materials) or changes in the aging characteristics of old vehicles (such as an increase in rusted connecting parts), resulting in low path optimization efficiency and high cost.

[0007] Therefore, it is necessary to provide a vehicle disassembly process intelligent scoring method and system to solve the above technical problems. SUMMARY

[0008] To solve the above technical problems, the application provides a vehicle disassembly process intelligent scoring method and system, which dynamically optimizes the disassembly path and process strategy through data-driven dynamic path optimization, multi-source risk quantitative evaluation and closed-loop evolution mechanism, and realizes safe and efficient optimal disassembly.

[0009] The application provides a vehicle disassembly process intelligent scoring method, which comprises the following steps:

[0010] A vehicle structure graph is generated based on vehicle topology data, and real-time data streams of disassembly stations are synchronously collected;

[0011] An ant colony optimization algorithm is executed based on the vehicle structure graph and a pre-stored benchmark disassembly path, and an optimal disassembly path and a path deviation degree are output, wherein the path deviation degree represents the difference degree between the pre-stored benchmark disassembly path and the optimal disassembly path;

[0012] The real-time data streams and the optimal disassembly path are taken as inputs, multi-dimensional feature extraction and risk marking are performed, and a process feature vector is generated;

[0013] The process feature vector is taken as an input, real-time risk scoring is performed according to a preset rule, and a dynamic risk score is generated;

[0014] A weight set is generated according to the process feature vector and the dynamic risk score, and a process comprehensive score is generated according to the weight set;

[0015] When the process comprehensive score triggers a dynamic condition, path evolution is performed based on the path deviation degree and the process comprehensive score, and the pre-stored benchmark disassembly path is updated according to the evolution result until the dynamic condition is met.

[0016] Preferably, the vehicle structure graph is generated based on vehicle topology data, and real-time data streams of disassembly stations are synchronously collected, comprising:

[0017] The VIN code of the vehicle is scanned, the standard vehicle structure graph and the component risk level library are obtained by associating the vehicle model database, the component risk level is mapped to the standard vehicle structure graph, and the vehicle structure graph is generated;

[0018] Targeted data collection is performed based on the vehicle structure graph, and real-time data streams including action video streams, gas concentration and torque values are obtained.

[0019] Preferably, the ant colony optimization algorithm is executed based on the vehicle structure graph and the pre-stored benchmark disassembly path, and the optimal disassembly path and the path deviation degree are output, comprising:

[0020] Initialize the pheromone concentration matrix based on the component risk level in the vehicle structure graph, and increase the pheromone weight of the corresponding topological element in the pheromone concentration matrix of the pre-stored benchmark disassembly path according to a preset proportion, and output a risk perception pheromone matrix;

[0021] Execute an ant colony optimization algorithm based on the risk perception pheromone matrix, output an optimal disassembly path, and calculate a path deviation degree based on the time and torque deviation of the pre-stored benchmark disassembly path.

[0022] Preferably, the execution of the ant colony optimization algorithm based on the risk perception pheromone matrix and the output of the optimal disassembly path comprise:

[0023] Input the risk perception pheromone matrix to construct a path selection probability model for component nodes in the vehicle structure graph, wherein the component risk level is positively correlated with the pheromone concentration;

[0024] Simulate the traversal of the ant colony through the vehicle structure graph based on the path selection probability model, dynamically calculate the transition probability of each component node, and select the next disassembly node;

[0025] After each ant traversal is completed, adjust the pheromone evaporation coefficient according to the path deviation degree feedback, and update the pheromone concentration of the corresponding disassembly path in the risk perception pheromone matrix;

[0026] Repeat the ant colony traversal and pheromone update until a preset condition is reached, and output the optimal disassembly path.

[0027] Preferably, the input of the real-time data stream and the optimal disassembly path, the multi-dimensional feature extraction and risk labeling, and the generation of the process feature vector comprise:

[0028] Align the action video stream timestamp of the pre-stored benchmark disassembly path based on the work step sequence of the optimal disassembly path, calculate the time sequence deviation of each work step, and output a work step time sequence deviation feature set;

[0029] Input the gas concentration and torque value of the real-time data stream:

[0030] When the gas concentration exceeds a preset concentration threshold, mark a gas exceeding event;

[0031] When the torque value deviates from the path standard value by a preset concentration threshold, mark a torque abnormal event;

[0032] In combination with the component risk label in the vehicle structure graph, if the marked event occurs on a component with a component risk level exceeding a preset level, upgrade it to a high-risk event label;

[0033] The final output includes a process feature vector of time sequence deviation, gas exceeding standard mark, torque abnormality mark and high-risk event mark.

[0034] Preferably, the process feature vector is inputted, real-time risk scoring is performed according to preset rules, a dynamic risk score is generated, including:

[0035] When the time sequence deviation exceeds a deviation threshold, a basic risk score is accumulated, when the deviation exceeds a continuous deviation threshold for a preset number of times, an additional risk score is accumulated, and a time sequence risk score is generated;

[0036] The event marks in the process feature vector are analyzed, and event risk score accumulation is performed to obtain an event risk score, specifically including:

[0037] Gas exceeding standard mark: triggering fixed risk score accumulation;

[0038] Torque abnormality mark: triggering fixed risk score accumulation;

[0039] High-risk event mark: triggering multiplied risk score accumulation;

[0040] The time sequence risk score and the event risk score are synthesized into a dynamic risk score, and a real-time alarm instruction is triggered when the dynamic risk score exceeds an alarm threshold.

[0041] Preferably, the weight set is generated according to the process feature vector and the dynamic risk score, and a process comprehensive score is generated according to the weight set, including:

[0042] When the frequency of the high-risk event mark exceeds a statistical benchmark value, the weight value of the high-risk event in the weight set is adjusted according to a first increase amplitude;

[0043] When the dynamic risk score reaches a preset threshold interval, the proportion of the time sequence deviation in the weight set is adjusted according to a second increase amplitude;

[0044] The path deviation degree is converted into a stability coefficient;

[0045] The adjusted dynamic risk score and the stability coefficient are fused to generate a process comprehensive score.

