Vehicle mutual-aid claim settlement method and system based on multi-modal data fusion

Abstract

The invention provides a vehicle mutual-aid claim settlement method and system based on multi-modal data fusion. Wherein vision and depth sensing data and maintenance records in a vehicle accident are integrated, part types and collision mechanical parameters are extracted, and surface microscopic deformation and internal hidden cracks are captured through dual-band scanning; and carrying out time-space association on the two to construct a comprehensive damage map. And generating a standard damage sequence by using the generative adversarial network based on a normal damage rule of the component, comparing the standard damage sequence with accident damage data layer by layer, extracting an abnormal region deviating from an evolution rule, and calculating a fraud risk index according to the spatial density of the abnormal region. And when the index exceeds a dynamic threshold value, triggering multi-source data to cooperatively verify feature consistency, and outputting a claim settlement result with a fraud mark. According to the technical scheme provided by the invention, accurate identification of maintenance fraud and collaborative optimization of claim settlement risk control efficiency are realized.

Description

Technical Field

[0001] This application relates to the technical field of vehicle mutual insurance claims settlement, and particularly to a vehicle mutual insurance claims settlement method and system based on multi-modal data fusion. Background Art

[0002] In the vehicle mutual insurance claims settlement scenario, problems such as difficult tracing of the damage causal chain, strong concealment of new fraud means, and lag in updating verification rules need to be solved. The method is required to: achieve a physical logic closed-loop verification of the damage formation mechanism and the rationality of repair operations; identify the correlation of fraud behaviors across time dimensions (such as repeated use of historical damages and abnormal timing of parts replacement); automatically optimize the verification strategy based on dynamic risk characteristics (such as the diffusion speed of new fraud technologies), generate traceable fraud determination bases, and ensure the timeliness and accuracy of claims settlement decisions.

[0003] The current mainstream solution adopts a physically driven accident simulation enhanced detection framework: simulating the vehicle damage propagation path under different collision conditions through a finite element simulation model to generate a theoretical damage distribution heat map; using a contrast learning mechanism to perform difference analysis on the actual damage detection data (such as 3D scan point cloud) and the simulation results to locate abnormal damage areas beyond physical laws; designing a fraud mode evolution map, combining the tool usage sequences in the repair shop equipment operation logs (such as the correlation between the working hours of the spray gun and the paint thickness), and identifying illegal operation nodes through graph embedding technology to output a fraud probability score.

[0004] However, the finite element simulation model has insufficient modeling accuracy for non-linear material deformation (such as local brittle fracture of composite materials), resulting in the deviation of the theoretical damage heat map from the actual detection data; the contrast learning mechanism relies on high-precision detection equipment, but some repair shops use low-resolution scanners, causing data noise interference and leading to mislocation of abnormal areas. In addition, the fraud mode evolution map can only model explicit operation chains and cannot identify hidden fraud behaviors caused by black box tools (such as illegally flashing ECU data), resulting in an increased missed detection rate of new electronic fraud means. Summary of the Invention

[0005] This application provides a vehicle mutual insurance claims settlement method and system based on multi-modal data fusion to solve the problems of insufficient fraud detection accuracy and poor multi-source data collaboration in the prior art.

[0006] In a first aspect, this application provides a vehicle mutual insurance claims settlement method based on multi-modal data fusion, including: Integrating the damage data composed of visual sensors, depth sensing units, and repair records in the vehicle accident scenario, extracting the types of vehicle components from the repair records, and obtaining the mechanical parameters of the accident collision through a depth detector; Perform dual - band scanning on the damage features of different dimensions in the damage data to capture the microscopic deformation features of surface damage and the hidden crack distribution of internal damage; Perform spatio - temporal correlation on the microscopic deformation features and the hidden crack distribution to form a comprehensive damage atlas including the correlation degree between surface damage and internal damage; Obtain the normal damage evolution law of vehicle components in a non - fraud scenario, and generate a standard damage sequence corresponding to the comprehensive damage atlas in combination with a generative adversarial network; Compare the standard damage sequence with the damage data layer by layer, extract the abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculate the fraud risk index based on the spatial density of the abnormal damage areas on the vehicle components; When the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of accident collisions, trigger the collaborative verification of multi - source data, and output a claim settlement determination result including a repair fraud mark by verifying the feature consistency of the damage data in the abnormal damage areas.

[0007] Optionally, comparing the standard damage sequence with the damage data layer by layer, extracting the abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculating the fraud risk index based on the spatial density of the abnormal damage areas on the vehicle components, includes: Compare the damage morphology change paths of vehicle components in the damage data with the damage morphology change paths of the standard damage sequence layer by layer. The layer - by - layer comparison divides the damage area of the vehicle into multiple sub - areas, and layer by layer compares the damage diffusion direction and the similarity of damage morphology within each sub - area; Based on the results of the layer - by - layer comparison, extract the sub - areas in the damage data where the deviation of the damage diffusion direction from the standard damage sequence exceeds a preset angle threshold, mark the sub - areas as deviation areas, and calculate the abnormal damage features with a damage morphology similarity lower than a preset similarity threshold within each deviation area; Statistically analyze the distribution positions and coverage ranges of the deviation areas on the vehicle components, and merge adjacent deviation areas according to the continuity of the damage diffusion direction to obtain a set of abnormal damage areas; Calculate the spatial density parameter of the abnormal damage area according to the coverage area of each deviation area in the abnormal damage area set and the total area of the corresponding component. The spatial density parameter is proportional to the coverage area; Perform weighted superposition on the spatial density parameter and the deviation amplitude of the damage morphology similarity within the deviation area to obtain a superposition value as the fraud risk index.

[0008] Optionally, obtain the normal damage evolution law of vehicle components in a non-fraud scenario, and generate a standard damage sequence corresponding to the comprehensive damage map in combination with a generative adversarial network, including: Obtain the normal damage evolution law of vehicle components in a non-fraud scenario, where the normal damage evolution law is obtained by integrating the type of vehicle components, the mechanical parameters of accident collisions, and the corresponding damage morphology change path, and the damage morphology change path records the whole process of the starting position, diffusion direction, and morphological branch change of damage diffusion of vehicle components after collision; Based on the correspondence between the type and mechanical parameters of vehicle components in the normal damage evolution law, combine the generator of the generative adversarial network to generate candidate damage diffusion rules for different types of vehicle components under specific mechanical parameters; Screen the candidate damage diffusion rules through the discriminator of the generative adversarial network, and retain the rules that conform to the spatio-temporal distribution characteristics of the damage morphology change path in the non-fraud scenario, and integrate them into dynamic generation rules; Input the starting position and diffusion direction in the comprehensive damage map into the dynamic generation rules, and combine the type of vehicle components to match the corresponding mechanical parameters to generate a standard damage sequence consistent with the damage diffusion direction and morphological branch change in the comprehensive damage map.

[0009] Optionally, input the starting position and diffusion direction in the comprehensive damage map into the dynamic generation rules, and combine the type of vehicle components to match the corresponding mechanical parameters to generate a standard damage sequence consistent with the damage diffusion direction and morphological branch change in the comprehensive damage map, including: Input the starting position coordinates of surface damage and the diffusion direction angle of internal damage in the comprehensive damage map into the dynamic generation rules, and the dynamic generation rules include the angle tolerance range of the damage diffusion direction and the change threshold of the number of morphological branches for different types of vehicle components under specific mechanical parameters; According to the type of vehicle components in the comprehensive damage map, match the corresponding mechanical parameter combinations from the dynamic generation rules, and the mechanical parameter combinations include collision angle, speed, and force range, which are used to constrain the generation path of the damage diffusion direction and morphological branches; Based on the matched mechanical parameter combinations, generate a standard damage sequence that is consistent with the starting position coordinates of surface damage in the comprehensive damage map and conforms to the diffusion direction angle of internal damage, and the number of morphological branches of the standard damage sequence is limited by the change threshold in the dynamic generation rules.

[0010] Optionally, perform dual-band scanning on the damage features of different dimensions in the damage data to capture the microscopic deformation features of surface damage and the hidden crack distribution of internal damage, including: Perform dual - band scanning on the surface damage area in the damage data and interrupt the first - band scanning. The first - band scanning focuses on the microscopic deformation details of the surface damage by adjusting the resolution parameters of the detection device, highlighting the deformation contour features of the surface damage and the texture differences in the deformation area, so as to extract the microscopic deformation features of the surface damage. Perform the second - band scanning in the dual - band scanning on the internal damage area in the damage data. The second - band scanning focuses on the hidden crack propagation path of the internal damage by adjusting the penetration parameters of the detection device, capturing the intermittent connection features of the internal cracks, so as to extract the distribution of the hidden cracks in the internal damage.

[0011] Optionally, perform spatio - temporal correlation on the microscopic deformation features and the hidden crack distribution to form a comprehensive damage map including the correlation degree between surface damage and internal damage, including: Perform damage path correlation on the deformation contour features in the microscopic deformation features and the intermittent connection features in the hidden crack distribution to obtain the spatio - temporal correlation relationship between the area with the maximum deformation degree in the deformation area of the surface damage and the position where the crack first appears in the intermittent connection features of the internal cracks. According to the spatio - temporal correlation relationship, screen out the part where the coverage area of the area with the maximum deformation degree exceeds the preset threshold and mark it as the key damage area, and continuously repair the intermittent connection features of the internal cracks corresponding to the key damage area according to the diffusion direction to form the internal crack propagation path features. Integrate the microscopic deformation features and the internal crack propagation path features to generate a comprehensive damage map including the correlation degree between the deformation intensity of the surface damage and the internal crack propagation path features.

[0012] Optionally, when the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, trigger the collaborative verification of multi - source data. By collaboratively verifying the feature consistency of the damage data in the abnormal damage area, output a claim settlement determination result including a maintenance fraud mark, including: When the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, generate a multi - source data verification request, which is used to trigger the data extraction of the comprehensive damage map and the historical maintenance records. According to the multi - source data verification request, extract the deformation contour features of the surface damage and the internal crack propagation path features of the abnormal damage area from the comprehensive damage map, and extract the damage repair features of the vehicle components in past accidents from the historical maintenance records. Perform multi-source feature consistency comparison on the deformation contour features, internal crack propagation path features, and damage repair features, and calculate the spatial coincidence degree parameter of the deformation contour features and the internal crack path features in the abnormal damage area, as well as the morphological difference parameter between the current damage features and the repair features; Compare the spatial coincidence degree parameter and the morphological difference parameter with a preset coincidence threshold and a preset difference threshold. If the spatial coincidence degree parameter is lower than the preset coincidence threshold or the morphological difference parameter exceeds the preset difference threshold, it is determined that there is multi-source data inconsistency in the abnormal damage area, a maintenance fraud flag is generated, and a claim settlement determination result including the maintenance fraud flag is output according to the maintenance fraud flag.

[0013] In a second aspect, the present application provides a vehicle mutual claim settlement system based on multi-modal data fusion, including: An integration module for integrating damage data composed of visual sensors, depth sensing units, and maintenance records in a vehicle accident scenario, extracting the types of vehicle components from the maintenance records, and obtaining the mechanical parameters of the accident collision through a depth detector; A capture module for performing dual-band scanning on damage features of different dimensions in the damage data to capture microscopic deformation features of surface damage and hidden crack distributions of internal damage; A correlation module for spatiotemporally correlating the microscopic deformation features and the hidden crack distributions to form a comprehensive damage map including the correlation degree between surface damage and internal damage; A generation module for obtaining the normal damage evolution law of vehicle components in a non-fraud scenario and generating a standard damage sequence corresponding to the comprehensive damage map in combination with a generative adversarial network; A comparison module for layer-by-layer comparison of the standard damage sequence and the damage data, extracting abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculating a fraud risk index based on the spatial density of the abnormal damage areas on the vehicle components; A verification module for triggering collaborative verification of multi-source data when the fraud risk index exceeds a determination threshold dynamically adjusted according to the types of vehicle components and the mechanical parameters of the accident collision, and outputting a claim settlement determination result including a maintenance fraud flag by collaboratively verifying the feature consistency of the damage data in the abnormal damage area.

[0014] In a third aspect, an embodiment of the present application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a vehicle mutual claim settlement method based on multi-modal data fusion as described in the first aspect above.

[0015] This application integrates damage data composed of visual sensors, depth sensing units, and maintenance records in vehicle accident scenarios, extracts the types of vehicle components from the maintenance records, and obtains the mechanical parameters of accident collisions through depth detectors, enabling the fusion of multi-source heterogeneous data to accurately capture the correlation between vehicle damage characteristics and accident mechanical characteristics; by performing dual-band scanning on damage characteristics of different dimensions in the damage data to capture the microscopic deformation characteristics of surface damage and the hidden crack distribution of internal damage, it can simultaneously achieve refined detection of surface visible damage and penetrative analysis of internal structure damage; by spatio-temporally correlating the microscopic deformation characteristics and the hidden crack distribution to form a comprehensive damage map including the correlation degree between surface damage and internal damage, it can establish a spatial mapping model of the damage propagation path and improve the comprehensiveness of damage cause analysis; by obtaining the normal damage evolution law of vehicle components in a non-fraudulent scenario and combining with a generative adversarial network to generate a standard damage sequence corresponding to the comprehensive damage map, it can construct a benchmark comparison framework based on the real damage evolution law and strengthen the ability to identify abnormal damage patterns; by layer-by-layer comparing the standard damage sequence with the damage data, extracting the abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculating the fraud risk index based on the spatial density of the abnormal damage areas on the vehicle components, it can quantify the abnormal aggregation characteristics of non-natural damage distribution and accurately locate potential fraud risk points; by triggering the collaborative verification of multi-source data when the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of accident collisions, and collaboratively verifying the feature consistency of the damage data in the abnormal damage areas to output a claim settlement determination result including a maintenance fraud mark, it can enhance the robustness of fraud determination based on a dynamic threshold mechanism and multi-source feature cross-verification, and achieve the precision of claim settlement decision-making and the automation of anti-fraud defense.

[0016] Furthermore, through the precise modeling of user behavior baselines and the deviation matching of key driving behavior features, a quantitative assessment of members' driving habits and accident proneness is achieved, effectively capturing the abnormal behavior patterns of high-risk users. By combining the dynamic adjustment of multi-dimensional coupled weights to the mutual assistance fund allocation ratio threshold, intelligent risk control of fund flows can be automatically triggered when the fraud risk index and behavior anomaly coefficient are superimposed, preventing the abnormal loss of mutual assistance pool funds. Relying on a multi-index collaborative determination mechanism of behavior anomalies, fraud risks, and fund compliance markings, high-risk accounts can be quickly frozen and manual review initiated when multiple abnormal features occur simultaneously, significantly improving the timeliness and accuracy of fraud behavior interception. By injecting the fraud features confirmed by the review back into the adversarial generation network, a continuous learning loop based on real fraud cases is formed, continuously optimizing the generation strategy of standard damage sequences and enhancing the system's adaptive recognition ability for new fraud techniques. Ultimately, a full-chain anti-fraud protection system covering behavior modeling, dynamic risk control, collaborative determination, and model optimization is constructed, realizing the safe management of mutual assistance funds and the pre-blocking of fraud risks while ensuring the stable operation of the vehicle mutual assistance plan.

[0017] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 The flowchart of a vehicle mutual assistance claim settlement method based on multi-modal data fusion provided by the present application is shown; Figure 2 The structural schematic diagram of a vehicle mutual assistance claim settlement system based on multi-modal data fusion provided by the present application is shown; Figure 3 The structural schematic diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to enable those skilled in the art of the present technology to better understand the solutions of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application.

[0021] In some of the processes described in the specification, claims, and the above-mentioned drawings of the present application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any order of execution. Additionally, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit that "first" and "second" are of different types.

[0022] Researchers have found that existing vehicle repair fraud detection methods rely on single-dimensional damage data or static threshold determination, making it difficult to dynamically correlate surface damage, internal cracks, and collision mechanical parameters, resulting in a high undetected rate of hidden fraud risks and a rigid determination logic. Based on this, a multi-source data collaborative detection method for vehicle repair fraud is provided. This method can accurately identify the risk of repair fraud through the dynamic modeling of damage maps and the comparison of abnormal sequences of generative adversarial networks. The technical solution of the present application is applicable to scenarios such as vehicle insurance claim review, repair service fraud warning, and accident damage dynamic assessment.

[0023] The entire R & D process embodies a closed-loop determination mechanism for multi-modal damage data fusion and dynamic verification, aiming to overcome the defects of data dimension fragmentation, lagging response to hidden damage, and static fraud determination in existing solutions. By fusing the spatio-temporal correlation of microscopic deformation and hidden cracks through dual-band scanning technology, it breaks through the dependence on a single damage feature in traditional detection; based on the standard damage sequence modeling of generative adversarial networks, it solves the problem that it is difficult for artificial rules to adapt to complex damage evolution laws; combined with a collaborative verification mechanism of abnormal region spatial density and dynamic thresholds, it eliminates the risk of misjudgment of local features in fraud determination; finally, through the consistency verification of multi-source data features, it realizes a two-way improvement in the accuracy and anti-interference ability of repair fraud identification. This method realizes full-process dynamic optimization from data fusion to fraud determination, significantly enhancing the prevention and control ability of repair fraud in complex scenarios.