[0046] Preferably, when the process comprehensive score triggers a dynamic condition, path evolution is performed based on the path deviation degree and the process comprehensive score, and the pre-stored benchmark disassembly path is updated according to the evolution result until the dynamic condition is met, including:

[0047] When the process comprehensive score is lower than a preset evolution triggering threshold, it is determined that the dynamic condition is triggered;

[0048] When the dynamic condition is triggered, a variation rate is calculated according to the path deviation degree and the process comprehensive score, taking the optimal disassembly path as a parent;

[0049] Single-path adaptive evolution is performed based on the variation rate, a random disturbance is introduced to simulate a variation operation, and a candidate evolution path is output;

[0050] The stability coefficient of the candidate evolution path on the pre-stored benchmark disassembly path is evaluated, specifically including:

[0051] If the stability coefficient of the candidate evolution path exceeds the stability coefficient of the pre-stored benchmark disassembly path by a preset percentage, the candidate evolution path is marked as an effective evolution path;

[0052] If no effective evolution path is generated, a manual path optimization instruction is sent to an operation terminal;

[0053] The effective evolution path is updated as the pre-stored benchmark disassembly path;

[0054] The ant colony optimization algorithm, real-time risk score, process comprehensive score generation and path evolution process are repeatedly executed until the process comprehensive score is higher than the preset evolution trigger threshold, and it is determined that the dynamic condition is met.

[0055] The application also provides a vehicle disassembly process intelligent scoring system for executing the vehicle disassembly process intelligent scoring method.

[0056] A data acquisition module is configured to generate a vehicle structure graph based on vehicle topology data and synchronously acquire real-time data streams of disassembly stations;

[0057] An optimal path planning module is configured to execute an ant colony optimization algorithm based on the vehicle structure graph and a pre-stored benchmark disassembly path, and output an optimal disassembly path and a path deviation degree, wherein the path deviation degree represents a difference degree between the pre-stored benchmark disassembly path and the optimal disassembly path;

[0058] A comprehensive processing module is configured to take the real-time data streams and the optimal disassembly path as inputs, perform multi-dimensional feature extraction and risk marking, and generate a process feature vector;

[0059] A scoring module is configured to take the process feature vector as an input, perform real-time risk scoring according to a preset rule, and generate a dynamic risk score;

[0060] A comprehensive evaluation module is configured to generate a weight set according to the process feature vector and the dynamic risk score, and generate a process comprehensive score according to the weight set;

[0061] A path evolution module is configured to perform path evolution based on the path deviation degree and the process comprehensive score when the process comprehensive score triggers a dynamic condition, and update the pre-stored benchmark disassembly path according to the evolution result until the dynamic condition is met.

[0062] Compared with the related art, the vehicle disassembly process intelligent scoring method and system provided by the application has the following beneficial effects:

[0063] The application generates a structure graph based on vehicle topology data, initializes the pheromone matrix of the ant colony algorithm in combination with the component risk level, so that the optimal disassembly path is automatically adapted to new vehicle manufacturing tolerances (such as assembly error compensation) or old vehicle component aging characteristics (such as rusted connecting piece torque adjustment); the pheromone evaporation coefficient is dynamically adjusted through path deviation degree feedback to realize rapid convergence and optimization of the path, significantly improve disassembly efficiency and reduce tool damage risk.

[0064] The application synchronously collects real-time data such as action video, gas concentration, torque value, aligns the process step timing of the optimal disassembly path, accurately marks the timing deviation, gas exceeding the standard, torque anomaly and high-risk events (such as abnormal operation on high-risk components), and quantifies the risk level through a dynamic risk scoring model to realize millisecond-level early warning and operation intervention, and ensure personnel and equipment safety.

[0065] The application dynamically adjusts the proportion of each risk factor (such as increasing the weight of high-risk events or the weight of timing deviation) according to the frequency of high-risk events and the dynamic risk scoring interval, eliminates the subjectivity of manual scoring, ensures that the scoring result is strongly related to the actual process quality, and provides a reliable basis for process improvement.

[0066] When the process comprehensive score triggers the evolution condition, the application calculates the mutation rate based on the path deviation degree and the score, generates candidate paths through single-path adaptive evolution, and updates the benchmark library with paths having better stability coefficients to form a closed loop of "execution-evaluation-evolution", prolong the adaptation period of the technical solution to new vehicle process iteration and old vehicle aging scenarios, and reduce the cost of manual optimization. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 A flowchart of a vehicle disassembly process intelligent scoring method provided by the application;

[0068] Figure 2 A module structure diagram of a vehicle disassembly process intelligent scoring method provided by the application. DETAILED DESCRIPTION

[0069] The application will be described in further detail below with reference to the drawings and embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of the application. In addition, it should be noted that, for the sake of brevity, only structures related to the application are shown and described in the drawings. In addition, the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0070] In addition, it should be noted that, for the sake of brevity, only structures related to the application are shown and described in the drawings. Before discussing the example embodiments in more detail, it should be mentioned that some of the example embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts depict the operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The processes can be terminated when their operations are completed, but can also have additional steps not included in the drawings. The processes can correspond to methods, functions, procedures, subroutines, etc.

[0071] Embodiment one

[0072] The application provides a vehicle disassembly process intelligent scoring method, referring to FIG. 1, the method comprises the following steps: Figure 1 As shown in the figure, the method comprises the following steps:

[0073] S1: generating a vehicle structure graph based on vehicle topology data, and synchronously collecting real-time data streams of the disassembly station.

[0074] Specifically, step S1 comprises the following steps:

[0075] S11: scanning the vehicle VIN code, associating the vehicle model database to obtain a standard vehicle structure graph and a component risk level library, and mapping the component risk level to the standard vehicle structure graph to generate a vehicle structure graph.