[0024] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0025] Figure 1 The following is a flowchart of a vehicle mutual claim settlement method based on multi-modal data fusion provided for the embodiments of the present application, as Figure 1As shown, the method includes: 101. Integrate the damage data composed of visual sensors, depth sensing units, and maintenance records in the vehicle accident scenario, extract the types of vehicle components from the maintenance records, and obtain the mechanical parameters of the accident collision through a depth detector; In this step, the damage data refers to the set of vehicle damage information collected by multiple sensors. The visual sensor refers to an optical acquisition device used to capture vehicle surface damage. The depth sensing unit refers to a detection device that measures the depth information of an object. The maintenance record refers to the document data of the vehicle's historical maintenance situation. The vehicle component type refers to the classification identifier of the damaged part. The mechanical parameter refers to the physical quantity that describes the magnitude and direction of the collision force.

[0026] At a certain traffic accident damage assessment center, a white SUV damaged in a rear-end collision was towed into the inspection workshop. The visual sensor collected high-definition images of the vehicle's front bumper, hood, and left front headlight area, clearly capturing the peeling of the paint on the bumper surface and the wrinkling deformation of the hood edge. The depth sensing unit synchronously scanned the deformation depth of the vehicle frame longitudinal beam and detected a ten-centimeter depression displacement in the middle of the longitudinal beam. The system retrieved the vehicle's historical maintenance record and found that the right rear door had been replaced due to a scratch three months ago. Combining the mechanical parameters of this collision obtained by the depth detector (the impact angle was a 45-degree oblique impact), it was initially determined that the front damage was directly related to this accident, and at the same time, the vehicle frame longitudinal beam was marked as a high-risk verification component.

[0027] In the embodiment of the present application, first, integrate the damage images collected by the visual sensor, the three-dimensional deformation data obtained by the depth sensing unit, and the historical maintenance information in the maintenance record in the vehicle accident scenario through a data fusion algorithm. Secondly, extract keyword fields (such as "left front door replacement") from the text of the maintenance record, parse out the specific vehicle component types (such as doors, bumpers), and analyze the pressure distribution and impact waveform at the moment of collision through a depth detector to generate the mechanical parameters of the accident collision (such as collision angle, force magnitude). Then, align the component type and the mechanical parameter according to the timestamp to form an original data set containing multi-dimensional damage features as the input source for subsequent analysis.

[0028] 102. Perform dual-band scanning on the damage features of different dimensions in the damage data to capture the microscopic deformation features of the surface damage and the hidden crack distribution of the internal damage; In this step, the dual-band scanning refers to a detection method that simultaneously collects damage features of different frequency bands. The microscopic deformation feature refers to the quantitative description of the fine deformation on the surface of an object. The hidden crack distribution refers to the spatial arrangement of invisible cracks inside the material.

[0029] In the embodiment of the present application, first, the integrated damage data in step 101 is preprocessed to separate the surface deformation data and the internal structure data. Secondly, a dual-band scan is initiated: by adjusting the resolution parameters of the optical sensor, high-precision scanning of the surface damage is performed, and the edge enhancement algorithm is used to enhance the microscopic deformation features (such as the width gradient of the paint surface crack and the depression depth distribution); at the same time, the ultrasonic wave band is switched, and the penetration parameters are adjusted to detect the internal damage area, and the hidden crack distribution (such as the metal interlayer fracture path) is captured through echo analysis. Then, the two types of data are mapped to the same three-dimensional coordinate system to ensure the spatial alignment of the surface and internal damage, providing a basis for correlation analysis.

[0030] For the damage data of the bumper and the longitudinal beam of the frame, the system initiates a dual-band scan: the short-wave infrared band penetrates the paint layer to identify three radial hidden cracks in the plastic bracket inside the bumper; the long-wave ultraviolet band captures the microscopic deformation features of the paint surface around the welding point of the longitudinal beam, indicating the lattice layer misalignment of the metal substrate. By comparing the results of the two-band scans, the system finds that the distribution density of the hidden cracks in the depressed area of the longitudinal beam far exceeds the surface deformation range, and the crack extension direction deviates from the stress conduction path of the mechanical parameters of this collision, triggering an internal damage anomaly warning.

[0031] 103. Spatially and temporally correlate the microscopic deformation features with the hidden crack distribution to form a comprehensive damage map including the correlation degree between the surface damage and the internal damage; In this step, spatio-temporal correlation refers to the correlation analysis of damage features in different dimensions in terms of time and space. The comprehensive damage map refers to a visualization model integrating the correlation between the surface and internal damage. The correlation degree refers to the correlation degree index between different damage features.

[0032] In the embodiment of the present application, first, based on the scan results of step 102, a spatio-temporal correlation model of microscopic deformation features (such as the surface depression coordinates) and hidden crack distribution (such as the starting point of the internal crack) is established. Secondly, the overlapping area of the two in the three-dimensional coordinates is calculated through the spatial matching algorithm. If the position with the deepest surface depression coincides with the starting point of the internal crack, it is marked as a high-correlation area. Then, combined with time series analysis (such as the damage diffusion speed), the time sequence relationship between the surface deformation and the internal crack is verified, and finally a comprehensive damage map is generated, and the correlation intensity between the surface and internal damage is displayed in a heat map layer, providing a basis for generating a standard damage sequence.

[0033] The system spatially and temporally correlates the paint peeling area on the bumper surface with the crack in the internal bracket, and confirms that there is a spatial offset of 15 centimeters between the surface impact point and the starting position of the internal crack. Combining with the lattice layer misalignment characteristics of the vehicle frame longitudinal beam, a comprehensive damage map covering the surface depression depth, the trend of hidden cracks, and the material fatigue degree is constructed. The map shows that the damaged area of the longitudinal beam presents an abnormal stress distribution pattern of "central compression - edge tension", which is significantly inconsistent with the "unidirectional gradient decreasing" damage law that should be generated by the frontal impact in this rear-end collision accident, indicating that there may be historical damage superposition or human intervention.

[0034] 104. Obtain the normal damage evolution law of vehicle components in a non-fraudulent scenario, and combine with a generative adversarial network to generate a standard damage sequence corresponding to the comprehensive damage map; In this step, the normal damage evolution law refers to the typical mode of damage development in the case of no fraud. The generative adversarial network refers to a machine learning model used to generate a standard damage sequence. The standard damage sequence refers to the data of the damage development process that conforms to the normal evolution law.

[0035] In the embodiment of the present application, first, extract the normal damage evolution law of the same type of vehicle component under specific mechanical parameters from the historical non-fraudulent case library (such as the average rate of crack propagation after a side collision of a car door). Secondly, input the law data into a generative adversarial network (GAN), train the generator to simulate the damage diffusion path under different collision conditions, and generate a candidate damage sequence that matches the spatio-temporal characteristics of the current comprehensive damage map. Then, the discriminator screens out the sequences that conform to the physical laws (such as excluding abnormal situations where cracks suddenly extend in the reverse direction), and retains the logically reasonable sequences as the standard damage sequences for subsequent fraud detection.

[0036] Based on a non-fraudulent accident case library of 100,000 cases, the generative adversarial network learns the normal damage evolution law of the vehicle frame longitudinal beam under a 45-degree oblique collision: usually manifested as a progressive deformation extending from the impact point to the end of the longitudinal beam, and the hidden cracks are mostly concentrated within 5 centimeters on both sides of the main stress conduction path. The system generates a standard damage sequence corresponding to the comprehensive damage map of this vehicle. The simulation shows that under the same collision conditions, two main cracks parallel to the vehicle body axis should appear on the longitudinal beam, rather than the six radially intersecting cracks detected currently, revealing a significant deviation between the actual damage and the theoretical model.

[0037] 105. Compare the standard damage sequence with the damage data layer by layer, extract the abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculate the fraud risk index based on the spatial density of the abnormal damage areas on the vehicle components; In this step, layer-by-layer comparison refers to an analytical method of gradually comparing by damage levels. The abnormal damage area refers to the damaged part that does not conform to the standard evolution law. The spatial density refers to the degree of distribution density of the abnormal area on the surface of the component. The fraud risk index refers to a numerical indicator that quantifies the possibility of fraud.

[0038] In the embodiment of the present application, first, the standard damage sequence generated in step 104 and the damage data in step 101 are compared layer by layer according to the time layer and the space layer, and the abnormal damage areas (such as a sudden increase in the crack width somewhere) deviating from the normal evolution logic are marked through a difference algorithm (such as contour similarity comparison). Secondly, the distribution density of the abnormal areas on the surface of the vehicle component is counted (such as the proportion of the abnormal area to the total area), and combined with the component type (such as bumpers being more prone to forged damage) and mechanical parameters (such as deep cracks should not be generated by low-speed collisions), the fraud risk index is calculated through a weighted formula. Finally, the decision threshold is dynamically adjusted according to the index level to ensure that the detection results adapt to different scenarios.