[0076] In this embodiment, the vehicle VIN code is scanned by an industrial-grade code scanning gun, and a vehicle model database API (SOAP protocol interface) is called to query a standard vehicle structure graph, which contains the topology connection relationship and material properties (such as metal / plastic type) of each component.

[0077] A component risk level library (MySQL table) is synchronously called, and risk labels (a total of four levels: low / medium / high / very high) such as explosiveness (such as airbag) and toxicity (such as battery electrolyte) are matched according to the component ID. The risk level is mapped to the corresponding node of the vehicle structure graph: the high-risk node is marked as red, the medium-risk node is marked as yellow, and a three-dimensional vehicle structure graph with risk labels is output in JSON format, and stored in a local cache area.

[0078] S12: Targeted data acquisition based on the vehicle structure map, to obtain real-time data streams including action video streams, gas concentration, and torque values.

[0079] In this embodiment, sensor groups are deployed based on high-risk nodes (red markers) of the vehicle structure map, and the following data is acquired:

[0080] Action video stream acquisition:

[0081] A multi-view industrial camera is installed at the disassembly station, with a frame rate of 75 fps.

[0082] Based on the positions of high-risk components in the map, dynamic focusing of the corresponding cameras is triggered (e.g., automatically enabling the top camera when disassembling the airbag).

[0083] Gas concentration monitoring:

[0084] Electrochemical sensors are installed near high-risk nodes such as battery areas and fuel tank areas.

[0085] Real-time detection of H2S, CO, and other gas concentrations (range 0-500 ppm, accuracy ±1%).

[0086] Torque value synchronization recording:

[0087] A torque sensor is integrated into the pneumatic wrench, and when disassembling high-risk components marked on the map, the peak torque and duration are recorded.

[0088] All data is transmitted to the edge computing gateway through a three-wire CAN bus, aligned by timestamp, packaged into structured real-time data streams, and pushed to the central processing unit at 10 ms intervals.

[0089] S2: Perform an ant colony optimization algorithm based on the vehicle structure map and the pre-stored baseline disassembly path, output the optimal disassembly path and path deviation, where the path deviation represents the difference between the pre-stored baseline disassembly path and the optimal disassembly path.

[0090] Specifically, step S2 includes the following steps:

[0091] S21: Initialize the pheromone concentration matrix based on the component risk level in the vehicle structure map, and increase the pheromone weight of the corresponding topological elements of the pre-stored baseline disassembly path in the pheromone concentration matrix by a predetermined proportion, output the risk-aware pheromone matrix.

[0092] In this embodiment, first, the component risk level in the vehicle structure map is analyzed, and initial pheromone values are assigned to different risk levels: high-risk components (such as high-voltage battery packs) are set to 0.8, medium-risk components (such as cooling pipelines) are set to 0.5, and low-risk components (such as interior panels) are set to 0.2.

[0093] An n x n pheromone matrix is constructed based on the component topology relationship (n is the total number of components), wherein the diagonal elements are forced to be zero to prohibit self-connection. Then, a pre-stored benchmark disassembly path is loaded, and the topology connection pheromone value involved in the path is increased by 50% (for example, "high-voltage maintenance switch -> high-voltage battery pack" is increased from 0.8 to 1.2). Finally, a CSV format matrix is output and stored in a Redis database, while a check is performed to ensure that all high-risk node connection values are greater than or equal to 0.8 and the benchmark path enhancement is effective.

[0094] S22: performing an ant colony optimization algorithm based on the risk-aware pheromone matrix, outputting an optimal disassembly path, and calculating a path deviation degree based on the time and torque deviation of the pre-stored benchmark disassembly path.

[0095] In this embodiment, after obtaining the risk-aware pheromone matrix, an ant colony optimization algorithm is performed to output an optimal disassembly path, and a path deviation degree is calculated simultaneously. The deviation degree quantifies the difference between the optimal path and the pre-stored benchmark path, which is achieved by the following steps:

[0096] Initialize 50 ant threads, each starting from the starting node (such as the front bumper of the vehicle) and traversing based on a path selection probability model to dynamically select the next node. The iteration is performed for 200 times or terminated when the optimal path remains unchanged for 20 consecutive times, and then the path deviation degree is calculated, including:

[0097] Time deviation: calculating the absolute difference percentage between the total time consumption of the optimal path and the benchmark time;

[0098] Torque deviation: the sum of the absolute values of the torque difference of each step accounts for the proportion of the total benchmark torque;

[0099] The path deviation degree is obtained by weighted summation of the time deviation and the torque deviation, with weights of 0.6 and 0.4, respectively. For example, the optimal path takes 182 seconds (the benchmark time is 185 seconds), and the torque deviation accumulates 4 Nm (the total benchmark torque is 100 Nm), then the time deviation = 0.54%, the torque deviation = 4%, and the path deviation degree = 0.6 x 0.54% + 0.4 x 4% = 1.92%.

[0100] The execution of the ant colony optimization algorithm specifically includes:

[0101] S22a: constructing a path selection probability model for component nodes in the vehicle structure graph based on the risk-aware pheromone matrix, wherein the component risk level is positively correlated with the pheromone concentration.

[0102] In this embodiment, a component transition probability model is constructed, and the probability is calculated by the formula: .

[0103] wherein Represents a node To the node Pheromon concentration (from the risk perception pheromone matrix). The heuristic factor is defined as the target node. The risk level is at its lowest. This represents the pheromone weight index, with a base value of 1, which increases to 1.5 when the current node is a high-risk component. This represents the heuristic factor weighting index, with a fixed value of 2. This represents the set of neighboring nodes that are reachable from the current node and have not been demoted.

[0104] The transition probability of each neighboring node is calculated using this formula. For example, the transition probability from the high-voltage maintenance switch to the high-voltage battery pack is calculated as: (1.2^1.5×(1 / 3)^2) / the sum of the same calculation formulas for all neighboring nodes.