[0039] By comparing the standard damage sequence with the current damage data layer by layer, the system identifies three abnormal damage areas on the longitudinal frame of the vehicle: the crack density at the left front welding point reaches three times the standard value, reverse tensile deformation appears in the middle depression area of the longitudinal beam, and the covered area of the hidden crack network exceeds the expected range by 60%. Based on the spatial density distribution of these abnormal areas, combined with the component type (high-strength steel frame) and the mechanical parameters of the accident collision (medium-strength impact), the calculated fraud risk index reaches 0.87, far exceeding the dynamically adjusted decision threshold of 0.65, triggering the manual review and multi-source data cross-verification process.

[0040] 106. When the fraud risk index exceeds the decision threshold dynamically adjusted according to the type of vehicle component and the mechanical parameters of the accident collision, trigger the collaborative verification of multi-source data, and output a claim settlement decision result including a maintenance fraud mark by verifying the feature consistency of the damage data in the abnormal damage area.

[0041] In this step, the decision threshold refers to the critical parameter value for triggering fraud determination. The multi-source data collaborative verification refers to the cross-verification process of integrating multiple data sources. The feature consistency refers to the degree of coincidence of the damage features reflected by different data sources. The claim settlement decision result refers to the final evaluation conclusion including a fraud mark.

[0042] In the embodiments of the present application, first, when the fraud risk index exceeds the dynamic threshold, multi-source data collaborative verification is triggered: the high-definition images of the visual sensor, the three-dimensional deformation data of the depth sensing unit, and the time nodes of the maintenance records are retrieved, and the feature consistency is cross-verified within the abnormal damage area (such as whether the visual image shows artificial scratching marks). Secondly, if data contradictions are found (such as the maintenance record shows component replacement but there is no corresponding damage), it is determined as a suspected maintenance fraud. Finally, the verification results are integrated to generate a claim settlement determination result, clearly marking the fraud type (such as forged damage, exaggerated maintenance scope), and outputting an evidence chain (such as timeline contradictions, physical law inconsistencies) for manual review.

[0043] The system starts multi-source data collaborative verification: the visual sensor rechecks and confirms that there are secondary grinding marks on the longitudinal beam welding point; the depth sensing unit detects that the metal fatigue characteristics of the abnormal damage area do not match the time axis of this collision; the maintenance records comparison finds that the vehicle had a "chassis inspection" at a non-cooperative maintenance point two weeks ago. The above evidence chain confirms that the feature consistency of the abnormal damage area points to an artificial damage expansion fraud behavior. The system outputs a claim settlement determination result: rejecting the claim for the replacement cost of the vehicle frame longitudinal beam, marking a "maintenance fraud mark" in the loss assessment report, and synchronizing the relevant evidence to the insurance anti-fraud database at the same time.

[0044] In summary, steps 101 to 106 achieve multi-modal data fusion analysis for vehicle damage detection and claim fraud identification. By integrating multi-source data such as visual sensors, depth sensing units, and maintenance records, the system can construct a comprehensive damage map covering surface micro-deformation and internal hidden cracks. Using the technical means of the adversarial generative network to generate standard damage sequences, it breaks through the limitations of traditional single-dimensional damage comparison, realizes dynamic modeling of damage evolution laws and intelligent identification of abnormal areas. Combining mechanical parameters to dynamically adjust the fraud determination threshold significantly improves the accuracy and adaptability of claim fraud detection, forming an intelligent analysis system for the entire process from data collection to fraud determination.

[0045] In order to improve the collaborative optimization ability of the monitoring accuracy of the charging pile operation status and the modeling of the user charging behavior portrait, this study constructs a multi-source monitoring signal fusion mechanism for the mechanical bearing vibration waveform and the charging current phase offset, combines the temperature gradient distribution data and the power grid load fluctuation characteristics, and designs a dynamic compensation strategy based on the reinforcement learning decision model. The experiment uses a joint analysis method of spatio-temporal trajectory coordinates and energy consumption characteristics, establishes a mapping relationship between the charging power adjustment interval division rule and the vibration suppression priority, and realizes the adaptive iteration of the strategy parameters through a closed-loop feedback mechanism.

[0046] In some embodiments, for the method described in claim 1, the method further includes: 201. Establish a user behavior baseline according to the payment records, driving behavior data, and historical accident data of members during the vehicle mutual assistance period; In step 201, the vehicle mutual assistance period refers to the time period during which members participate in vehicle mutual assistance protection. The payment record refers to the historical data of members' regular payment of mutual assistance fees. The driving behavior data refers to the operation information reflecting members' driving habits. The historical accident data refers to the records of traffic accidents that members have had in the past. The user behavior baseline refers to the normal behavior standard model established based on members' historical data.

[0047] In the embodiments of the present application, first, integrate the payment records (such as the on-time payment frequency), driving behavior data (such as average speed, number of hard brakes) and historical accident data (such as accident time, liability determination) of members during the vehicle mutual assistance period, and divide users into different risk levels through a clustering algorithm. Secondly, extract common features for each type of user (such as low-risk users generally pay in a timely manner and have few hard brakes), calculate the mean and variance of each feature through a statistical model, and construct a user behavior baseline reflecting normal behavior (such as "low-risk users have ≤ 3 hard brakes per month"). Finally, store the baseline in the database and bind it to the user ID as the basis for subsequent deviation analysis.

[0048] 202. Extract key features from the driving behavior data of the member vehicles collected and the vehicle state data at the time of accident trigger, and perform deviation matching between the key features and the user behavior baseline to generate a behavior anomaly coefficient. The key features include hard acceleration, high-frequency hard braking, and night accident tendency features; In step 202, the time of accident trigger refers to the time point when a vehicle collision accident occurs. The vehicle state data refers to the operating parameters of the vehicle at the time of the accident. The key features refer to the core behavior indicators used to evaluate risks. Hard acceleration refers to the driving behavior of significantly increasing the vehicle speed within a short period of time. High-frequency hard braking refers to the driving behavior of frequently performing emergency braking. The night accident tendency feature refers to the driving characteristics that are prone to accidents during the night period. Deviation matching refers to the process of comparing the current features with the baseline for differences. The behavior anomaly coefficient refers to a value that quantifies the degree of deviation of driving behavior from the normal.

[0049] In the embodiments of the present application, first, extract dynamic features such as hard acceleration (such as acceleration exceeding 2 m / s²) and high-frequency hard braking (such as ≥ 5 brakes within 10 minutes) from the real-time collected driving behavior data, and extract the night accident tendency feature (such as the proportion of night accidents exceeding 60%) from the vehicle state data at the time of accident trigger. Secondly, compare the above key features item by item with the user behavior baseline in step 201, calculate the single-item deviation score through a deviation algorithm (such as the difference ratio between the actual number of hard brakes and the baseline value), and then generate a comprehensive behavior anomaly coefficient through weighted summation. Finally, associate and store the coefficient with the user ID for subsequent risk decision-making.

[0050] 203. When receiving the damage assessment form from the cooperative repair shop, calculate the multi-dimensional coupling weights of the behavior anomaly coefficient and the fraud risk index based on the member's vehicle accident data, vehicle damage data, damage assessment amount, and the fraud risk index, so as to adjust the proportion threshold for transferring funds from the mutual aid pool to the repair shop according to the multi-dimensional coupling weights. In step 203, the cooperative repair shop refers to a repair institution that has signed an agreement with the mutual aid platform. The damage assessment form refers to a vehicle loss assessment report issued by the repair shop. The vehicle damage data refers to the detailed information on the damaged parts and degrees of the vehicle. The damage assessment amount refers to the total repair cost estimated by the repair shop. The multi-dimensional coupling weight refers to the comprehensive influence coefficient of multiple risk indicators. The mutual aid pool refers to a guarantee fund pool formed by the joint contributions of members. The proportion threshold refers to the upper limit proportion of paying the repair cost from the mutual aid pool.

[0051] In the embodiment of the present application, first, when receiving the damage assessment form from the cooperative repair shop, retrieve the member's vehicle accident data (such as the collision location), vehicle damage data (such as the damage depth), and the fraud risk index generated in step 105 from the system. Secondly, analyze the correlation between the behavior anomaly coefficient (such as coefficient 0.8), the fraud risk index (such as index 75), and the damage assessment amount through a decision tree model, and calculate the multi-dimensional coupling weights of the three (such as the behavior anomaly accounts for 40%, the fraud risk accounts for 50%, and the damage assessment amount accounts for 10%). Then, adjust the proportion threshold for transferring funds from the mutual aid pool to the repair shop according to the weights (such as the original threshold of 80% is lowered to 65%) to limit the compensation ratio of high-risk users. Finally, synchronize the adjusted threshold to the financial system for real-time effect.

[0052] 204. If at least two of the behavior anomaly coefficient, the fraud risk index, and the fund flow compliance mark exceed the proportion threshold, freeze the mutual aid rights and interests of the member and initiate manual review. At the same time, inject the fraud behavior characteristics confirmed by the review into the adversarial generation network in reverse to optimize the generation strategy of the standard damage sequence.