[0105] S22b: Based on the path selection probability model, simulate an ant colony traversing the vehicle structure map, dynamically calculate the transition probability of each component node, and select the next disassembly node.

[0106] In this embodiment, 50 ant threads are initialized for parallel traversal. Each ant starts from the entry node and selects the next node based on a probability model: first, the reachable neighbor list of the current node is obtained, filtered for nodes that have been dismantled, then the transition probability of each neighbor is calculated using a probability formula, and finally, the selected node is determined using a roulette wheel algorithm. The dismantling sequence, time cost, and torque value of each ant are recorded, where the time cost calculation formula is: .

[0107] S22c: After each ant traversal is completed, the pheromone evaporation coefficient is adjusted according to the path deviation feedback, and the pheromone concentration of the corresponding dismantling path in the risk perception pheromone matrix is ​​updated.

[0108] In this embodiment, after each ant completes its path traversal, a dynamic pheromone update is performed, comprising two stages: global evaporation and path reinforcement. First, the evaporation coefficient is dynamically adjusted based on the path deviation. When the deviation is <10% =0.05 (slow evaporation), 10-30% =0.10 (normal volatility), >30% =0.20 (rapid evaporation), and apply the evaporation formula. Achieve global pheromone attenuation. Then, strengthen the risk perception path using a modified formula. Calculate the enhancement amount, where pheromone intensity Dynamically assigned values ​​based on operating conditions (normal operating condition = 10, high-risk event = 15, training mode = 5), target node risk coefficient. According to the component risk level setting (high risk = 1.5, medium risk = 1.0, low risk = 0.8), The actual time consumption (seconds) of the ant path. The reinforcement logic embodies the high-risk priority principle, for example, the reinforcement amplitude is increased by 50% when disassembling the high-voltage battery pack ( =1.5) and reduced by 20% when disassembling the interior panel ( =0.8), while the path time consumption is shorter, the reinforcement value is greater.

[0109] S22d: Repeat the ant colony traversal and pheromone update until the preset condition is reached, and output the optimal disassembly path.

[0110] In this embodiment, a double convergence condition is set: the optimal path remains unchanged for 20 consecutive times or the path deviation degree ≤2% is maintained for 5 iterations. The optimal path is recorded in each iteration, and the evaluation index is: path score = basic score - deviation degree × 100. If the upper limit of 200 iterations is reached, the current optimal path is forced to output. The final output includes a structured report containing the path sequence, time cost (calculation formula: benchmark time × (1 + deviation degree)), torque deviation details, and the number of iterations.

[0111] S3: Perform multi-dimensional feature extraction and risk labeling based on the real-time data stream and the optimal disassembly path as input, and generate a process feature vector.

[0112] Specifically, step S3 includes the following steps:

[0113] S31: Align the action video stream time stamp of the pre-stored reference disassembly path based on the work step sequence of the optimal disassembly path, and calculate the time sequence deviation of each work step, and output the work step time sequence deviation feature set.

[0114] In this embodiment, the work step sequence of the optimal disassembly path is used as a reference, and the time stamp alignment algorithm is used to associate the real-time collected action video stream with the video time stamp of the pre-stored reference disassembly path. In specific implementation, first, extract the start time stamp and end time stamp of each work step in the optimal disassembly path to form a work step time interval; then, through video frame synchronization technology (such as key frame feature point matching or time stamp difference calculation), the action start frame in the real-time video stream is aligned with the time interval of the reference path. For each work step, calculate the absolute difference between the real-time action start time and the reference path start time, and the relative deviation proportion of the work step duration ((real-time duration - reference duration) / reference duration × 100%). Finally, output a feature set containing the absolute value and relative deviation proportion of the time sequence deviation of each work step.

[0115] S32: Take the gas concentration and torque value of the real-time data stream as input:

[0116] When the gas concentration exceeds the preset concentration threshold, mark the gas exceeding event;

[0117] When the torque value deviates from the path standard value by a preset concentration threshold, mark the torque abnormal event;

[0118] In combination with the component risk label in the vehicle structure map, if the marked event occurs on a component with a component risk level exceeding a preset level, it is upgraded to a high-risk event label.

[0119] In this embodiment, based on the gas concentration and torque value data in the real-time data stream, event marking is performed through a threshold comparison algorithm.

[0120] For gas concentration, the preset concentration threshold is set according to the component type and the disassembly environment safety standard (such as the carbon monoxide concentration threshold for disassembling the fuel system is set to 50 ppm); the real-time gas concentration data is compared with the preset concentration threshold in real time after being collected by the sensor, and if it exceeds the preset concentration threshold, it is marked as a "gas exceeding event" and the exceeding duration and peak concentration are recorded. For torque value, first extract the path standard torque value of each work step from the optimal disassembly path (such as the preset bolt disassembly torque is 20 N·m), then calculate the absolute deviation (real-time torque-standard torque) and relative deviation ratio (absolute deviation / standard torque x 100%) of real-time torque value and standard value; if the absolute deviation exceeds the preset concentration threshold (such as ±3 N·m) or the relative deviation exceeds the preset ratio (such as ±15%), it is marked as a "torque abnormal event" and the deviation direction (overrun or insufficient) and maximum deviation value are recorded.

[0121] S33: Finally output the process feature vector including time sequence deviation, gas exceeding label, torque abnormal label and high-risk event label.

[0122] In this embodiment, in combination with the component risk label in the vehicle structure map, risk upgrade judgment is performed on the marked gas exceeding event and torque abnormal event. In specific implementation, first extract the risk level of each disassembly component from the vehicle structure map (such as the fuel line component is marked as high risk and the ordinary bolt is marked as low risk), and the preset risk level threshold is set to "high risk" (which can be adjusted to medium risk or custom level according to actual needs). When the gas exceeding event or the torque abnormal event occurs on a component with a risk level exceeding the preset threshold, the event label is upgraded to "high-risk event", and a high-risk event label field is added to the process feature vector (such as marked as "1" indicating the existence of a high-risk event, and "0" indicating none);

[0123] The features generated in the above steps are structured and integrated to output the process feature vector. In specific implementation, the feature vector contains the following dimensions:

[0124] Timing deviation feature: absolute value and relative deviation proportion of timing deviation of each process step (array form, length consistent with process step number);

[0125] Gas exceeding standard flag: binary value (1 represents exceeding standard, 0 represents no);

[0126] Torque abnormality flag: binary value (1 represents abnormality, 0 represents no);

[0127] High-risk event flag: binary value (1 represents high-risk event, 0 represents no).

[0128] All features are processed by data standardization (such as timing deviation normalization to interval) and packaged as a fixed-dimension vector for subsequent steps.

[0129] S4: inputting the process feature vector, performing real-time risk scoring according to a preset rule, and generating a dynamic risk score.

[0130] Specifically, step S4 includes the following steps:

[0131] S41: when the timing deviation exceeds the deviation threshold, accumulate the basic risk score, when the deviation exceeds the continuous deviation threshold for a continuous preset number of times, accumulate the additional risk score, and generate a timing risk score.

[0132] In this embodiment, based on the process step timing deviation feature set in the process feature vector, the timing risk score is calculated according to the following rules:

[0133] Threshold setting: the preset absolute timing deviation threshold is 5 seconds (i.e., the difference between the real-time start time of any process step and the reference path start time exceeds 5 seconds), and the relative timing deviation threshold is 10% (i.e., the relative deviation proportion of the process step duration exceeds 10%).

[0134] Basic risk score accumulation: when the absolute timing deviation or the relative timing deviation of a single process step exceeds the corresponding threshold, accumulate the basic risk score 2 points / time.

[0135] Continuous deviation additional risk score: maintain a continuous deviation counter, when the timing deviation exceeds the threshold for the same process step for 2 consecutive times, the counter is incremented by 1; if the counter reaches 3 times (i.e., the same process step deviation exceeds the threshold for 3 consecutive times), an additional 5 points are accumulated, and the counter is reset.

[0136] Timing risk score output: the sum of the basic risk score and the additional risk score is taken as the timing risk score, ranging from 0 to 50 points (for example, a certain disassembly process has 4 single deviations and 1 continuous 3-time deviation, the timing risk score = 4 x 2 + 5 = 13 points).

[0137] S42: Analyze the event flag in the process feature vector and perform event risk score accumulation to obtain the event risk score, specifically including:

[0138] Gas over-limit flag: trigger fixed risk score accumulation;

[0139] Torque anomaly flag: trigger fixed risk score accumulation;

[0140] High-risk event flag: trigger multiplied risk score accumulation.

[0141] In this embodiment, the event flag field in the process feature vector is analyzed, and the event risk score is calculated according to the following rules:

[0142] Gas over-limit flag: every time a gas over-limit event (such as CO concentration > 50 ppm) is triggered, 10 points are fixedly accumulated, and every 10 seconds of over-limit duration is additionally added 2 points (for example, an over-limit event lasting 30 seconds scores 10 + 2 x 3 = 16 points).

[0143] Torque anomaly flag: every time a torque anomaly event (such as absolute deviation > 3 N·m or relative deviation > 15%) is triggered, 8 points are fixedly accumulated, and 3 points are additionally added when the deviation direction is over-torque (for example, an event of over-torque 4 N·m scores 8 + 3 = 11 points).

[0144] High-risk event flag: when a gas over-limit or torque anomaly event occurs on a high-risk component (risk level ≥ high), multiplied risk score is triggered, and the multiplication coefficient is 1.5 times (for example, a torque anomaly event of a high-risk component scores 8 x 1.5 = 12 points, and 3 points are additionally added when over-torque, for a total of 15 points).

[0145] Event risk score output: all event scores are accumulated, ranging from 0 to 100 points (for example, two gas over-limit events score 16 + 10 = 26 points, and a high-risk torque anomaly event scores 15 points, for a total event risk score = 26 + 15 = 41 points).

[0146] S43: Synthesize the time sequence risk score and the event risk score into a dynamic risk score, and trigger a real-time alarm instruction when the dynamic risk score exceeds an alarm threshold.

[0147] In this embodiment, the dynamic risk score synthesis: a weighted summation model is adopted, the time sequence risk score weight is 0.3, the event risk score weight is 0.7, and the calculation formula is: dynamic risk score = time sequence risk score x 0.3 + event risk score x 0.7 (for example, time sequence risk score 13 points, event risk score 41 points, dynamic risk score = 13 x 0.3 + 41 x 0.7 ≈ 3.9 + 28.7 = 32.6 points).

[0148] Alarm threshold determination: preset three-level alarm thresholds:

[0149] Yellow warning (≥ 50 points): Trigger interface pop-up prompt "Risk is rising, please pay attention to operation";

[0150] Orange alert (≥ 70 points): Trigger sound alarm and pause disassembly workstation operation;

[0151] Red alert (≥ 90 points): Lock the disassembly device and notify the safety officer to intervene.

[0152] Alarm instruction execution: When the dynamic risk score exceeds any threshold, the system sends an alarm instruction to the operation terminal through the MQTT protocol, including score details, trigger event list and suggested treatment measures (such as "check torque sensor calibration" or "emergency ventilation").

[0153] S5: Generating a weight set according to the process feature vector and the dynamic risk score, and generating a process comprehensive score according to the weight set.

[0154] Specifically, step S5 includes the following steps:

[0155] S51: When the frequency of the high-risk event marker exceeds the statistical reference value, adjust the weight value of the high-risk event in the weight set according to the first increase amplitude.

[0156] In this embodiment, the dynamic adjustment of the high-risk event weight includes:

[0157] Statistical reference value setting: Based on historical disassembly data (in the last 30 days, ≥ 100 disassembly records), calculate the average frequency (times / hour) of high-risk event markers, and set it as the statistical reference value (for example, the average frequency is 1.5 times / hour).