[0053] In step 204, the fund flow compliance mark refers to the identification of the legality of the use of the repair cost. The mutual aid rights and interests refer to the qualification rights of members to enjoy mutual aid guarantees. The manual review refers to the review of fraud behaviors by professional personnel. The reverse injection refers to the process of feeding back the confirmed fraud characteristics to the model. The generation strategy refers to the simulation rules and methods of the standard damage sequence.

[0054] In the embodiment of the present application, first, monitor the abnormal behavior coefficient, fraud risk index and compliance mark of funds flow (such as abnormal transfer records of repair shop accounts). If at least two of the three exceed the adjusted ratio threshold of step 203 (such as abnormal behavior coefficient>0.7 and abnormal funds mark), the risk control rules are triggered. Secondly, freeze the mutual aid rights of the member (such as suspending the claim application), and push the case to the manual review queue, and the auditor verifies the original evidence such as accident photos and maintenance records. Then, if the review confirms the fraudulent behavior (such as falsifying the accident scene), extract the fraud features (such as the damage morphology does not conform to the collision logic), reverse inject the adversarial generative network of step 104, and optimize the generation strategy of the standard damage sequence (such as adding recognition rules for forged cracks). Finally, update the risk control model and lift the rights and interests restrictions of the misjudged users to complete the closed-loop optimization.

[0055] Here is a specific example: In the intelligent claims settlement system of the Urban New Energy Vehicle Mutual Aid Alliance, the multi-source data fusion system effectively curbs insurance fraud. After a shared car platform was connected to the mutual aid system, the system integrated the three-year charging records, sharp turn frequency and historical scratch data of 3,000 members (step 201), and established a user behavior baseline covering low-speed and high-speed commuting in urban areas. When a member's vehicle triggered an accident on a certain road section in the early morning, the on-board sensor captured five consecutive emergency brakes and the steering wheel angle exceeded the baseline by twice (step 202), generating a behavioral anomaly coefficient of up to 85 points. After the system received the front suspension assembly replacement assessment form submitted by the cooperative repair shop, combined with the abnormal characteristics of the damage diffusion direction of the laser scan (step 203), the behavior coefficient was coupled with the fraud risk index for calculation, and the mutual aid fund payment ratio was reduced from 100% to 55%. The review found that the member's three night accidents within three months all involved damage to the same component (step 204). The system immediately froze his rights and interests and injected the abnormal characteristics into the adversarial network. The trained model added a new identification dimension for the correlation between high-frequency steering wheel vibration and suspension damage, and successfully intercepted two subsequent insurance fraud cases involving fake steering gear failures, forming a closed-loop risk control system from user portrait construction to fraud model evolution.

[0056] In summary, 201 to 204 have realized the risk assessment and anti-fraud decision-making system for mutual aid members of vehicles based on multi-dimensional data fusion. By establishing a user behavior baseline model, integrating driving behavior, accident history and payment record data, extracting dynamic features such as sudden acceleration and high-frequency sudden braking to generate behavioral abnormality coefficients; constructing a multi-dimensional coupling weight algorithm for the amount of damage assessment, vehicle injury data and fraud risk index, and dynamically adjusting the threshold of mutual aid fund allocation ratio; introducing an adversarial generative network to optimize the standard damage sequence generation strategy, triggering the freezing mechanism when the abnormal index exceeds the threshold and reversely injecting the manual review results to form a closed-loop feedback, completing the iterative update of the risk model and the adaptive optimization of the anti-fraud strategy, and improving the compliance of the mutual aid fund flow and the real-time risk identification.

[0057] In some embodiments, as described in step 105, the standard damage sequence is compared layer by layer with the damage data, the abnormal damage areas deviating from the evolution law of the standard damage sequence are extracted from the damage data, and the fraud risk index is calculated based on the spatial density of the abnormal damage areas on the vehicle components, including: 301. Compare the damage form change path of the vehicle components in the damage data with the damage form change path of the standard damage sequence layer by layer. The layer-by-layer comparison divides the damage area of the vehicle into multiple sub-areas, and layer by layer compares the damage diffusion direction and the similarity of damage forms within each sub-area; In step 301, the damage form change path refers to the morphological evolution trajectory during the damage development process. The layer-by-layer comparison refers to a method of step-by-step comparison according to the damage level. The sub-area refers to a local analysis unit divided from the damage part. The damage diffusion direction refers to the spatial trend of damage expansion. The damage form similarity refers to the quantitative value of the similarity degree between damage features.

[0058] In the embodiments of the present application, first, the damage form change path of the vehicle components in the damage data is spatially segmented from the damage form change path of the standard damage sequence, and the vehicle damage area is divided into multiple sub-areas (such as a 10 cm × 10 cm grid). Secondly, the damage diffusion direction (such as the crack extension angle) and the form similarity (such as the depression depth gradient) are analyzed layer by layer within each sub-area, and the difference value between the two in each layer is calculated through the direction comparison algorithm. Then, the difference value is mapped to the abnormal score within the sub-area to generate the layer-by-layer comparison result. Finally, the comparison results of each layer are summarized according to the spatial position to form a complete abnormal distribution map of the sub-area.

[0059] 302. Based on the result of the layer-by-layer comparison, extract the sub-areas in the damage data where the damage diffusion direction deviates from the standard damage sequence by more than a preset angle threshold, mark the sub-areas as deviation areas, and calculate the abnormal damage features with a damage form similarity lower than the preset similarity threshold within each deviation area; In step 302, the preset angle threshold refers to the critical angle value for judging the direction deviation. The deviation area refers to the local area where the damage diffusion direction is abnormal. The abnormal damage feature refers to the damage characteristic that does not conform to the standard sequence.

[0060] In the embodiments of the present application, first, the deviation angle between the damage diffusion direction of the sub-region and the standard damage sequence is extracted from the hierarchical comparison result. If the deviation exceeds the preset angle threshold (such as 30 degrees), the sub-region is marked as a deviated region. Secondly, the damage morphology similarity is further analyzed within the deviated region, and the similarity value with the standard sequence is calculated through a morphology matching algorithm (such as contour comparison). Then, the regions with similarity lower than the preset similarity threshold are screened out, and their abnormal damage features (such as irregular crack bifurcation, discontinuous depression) are extracted. Finally, the position of the deviated region and the type of abnormal features are recorded to provide data for subsequent merging.

[0061] 303. Statistically analyze the distribution position and coverage range of the deviated regions on the vehicle component, and merge adjacent deviated regions according to the continuity of the damage diffusion direction to obtain a set of abnormal damage regions; In step 303, the distribution position refers to the spatial coordinates of the abnormal region on the component. The coverage range refers to the affected area of the abnormal region. The set of abnormal damage regions refers to the summary data set of all deviated regions.

[0062] In the embodiments of the present application, first, a spatial distribution heat map is drawn according to the distribution position of the deviated regions on the vehicle component. Secondly, it is detected whether the damage diffusion directions of adjacent deviated regions are continuous (such as the crack extension direction of the previous region is consistent with that of the next region). If continuous, they are merged into a larger abnormal damage region through a region merging algorithm (such as morphological dilation). Then, the coverage range (such as area, shape) of the merged region is statistically analyzed to generate a set of abnormal damage regions. Finally, the boundary coordinates and core abnormal features of each set are marked for density calculation.

[0063] 304. Calculate the spatial density parameter of the abnormal damage region according to the coverage area of each deviated region in the set of abnormal damage regions on the vehicle component and the total area of the corresponding component. The spatial density parameter is proportional to the coverage area; In step 304, the coverage area refers to the actual influence range of the abnormal region. The total area refers to the overall surface area of the component to be analyzed. The spatial density parameter refers to the quantization value of the distribution density of the abnormal region.

[0064] In the embodiments of the present application, first, the coverage area of each deviated region in the set of abnormal damage regions on the vehicle component is calculated and converted into a ratio with the total area of the corresponding component. Secondly, a spatial density parameter is generated through a density calculation formula (such as coverage area / total area of the component), and the parameter value increases as the coverage area increases. Then, the density parameter is normalized and mapped to the range of 0-1 for subsequent weighting. Finally, the density parameter is associated and stored with the position information of the deviated region to form a spatial density distribution map.

[0065] 305. Weightedly superimpose the spatial density parameter and the deviation amplitude of the similarity of the damage morphology within the deviation region to obtain a superimposed value as the fraud risk index.

[0066] In step 305, the deviation amplitude refers to the degree of difference between the damage characteristics and the standard sequence. Weighted superposition refers to a calculation method of numerically accumulating according to the weight distribution. The superimposed value refers to the comprehensive evaluation value after weighted calculation.