[0158] Frequency monitoring and trigger determination: Real-time statistics of the number of high-risk event markers in the current disassembly period, calculate the frequency (total number / time length). When the real-time frequency exceeds the statistical reference value (such as ≥ 2 times / hour), trigger the weight adjustment mechanism.

[0159] First increase amplitude calculation: Use a linear increasing model, adjustment amplitude=(real-time frequency-reference value)×0.05 (for example, real-time frequency 2.5 times / hour, reference value 1.5 times / hour, adjustment amplitude=1×0.05=0.05).

[0160] Weight value adjustment: The initial weight of high-risk event in the weight set is 0.3 (default value), the adjusted weight=initial weight+adjustment amplitude (such as 0.3+0.05=0.35), while other weights (time sequence deviation, stability coefficient) are compressed in proportion to ensure the total weight sum is 1.

[0161] S52: When the dynamic risk score reaches the preset threshold interval, adjust the proportion of time sequence deviation in the weight set according to the second increase amplitude.

[0162] In this embodiment, the timing deviation weight dynamic adjustment includes:

[0163] Preset threshold interval setting: set the threshold interval of the dynamic risk score according to the risk level, for example:

[0164] Low risk area (<50 points): timing deviation weight 0.2;

[0165] Medium risk area (50-70 points): timing deviation weight 0.3;

[0166] High risk area (>70 points): timing deviation weight 0.4.

[0167] Second increase amplitude calculation: when the dynamic risk score enters the medium risk area, the timing deviation weight is increased from 0.2 to 0.3 (increase amplitude +0.1); when it enters the high risk area, the weight is further increased to 0.4 (increase amplitude +0.2).

[0168] Weight proportion adjustment: realize the weight transition through linear interpolation, for example, when the score increases from 49 points to 51 points, the timing deviation weight gradually increases from 0.2 to 0.3, avoiding sudden changes.

[0169] S53: Convert path deviation degree into stability coefficient.

[0170] In this embodiment, the stability coefficient function is defined: a nonlinear decay model is used, stability coefficient = 1-(path deviation degree x s), where s is the decay factor (s=0.02). For example, when the path deviation degree is 2%, the stability coefficient is 1-0.02x2=0.96.

[0171] Boundary constraint: the minimum value of the stability coefficient is not less than 0.5 (when the path deviation degree is ≥25%, it is forced to be 0.5), ensuring that the basic stability evaluation is still retained in the case of extreme deviation.

[0172] S54: Fuse the adjusted dynamic risk score and the stability coefficient to generate the process comprehensive score.

[0173] In this embodiment, the weight set fusion: the adjusted dynamic risk score (weight proportion 0.6) and the stability coefficient (weight proportion 0.4) are substituted into the weighted summation formula:

[0174] Process comprehensive score = dynamic risk score x 0.6 + stability coefficient x 100 x 0.4

[0175] (Note: the stability coefficient is a decimal number between 0 and 1, which needs to be multiplied by 100 to convert it to the percentage system)

[0176] For example, dynamic risk score = 65 points, stability coefficient = 0.85, then process comprehensive score = 65 x 0.6 + 85 x 0.4 = 39 + 34 = 73 points.

[0177] Score range correction: if the comprehensive score exceeds 100 points, it is forced to be 100 points; if it is less than 0 points, it is set to 0 points to avoid numerical overflow.

[0178] S6: When the process comprehensive score triggers a dynamic condition, perform path evolution based on the path deviation and process comprehensive score, and update the pre-stored benchmark disassembly path according to the evolution result until the dynamic condition is met.

[0179] Specifically, step S6 includes the following steps:

[0180] S61: When the process comprehensive score is lower than a preset evolution trigger threshold, it is determined that the dynamic condition is triggered.

[0181] In this embodiment, when the preset evolution trigger threshold, first extract the process comprehensive score of the last 60 disassembly tasks from the database, calculate the average (μ) and standard deviation (σ), and set the trigger threshold to μ-1.5σ (for example, if μ=80 points, σ=5 points, then the threshold =72.5 points), and the minimum trigger threshold is not less than 50 points. Real-time monitoring, maintain a sliding window of length 10, real-time calculation of the average value of the scores in the window, if the average value is lower than the threshold for 3 times in a row, or a single score is lower than the threshold and the high-risk event frequency is more than 1.5 times the historical average frequency, then it is determined that the dynamic condition is triggered.

[0182] S62: When the dynamic condition is triggered, take the optimal disassembly path as the parent, and calculate the mutation rate according to the path deviation and process comprehensive score.

[0183] In this embodiment, the mutation rate (M) is calculated by the formula M= path deviation (D) x 0.3 + (statistical benchmark value (B) - process comprehensive score (S)) x 0.05 (for example, D=5%, B=80 points, S=65 points, M=2.25%). If the path deviation is >10%, the weight 1 is increased to 0.4; if the process comprehensive score is <60 points, the weight 2 is increased to 0.08.

[0184] The final mutation rate is limited by the function Constrained to a reasonable range, the default mutation rate is set to 8% under abnormal data.

[0185] S63: Perform single-path adaptive evolution based on the mutation rate, simulate mutation operations by introducing random perturbations, and output candidate evolution paths.

[0186] In this embodiment, the optimal disassembly path is taken as the parent, 20% of the steps are randomly selected for sequential exchange (e.g. "disassemble the tire" is adjusted to "disassemble the brake disc"), and the torque threshold, time window and other parameters are randomly offset by ±5%. The mutated path needs to pass the topological connection check (e.g. "the engine must be disconnected before disassembly") and the safety rule check (e.g. the high-voltage battery pack is discharged before disassembly), and if the check fails, it will be rolled back to the parent path. Finally, 10 candidate paths are generated, of which 3 are low mutation rate (M=1-3%), 5 are medium mutation rate (M=4-8%), and 2 are high mutation rate (M=9-15%).

[0187] S64: Evaluate the stability coefficient of the candidate evolution path to the pre-stored reference disassembly path, specifically including:

[0188] If the stability coefficient of the candidate evolution path exceeds the pre-stored reference disassembly path by a preset percentage, mark the candidate evolution path as an effective evolution path;

[0189] If no effective evolution path is generated, send a manual path optimization instruction to the operation terminal.

[0190] In this embodiment, the path deviation degree is calculated for each candidate path in S22, and the stability coefficient is converted (the method is the same as S53). If the stability coefficient of the candidate path exceeds 1.1 times the pre-stored reference path (e.g. the reference coefficient is 0.8, the candidate needs to be ≥0.88), it is marked as effective; if no effective path is generated for 3 consecutive times, the threshold is reduced to 1.05 times. If it is still invalid, send a manual optimization instruction to the operation terminal through the MQTT protocol, including path ID, mutation point and engineer adjustment interface.

[0191] S65: Update the effective evolution path as the pre-stored reference disassembly path.

[0192] In this embodiment, the exponential weighted average method is used to update the reference path parameters for the effective candidate path (e.g. the new path time cost accounts for 60%, and the original path accounts for 40%), and the torque threshold, step sequence and other parameters are updated locally according to the mutation point. The updated path is stored in JSON format in the MySQL database and synchronized to the Redis cache area of the edge computing gateway to ensure that the next task can be directly called.

[0193] S66: Repeat the ant colony optimization algorithm, real-time risk scoring, process comprehensive scoring generation and path evolution process until the process comprehensive score is higher than the preset evolution trigger threshold, and determine that the dynamic condition is met.

[0194] In the present embodiment, the process comprehensive score is regenerated after each path update. If the score is higher than the preset threshold (e.g., ≥72.5) for 5 consecutive times, it is determined that the dynamic condition is met. If the iteration number reaches 50 times and still does not meet the standard, the process is forcibly terminated, the candidate path with the highest stability coefficient is retained, and an alarm is sent to the administrator through SMS / email, containing the path ID, stability coefficient and last score, requesting the engineer to intervene in optimization.

[0195] Embodiment Two

[0196] The present application also provides a vehicle disassembly process intelligent scoring system for executing the vehicle disassembly process intelligent scoring method. Figure 2 As shown in the figure, the system comprises:

[0197] The data acquisition module 100 is configured to generate a vehicle structure map based on vehicle topology data and synchronously acquire real-time data streams of the disassembly stations.

[0198] The optimal path planning module 200 is configured to execute an ant colony optimization algorithm based on the vehicle structure map and the pre-stored benchmark disassembly path, and output an optimal disassembly path and a path deviation degree, wherein the path deviation degree represents the difference between the pre-stored benchmark disassembly path and the optimal disassembly path.

[0199] The comprehensive processing module 300 is configured to input the real-time data streams and the optimal disassembly path, perform multi-dimensional feature extraction and risk labeling, and generate a process feature vector.

[0200] The scoring module 400 is configured to input the process feature vector, perform real-time risk scoring according to a preset rule, and generate a dynamic risk score.

[0201] The comprehensive evaluation module 500 is configured to generate a weight set according to the process feature vector and the dynamic risk score, and generate a process comprehensive score according to the weight set.

[0202] The path evolution module 600 is configured to execute path evolution based on the path deviation degree and the process comprehensive score when the process comprehensive score triggers a dynamic condition, and update the pre-stored benchmark disassembly path according to the evolution result until the dynamic condition is met.

[0203] The present application is described with reference to the accompanying drawings, which are meant to be exemplary and illustrative and not limiting in any way. The present application is described with reference to flowcharts and / or block diagrams illustrating the architecture, functionality, and operation of possible implementations of methods, apparatus (systems) and computer program products according to the present application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more of the flow or blocks Figure 1 one or more of the flow or blocks

[0204] It is understood by those skilled in the art that all or a portion of the steps of various methods described above can be performed by program instructions. Such program instructions can be stored in a computer readable medium including a storage media having stored thereon instructions which can be used to program a computer to perform a process. The computer readable medium can include, but is not limited to, floppy diskettes, optical disks, CD-ROMs, and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, magnet or optical cards, or other type of media suitable for storing electronic instructions. Moreover, it is understood that the disclosure can be practiced with additional data storage devices, including database(s) and data library(s), or in additional data transmission mediums, including network(s) and / or data line(s).

[0205] It is further noted that the terms "comprise" "comprising" or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without further constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