[0067] In the embodiment of the present application, first, obtain the spatial density parameter from step 304, and extract the deviation amplitude of the similarity of the damage morphology (such as the similarity difference) from step 302. Secondly, superimpose the two through a weighted formula: the density parameter has a higher weight (such as 60%), and the deviation amplitude has a lower weight (such as 40%) to generate a comprehensive superimposed value. Then, perform interval mapping (such as 0-100) on the superimposed value to convert it into a fraud risk index. The higher the value, the greater the likelihood of fraud. Finally, associate the index with the vehicle component type and output it to the risk determination system to trigger the subsequent verification process.

[0068] The following is a specific example: In the intelligent auto insurance claim assessment system, the multi-modal damage analysis system accurately identifies the authenticity of accidents. When a five-car rear-end collision occurs on an urban expressway, the system obtains the three-dimensional damage path of the damaged vehicle through an in-vehicle laser scanning device, conducts hierarchical comparison with the standard accident damage database, and divides the front cabin of the target vehicle into twelve detection sub-regions. The algorithm detects that the damage diffusion direction in the third sub-region (left front longitudinal beam) deviates from the rear-end collision accident model by more than the safety threshold, marks it as a deviation region, and extracts its spiral crack characteristics. The system further scans and finds that the adjacent fifth (front fender) and seventh (right wing panel) sub-regions show abnormal diffusion with direction continuity, and combines them to form an abnormal damage set across components. By calculating the coverage ratio of this set in the engine compartment, a spatial density heat map showing that the core risk focuses on the left front side of the vehicle is generated. Finally, the system fuses the density parameter and the deviation degree of the crack morphology to generate a high-risk index to trigger a manual re-inspection: the claim settlement center retrieves the roadside monitoring during the accident period and finds that the old damage on the left front side of the vehicle does not match the current reported time. At the same time, a survey drone is dispatched for secondary millimeter-wave scanning to confirm the stress superposition characteristics at different time nodes in the abnormal region. The system generates a fraud warning report accordingly, automatically freezes the suspicious claim process, and synchronously uploads the multi-vehicle collision mechanics model to the insurance blockchain for evidence storage. After the case is closed, the system updates the standard damage sequence, increases the feature recognition dimension for spliced accidents, and forms a closed-loop anti-fraud mechanism from multi-source data fusion to risk intelligent judgment.

[0069] In summary, steps 301 to 305 achieve hierarchical and refined analysis of the vehicle damage area and quantitative assessment of fraud risk. By dividing the damage area into multiple layers of sub-areas and comparing the damage diffusion direction and morphological similarity layer by layer, the system can accurately locate abnormal areas that deviate from the standard damage pattern. The weighted superposition algorithm of the spatial density parameter and the morphological deviation amplitude effectively balances the dual effects of the damage distribution range and the morphological abnormality degree, and constructs a multi-dimensional fraud risk index calculation model. This technical solution not only retains the detailed information of local damage features but also strengthens the quantitative assessment ability of the overall damage distribution, significantly improving the scientificity and reliability of fraud risk determination.

[0070] In some embodiments, as described in step 104, obtaining the normal damage evolution law of vehicle components in a non-fraud scenario and generating a standard damage sequence corresponding to the comprehensive damage atlas in combination with a generative adversarial network includes: 401. Obtain the normal damage evolution law of vehicle components in a non-fraud scenario. The normal damage evolution law is obtained by integrating the type of vehicle components, the mechanical parameters of accident collisions, and the corresponding damage morphology change path. The damage morphology change path records the entire process of the starting position, diffusion direction, and morphological branch changes of damage diffusion after the vehicle components are collided. In step 401, the normal damage evolution law refers to the typical mode of damage development of vehicle components in real accidents. The type of vehicle components refers to the classification identifier of the damaged parts. The mechanical parameters refer to the set of physical quantities describing the magnitude and direction of the collision force. The damage morphology change path refers to the complete trajectory of morphological evolution during the damage development process. The starting position refers to the spatial coordinates where the damage initially occurs. The diffusion direction refers to the spatial trend direction of damage expansion. The morphological branch change refers to the secondary cracks or deformations generated during the damage development process.

[0071] In the embodiments of the present application, first, extract the accident collision mechanical parameters (such as collision angle, impact force magnitude) and the complete damage morphology change path data corresponding to different types of vehicle components (such as car doors, hoods) from the historical non-fraud case library. Secondly, correlate the three through data fusion technology: analyze the starting position (such as the left end of the bumper), diffusion direction (the crack extends to the right), and morphological branch changes (such as the time and position of bifurcation) of damage diffusion of the same component under specific mechanical parameters. Then, adopt a time series modeling method to integrate the damage evolution timeline of different components under different mechanical conditions to form a standardized normal damage evolution law, which includes the key nodes and morphological change thresholds of the entire process of damage diffusion.

[0072] 402. Based on the corresponding relationship between the type of vehicle components and the mechanical parameters in the normal damage evolution law, combine the generator of the generative adversarial network to generate candidate damage diffusion rules for different types of vehicle components under specific mechanical parameters; In step 402, the adversarial generative network refers to a machine learning framework that includes a generator and a discriminator. The generator refers to a neural network module used to simulate the damage diffusion rule. The candidate damage diffusion rule refers to the possible damage development patterns simulated by the generator. The specific mechanical parameters refer to the combination of physical quantities of the collision force within a defined range.

[0073] In the embodiments of the present application, first, based on the normal damage evolution law generated in step 401, the mapping relationship between the type of vehicle components and the mechanical parameters is extracted (for example, a certain type of vehicle bumper will inevitably produce a Y-shaped crack under a 30-degree angle collision). Secondly, this type of relationship is input into the generator of the adversarial generative network, and the generator is trained through supervised learning to understand the influence mode of mechanical parameters on damage diffusion. Then, according to the input combination of component type and mechanical parameters (such as "SUV car door + frontal impact"), the generator outputs various possible candidate damage diffusion rules (such as crack propagation rate, bifurcation density), simulating the possibility of damage development under different collision conditions. Finally, the candidate rules are stored in the temporary rule pool for screening.

[0074] 403. Screen the candidate damage diffusion rules through the discriminator of the adversarial generative network, and retain the rules that conform to the spatio-temporal distribution characteristics of the damage morphology change path in the non-fraudulent scenario, and integrate them into the dynamically generated rules; In step 403, the discriminator refers to a neural network module used to evaluate the authenticity of the rules. The spatio-temporal distribution characteristics refer to the development law of damage in time and space. The dynamically generated rules refer to the standardized damage development rule library retained after screening.

[0075] In the embodiments of the present application, first, the candidate damage diffusion rules generated in step 402 are input into the discriminator of the adversarial generative network. The discriminator evaluates whether the candidate rules conform to the physical laws of the non-fraudulent scenario by comparing the spatio-temporal characteristics of the historical real damage morphology change path (such as the curve of crack length increasing with time, the distribution of bifurcation positions). Secondly, the rules with a high matching degree with the real data are screened out (such as the crack propagation direction deviation is less than 5%), and the obviously abnormal rules (such as sudden reverse diffusion) are eliminated. Then, the retained rules are classified according to the component type and dynamically combined into a dynamically generated rule library that can adapt to different mechanical conditions to ensure that the rule library covers common collision scenarios.

[0076] 404. Input the starting position and diffusion direction in the comprehensive damage map into the dynamically generated rules, and combine the mechanical parameters corresponding to the type of vehicle components to generate a standard damage sequence that is consistent with the damage diffusion direction and morphological branch change in the comprehensive damage map.

[0077] In step 404, the standard damage sequence refers to the data of the damage evolution process that conforms to the development law of real accidents. The morphological branch change refers to the secondary structural change characteristics generated during the damage development process. Consistency refers to the state of high coincidence between the damage characteristics and the standard rules.

[0078] In the embodiment of the present application, first, the starting position (such as the dent point in the middle of the engine hood) and the propagation direction (the crack extends towards the front of the vehicle) of the current damage are extracted from the comprehensive damage atlas. Secondly, according to the type of the damaged component (such as "sedan engine hood"), the corresponding mechanical parameters (such as the collision intensity of 50 kN) are matched from the normal damage evolution law in step 401. Then, the dynamic generation rule library in step 403 is called, and the starting position, propagation direction, and mechanical parameters are input to generate a standard damage sequence that is consistent with the current damage propagation logic (such as predicting the crack propagation path and bifurcation position in the next 24 hours). Finally, the generated sequence is compared with the comprehensive damage atlas in real time to verify its consistency and provide a benchmark for fraud detection.

[0079] The following is a specific example: In the intelligent auto insurance anti-fraud evaluation system, the dynamic damage modeling system accurately identifies abnormal accident patterns. When an oblique collision between an SUV and a sedan occurs on an urban expressway, the system retrieves the accident database for the past ten years and integrates the deformation paths of the front bumper of the sedan in the speed range of 25 - 55 km / h, including the wrinkling directions of different materials and the law of solder joint fracture sequence. Through the simulation of the adversarial network generator, a candidate damage rule set for the high-strength steel front longitudinal beam during a 30-degree side impact is generated, covering twelve possible crack bifurcation patterns. The discriminator screens out eight rules that do not conform to the mechanical conduction law based on two thousand real side collision data and retains four groups of dynamically generated rules that match the physical characteristics. When the three-dimensional scanner transmits the Y-shaped damage atlas on the left side of the accident vehicle, the system inputs the damage starting point coordinates and the propagation vector into the dynamic rule library to match and generate the standard damage sequence for this vehicle type at the equivalent impact angle - a tree-shaped crack that should show unidirectional extension. The deviation between the actually detected bidirectional cross crack and the standard sequence triggers an alarm, and abnormal fluctuations are found in the impact acceleration curve in combination with the in-vehicle EDR data. The inspection drone captures rust cracks on the shock tower base that are not related to this collision, and the system automatically generates a spatio-temporal evolution comparison diagram to confirm that 30% of the damage belongs to the superposition of historical old injuries. After the case is closed, the newly added deformation rules of composite materials are injected into the model library to improve the recognition accuracy of the deformation of the battery compartment of new energy vehicles, forming an anti-fraud closed loop from multi-source law learning to intelligent loss assessment decision-making.

[0080] In summary, steps 401 to 404 achieve the intelligent generation and dynamic adaptation of the standard damage sequence. By integrating vehicle component types, mechanical parameters, and damage morphology change paths, the system constructs a damage evolution knowledge base in a fraud-free scenario. The introduction of the generative adversarial network effectively solves the problem of balancing the authenticity and diversity of the generation of standard damage rules, and the screening mechanism of the discriminator ensures the spatio-temporal consistency between the generated rules and the real damage diffusion law. This technology breaks through the limitations of traditional static rule bases and forms a standard damage sequence generation mechanism that can dynamically adapt to different collision scenarios, providing a highly reliable comparison benchmark for abnormal damage detection.

[0081] In some embodiments, as described in step 404, inputting the starting position and diffusion direction in the comprehensive damage map into the dynamic generation rule, and combining the type of vehicle component to match the corresponding mechanical parameters to generate a standard damage sequence consistent with the damage diffusion direction and morphological branch change in the comprehensive damage map, includes: 501. Input the starting position coordinates of the surface damage and the diffusion direction angle of the internal damage in the comprehensive damage map into the dynamic generation rule, and the dynamic generation rule includes the angle tolerance range of the damage diffusion direction and the change threshold of the number of morphological branches under specific mechanical parameters for different vehicle component types; In step 501, the angle tolerance range refers to the allowable deviation interval of the damage diffusion direction. The number of morphological branches refers to the number of secondary cracks or deformations generated during the damage development. The change threshold refers to the critical value that limits the number of morphological branches.

[0082] In the embodiment of the present application, first, extract the starting position coordinates of the surface damage (such as the longitude and latitude of the dent point at the left front end of the engine hood) and the diffusion direction angle of the internal damage (such as a 30-degree deviation angle of the crack extending to the right) from the comprehensive damage map. Secondly, input these parameters into a preset dynamic generation rule database, which stores the angle tolerance range (allowing a deviation of ±5 degrees) and the change threshold of the number of morphological branches (such as a maximum of 3 bifurcations) under specific mechanical parameters (such as a 50kN impact force) for different vehicle component types (such as doors, bumpers). Then, filter out the rule entries that match the current component type and damage angle through the rule matching engine to ensure that the subsequent generation process meets the physical constraint conditions. Finally, load the filtered rules into the damage sequence generation module to provide a parameter benchmark for the standardized damage simulation.

[0083] 502. According to the type of vehicle component in the comprehensive damage map, match the corresponding mechanical parameter combinations from the dynamic generation rule, and the mechanical parameter combinations include the collision angle, speed, and force range, which are used to constrain the generation path of the damage diffusion direction and morphological branches In step 502, the mechanical parameter combination refers to a specific combination of the collision angle, speed, and acting force. The collision angle refers to the angle between the acting force and the surface of the component. The acting force range refers to the effective interval of the magnitude of the collision force. The generation path refers to the simulated development trajectory of the standard damage sequence.

[0084] In the embodiment of the present application, first, according to the vehicle component type recorded in the comprehensive damage atlas (such as "SUV rear bumper"), the associated mechanical parameter combination is retrieved from the dynamic generation rule library. Secondly, the key parameters in the combination are extracted: the collision angle (such as 30 degrees front), the speed (such as 20 km / h), and the acting force range (such as 40 - 60 kN). These parameters are used to limit the logical boundaries of damage diffusion (for example, cracks shall not extend in the reverse direction under high-speed collisions). Then, the combinations that do not conform to the current damage characteristics are excluded through the parameter verification algorithm (such as low-speed parameters not matching the high-speed damage pattern) to ensure that the selected mechanical parameters are consistent with the actual situation of the damage atlas. Finally, the successfully matched parameter combination is marked as valid input and passed to the damage sequence generator.

[0085] 503. Based on the matched mechanical parameter combination, generate a standard damage sequence that is consistent with the starting position coordinates of the surface damage in the comprehensive damage atlas and whose internal damage diffusion direction angle is in line with the requirements. The number of morphological branches of the standard damage sequence is limited by the change threshold in the dynamic generation rule.

[0086] In step 503, "in line with" refers to the degree of matching between the damage characteristics and the standard rules. "Limited" refers to the constraint on the range of feature changes through the threshold.

[0087] In the embodiment of the present application, first, based on the mechanical parameter combination matched in step 502, the change threshold of the number of morphological branches of the current component type is obtained from the dynamic generation rule (such as allowing a maximum of 2 bifurcations). Secondly, starting from the starting position coordinates of the surface damage, along the path of the internal damage diffusion direction angle (such as extending along the 30-degree direction), a damage diffusion simulation path is generated in combination with the mechanical parameters. Then, the path deviation is corrected according to the angle tolerance range in the rule (such as automatically adjusting to 30 degrees when the actual angle is 28 degrees), and the number of bifurcations is determined according to the change threshold of the number of morphological branches (such as only generating 1 branch). Finally, a standard damage sequence that is completely consistent with the damage position, direction, and branch logic in the comprehensive damage atlas is output. This standard damage sequence includes the details of damage evolution in the time dimension (such as the crack extending 5 cm in the first hour and bifurcating in the second hour), serving as a comparison benchmark for fraud detection.

[0088] The following is a specific example: In the intelligent loss assessment system for new energy vehicle insurance, the dynamic damage modeling system accurately identifies abnormal damage to the battery compartment. When a rear-end collision of a new energy vehicle occurs in a certain tunnel, the laser scanner obtains the starting point of the depression on the rear surface of the vehicle and the crack propagation angle inside the battery compartment, and inputs them into the dynamically generated rule base covering the structure of the ternary lithium battery compartment. The system matches the mechanical parameter combination under the rear-end collision condition of 45 km / h according to the type of the aluminum battery compartment cover, and stipulates that the crack should conduct to the anti-collision beam in a 60-degree fan shape and the number of branches should not exceed three. Based on the standard damage sequence generated by these parameters, fishbone-shaped cracks that spread evenly from the impact point to both sides should be presented. The actual detection finds that there are five intersecting cracks inside the battery compartment and the propagation angle reaches 85 degrees, exceeding the morphological branch threshold of the dynamic rule base. The system automatically retrieves the in-vehicle sensor data at the moment of the accident, and reconstructs the collision process in combination with the tunnel monitoring video. It is found that there are traces of electrolyte crystallization on the west side of the battery compartment that are not caused by this impact. The intelligent loss assessment platform simultaneously starts the battery pressure test, verifies that the internal resistance value of the abnormal crack area increases abnormally, and confirms that it is the superposition of hidden damage caused by historical short circuits. Based on this, the claims settlement center rejects unreasonable claims and supplements the deformation characteristics of the new composite material battery compartment to the dynamic rule base, forming a closed-loop loss assessment mechanism from multi-source data fusion to intelligent damage comparison.

[0089] In summary, steps 501 to 503 achieve the accurate parameterization matching of the standard damage sequence and the collision scenario. By inputting the starting position and propagation direction of the comprehensive damage map into the dynamically generated rules, the system can generate a highly scenario-based standard damage sequence in combination with the vehicle component type and mechanical parameters. The introduction of the angle tolerance range and the morphological branch threshold not only preserves the natural physical law of damage propagation but also gives an elastic space for rule application. This technical solution significantly improves the matching accuracy between the standard damage sequence and the actual collision damage, provides a highly adaptable comparison basis for abnormal area identification, and enhances the scenario adaptation ability of fraud detection.