Claims

1. A vehicle teardown process intelligence scoring method, characterized in that, The method comprises the following steps: generating a vehicle structure graph based on vehicle topology data, and synchronously collecting real-time data streams of the disassembly stations; performing an ant colony optimization algorithm based on the vehicle structure graph and a pre-stored benchmark disassembly path, and outputting an optimal disassembly path and a path deviation degree, wherein the path deviation degree represents the difference between the pre-stored benchmark disassembly path and the optimal disassembly path; Specifically, the execution of the ant colony optimization algorithm comprises: initializing a pheromone concentration matrix based on the component risk level in the vehicle structure graph, and increasing the pheromone weight of the pre-stored benchmark disassembly path in the pheromone concentration matrix according to a preset proportion, and outputting a risk-aware pheromone matrix; performing the ant colony optimization algorithm based on the risk-aware pheromone matrix, outputting the optimal disassembly path, and calculating the path deviation degree based on the time and torque deviation of the pre-stored benchmark disassembly path; inputting the real-time data stream and the optimal disassembly path, performing multi-dimensional feature extraction and risk labeling, and generating a process feature vector; Specifically, the generation of the process feature vector comprises: aligning the action video stream timestamp of the pre-stored benchmark disassembly path based on the work step sequence of the optimal disassembly path, calculating the time sequence deviation of each work step, and outputting a work step time sequence deviation feature set; inputting the gas concentration and torque value of the real-time data stream: when the gas concentration exceeds a preset concentration threshold, mark a gas over-standard event; when the torque value deviates from the path standard value by a preset concentration threshold, mark a torque abnormal event; combined with the component risk label in the vehicle structure graph, if the marked event occurs on a component whose risk level exceeds a preset level, upgrade it to a high-risk event label; finally output the process feature vector including the time sequence deviation, gas over-standard label, torque abnormal label and high-risk event label; inputting the process feature vector, performing real-time risk scoring according to a preset rule, and generating a dynamic risk score; Specifically, the generation of the dynamic risk score comprises: when the time sequence deviation exceeds a deviation threshold, accumulate a basic risk score, and when the deviation exceeds a continuous deviation threshold for a continuous preset number of times, accumulate an additional risk score, and generate a time sequence risk score; analyze the event labels in the process feature vector and accumulate event risk scores to obtain an event risk score, specifically including: gas over-standard label: trigger fixed risk score accumulation; torque abnormal label: trigger fixed risk score accumulation; high-risk event label: trigger multiplication risk score accumulation; synthesize the time sequence risk score and the event risk score into a dynamic risk score, and trigger a real-time alarm instruction when the dynamic risk score exceeds an alarm threshold; generate a weight set according to the process feature vector and the dynamic risk score, and generate a process comprehensive score according to the weight set; Specifically, the generation of the process comprehensive score comprises: when the frequency of the high-risk event label exceeds a statistical benchmark value, adjust the weight value of the high-risk event in the weight set according to a first increase amplitude; when the dynamic risk score reaches a preset threshold interval, adjust the proportion of the time sequence deviation in the weight set according to a second increase amplitude; convert the path deviation degree into a stability coefficient; fusing the adjusted dynamic risk score and the stability coefficient to generate a process comprehensive score; when the process comprehensive score triggers a dynamic condition, performing path evolution based on the path deviation degree and the process comprehensive score, and updating the pre-stored benchmark disassembly path according to the evolution result until the dynamic condition is met.

2. The vehicle teardown process intelligence scoring method of claim 1, wherein, the vehicle structure graph is generated based on the vehicle topology data, and real-time data streams of the disassembly station are synchronously collected, including: scanning a vehicle VIN code, associating with a vehicle model database to obtain a standard vehicle structure graph and a component risk level library, and mapping the component risk level to the standard vehicle structure graph to generate the vehicle structure graph; targeted data collection is performed based on the vehicle structure graph to obtain real-time data streams including action video streams, gas concentration and torque values.

3. The vehicle teardown process intelligence scoring method of claim 2, wherein, the ant colony optimization algorithm is executed based on the risk perception pheromone matrix to output an optimal disassembly path, including: taking the risk perception pheromone matrix as input, a path selection probability model for component nodes in the vehicle structure graph is constructed, wherein the component risk level is positively correlated with the pheromone concentration; based on the path selection probability model, an ant colony is simulated to traverse the vehicle structure graph, the transition probability of each component node is dynamically calculated, and the next disassembly node is selected; after each ant traversal is completed, the pheromone evaporation coefficient is adjusted according to the path deviation degree feedback, and the pheromone concentration of the corresponding disassembly path in the risk perception pheromone matrix is updated; the ant colony traversal and pheromone update are repeatedly executed until a preset condition is reached, and the optimal disassembly path is output.

4. The vehicle teardown process intelligence scoring method of claim 3, wherein, when the process comprehensive score triggers a dynamic condition, performing path evolution based on the path deviation degree and the process comprehensive score, and updating the pre-stored benchmark disassembly path according to the evolution result until the dynamic condition is met, including: when the process comprehensive score is lower than a preset evolution trigger threshold, it is determined that the dynamic condition is triggered; when the dynamic condition is triggered, taking the optimal disassembly path as the parent, calculating the mutation rate according to the path deviation degree and the process comprehensive score; based on the mutation rate, single-path adaptive evolution is performed, a candidate evolution path is output by introducing random disturbance to simulate mutation operation; the stability coefficient of the candidate evolution path to the pre-stored benchmark disassembly path is evaluated, specifically including: if the stability coefficient of the candidate evolution path exceeds the preset percentage of the stability coefficient of the pre-stored benchmark disassembly path, the candidate evolution path is marked as an effective evolution path; if no effective evolution path is generated, an artificial path optimization instruction is sent to an operation terminal; the effective evolution path is updated as the pre-stored benchmark disassembly path; the ant colony optimization algorithm, real-time risk score, process comprehensive score generation and path evolution process are repeatedly executed until the process comprehensive score is higher than the preset evolution trigger threshold, and it is determined that the dynamic condition is met.

5. A vehicle disassembly process intelligent scoring system for implementing a vehicle disassembly process intelligent scoring method according to any one of claims 1 to 4, characterized in that, the system includes: a data collection module for generating a vehicle structure graph based on vehicle topology data and synchronously collecting real-time data streams of a disassembly station; An optimization path planning module is configured to execute an ant colony optimization algorithm based on the vehicle structure graph and a pre-stored benchmark disassembly path, and output an optimal disassembly path and a path deviation degree, wherein the path deviation degree represents a difference degree between the pre-stored benchmark disassembly path and the optimal disassembly path; A comprehensive processing module is configured to perform multi-dimensional feature extraction and risk labeling by taking the real-time data stream and the optimal disassembly path as inputs, and generate a process feature vector; A scoring module is configured to perform real-time risk scoring according to a preset rule by taking the process feature vector as an input, and generate a dynamic risk score; A comprehensive evaluation module is configured to generate a weight set according to the process feature vector and the dynamic risk score, and generate a process comprehensive score according to the weight set; A path evolution module is configured to execute path evolution based on the path deviation degree and the process comprehensive score when the process comprehensive score triggers a dynamic condition, and update the pre-stored benchmark disassembly path according to an evolution result until the dynamic condition is met.

Images