[0090] In some embodiments, as described in step 102, double-band scanning is performed on the damage features of different dimensions in the damage data to capture the microscopic deformation features of surface damage and the hidden crack distribution of internal damage, including: 601. Interrupt the first-band scanning during the double-band scanning of the surface damage area in the damage data. The first-band scanning focuses on the microscopic deformation details of the surface damage by adjusting the resolution parameters of the detection device, highlighting the deformation contour features of the surface damage and the texture differences in the deformation area to extract the microscopic deformation features of the surface damage; In step 601, the dual-band scanning refers to a scanning method that simultaneously adopts two different detection modes. The first-band scanning refers to a high-resolution detection mode for surface damage. The resolution parameter refers to the index of the detection device's ability to identify details. The microscopic deformation details refer to the fine structural change characteristics of the surface damage. The deformation contour feature refers to the geometric shape characteristics of the damage edge. The texture difference refers to the surface texture difference between the damaged area and the normal area.

[0091] In the embodiment of the present application, first, start the dual-band scanning process for the damage data. When performing the first-band scanning, actively interrupt other detection processes, centrally control the resolution parameter of the detection device (such as increasing it to micron-level accuracy), and perform high-precision focusing on the vehicle surface damage area. Secondly, focus on the microscopic deformation details of the surface damage through an optical sensor and an image enhancement algorithm - for example, magnify the local image of the paint crack, use an edge enhancement algorithm to sharpen the contour edge of the concave area, and enhance the texture difference in the surface deformation area (such as the reflectivity contrast between the scratch and the normal paint surface). Then, perform pixel-level analysis on the processed image: segment the boundary contour of the deformation area based on the watershed algorithm, and mark different deformation gradients (such as slight depression and deep crack) through a clustering algorithm. Finally, associate and store the extracted microscopic deformation features (such as crack length, deformation depth distribution) with the position coordinates to provide a data basis for subsequent damage correlation analysis.

[0092] 602. Perform the second-band scanning in the dual-band scanning on the internal damage area in the damage data. The second-band scanning focuses on the hidden crack propagation path of the internal damage by adjusting the penetration parameter of the detection device, captures the intermittent connection characteristics of the internal crack, and extracts the hidden crack distribution of the internal damage.

[0093] In step 602, the second-band scanning refers to a penetrative detection mode for internal damage. The penetration parameter refers to the index of the detection device's ability to detect the internal structure. The hidden crack propagation path refers to the extension trajectory of the crack inside the material. The intermittent connection feature refers to the morphological characteristics of the discontinuous distribution of the crack.

[0094] In the embodiment of the present application, first, after the first-band scanning in step 601 is completed, switch to the second-band scanning of the dual-band scanning, adjust the penetration parameter of the detection device (such as increasing the ultrasonic frequency to 20 MHz), and perform a penetration detection on the internal damage area. Secondly, by adjusting the echo absorption rate of the ultrasonic pulse, focus on the diffusion path of the hidden crack of the internal damage: use the time series analysis method to track the extension trajectory of the crack in different material layers, and capture the discontinuous connection characteristics of the internal crack (such as the jump fracture point of the crack between different metal layers). Then, combine the image reconstruction algorithm (such as back-projection reconstruction), convert the ultrasonic echo data into three-dimensional space coordinates, and mark the branch nodes and connection states of the crack. Finally, extract the complete crack distribution information (such as the length of the main crack path, the number of branches), associate it with the microscopic deformation characteristics of the surface damage, form a combined map of internal and external damage, and input it into the comprehensive analysis module of the next stage.

[0095] The following is a specific example: In the intelligent loss assessment system for commercial vehicle insurance, the multi-band damage detection system accurately identifies the structural damage of the vehicle frame. When a side scrape accident occurs to a logistics truck, the system activates the dual-band scanning device: first, adopt the high-resolution optical mode to focus on the surface scratch area of the outer steel plate of the vehicle frame, and enhance the identification of stress-induced wrinkle textures and micron-level metal fatigue marks that are invisible to the naked eye. Subsequently, switch to the depth penetration mode to scan the inside of the vehicle frame girder, and capture a hidden radioactive crack network and three discontinuous stress fracture points. After the multi-modal data fusion, the system finds that there is an abnormal angle between the trend of the surface wrinkle texture and the diffusion path of the internal crack, and this feature does not conform to the mechanical conduction law of the side scrape accident. The intelligent analysis platform retrieves the vehicle's historical maintenance records and finds that there are early welding repair marks inside the girder, forming a superimposed interference pattern with the cracks in this accident. The system automatically generates a three-dimensional damage spatio-temporal evolution map, marking that part of the damaged area is caused by the deterioration of historical repair defects. The insurance company rejects the excessive claim based on this report and updates the new damage mode of the composite structure vehicle frame to the anti-fraud model, realizing the three-dimensional loss assessment ability from millimeter-level surface deformation to centimeter-level internal cracks.

[0096] To sum up, steps 601 to 602 achieve multi-dimensional hierarchical scanning and accurate extraction of vehicle damage characteristics. By using the dual-band scanning technology to focus on the surface microscopic deformation and internal hidden cracks respectively, the system breaks through the resolution and penetration limitations of traditional detection methods. The high-resolution scanning in the first band enhances the ability to identify the texture differences of surface damage, and the depth penetration scanning in the second band reveals the diffusion path characteristics of internal cracks. This technical solution realizes the full-dimensional damage feature capture from the surface to the inside, providing high-precision basic data support for comprehensive damage analysis.

[0097] In some embodiments, as described in step 103, the spatio-temporal correlation between the microscopic deformation features and the distribution of the hidden cracks is carried out to form a comprehensive damage map including the correlation degree between surface damage and internal damage, including: 701. Carry out damage path correlation between the deformation contour features in the microscopic deformation features and the discontinuous connection features in the distribution of the hidden cracks, and obtain the spatio-temporal correlation relationship between the region with the largest deformation degree in the deformed region of the surface damage and the position where the crack first appears in the discontinuous connection features of the internal cracks; In step 701, the damage path correlation refers to an analysis method for establishing the correlation between the development of surface and internal damage. The region with the largest deformation degree refers to the local range where the surface damage is the most serious. The position where the crack first appears refers to the initial occurrence point of the internal damage. The spatio-temporal correlation relationship refers to the corresponding relationship between the development of damage in space and time.

[0098] In the embodiments of the present application, first, the deformation contour features (such as the depth gradient of the surface depression) in the extracted microscopic deformation features are spatially aligned with the discontinuous connection features (such as the crack bifurcation point coordinates) in the extracted distribution of the hidden cracks, and the deformed region of the surface damage is correlated with the starting position coordinates of the internal cracks through the three-dimensional coordinate system mapping technology. Secondly, based on the time stamp, the time correlation between the surface area with the largest deformation degree (such as the position with the deepest depression) and the position where the internal crack first appears (such as the fracture point between metal layers) after the accident collision is analyzed, for example, whether the surface deformation is earlier than the generation of the internal crack. Then, the spatial overlap degree and the time series matching degree between the two are calculated through the spatio-temporal correlation algorithm. If the surface deformation region and the starting point of the internal crack are at the same coordinate and the time line is continuous, it is marked as the spatio-temporal correlation relationship. Finally, an damage network map including the correlation between the deformation intensity and the crack propagation path is output, providing input for the screening of key regions.

[0099] 702. According to the spatio-temporal correlation relationship, screen out the part of the area covered by the region with the largest deformation degree that exceeds the preset threshold and mark it as the key damage region, and continuously repair the discontinuous connection features of the internal cracks corresponding to the key damage region in the diffusion direction to form the internal crack diffusion path feature; In step 702, the preset threshold refers to the critical area value for judging the key damage region. The key damage region refers to the local range that plays a decisive role in the overall damage development. The continuous repair refers to a processing method for logically complementing discontinuous cracks. The internal crack diffusion path feature refers to the complete crack development trajectory after repair.

[0100] In the embodiments of the present application, first, according to the spatio-temporal correlation relationship generated in step 701, the part with an area exceeding a preset threshold (such as the area of a single region ≥ 10 cm²) in the region with the largest surface deformation degree is screened out, and the boundary coordinates of these regions are extracted through an image segmentation algorithm (such as the region growing method). Secondly, the regions that meet the conditions are marked as key damage regions, and the discontinuous connection characteristics of the corresponding internal cracks (such as the coordinates of crack bifurcation points) are associated. Then, the discontinuous parts of the internal cracks are repaired: based on the crack propagation direction (such as extending along 30 degrees), an interpolation algorithm is used to complete the missing path nodes (such as filling the coordinates between the fracture points) to form a continuous internal crack propagation path feature. Finally, the repaired path is bound to the key damage region to generate a traceable damage propagation logic chain.

[0101] 703. Integrate the microscopic deformation characteristics and the internal crack propagation path characteristics to generate a comprehensive damage map including the correlation degree between the deformation intensity of the surface damage and the internal crack propagation path characteristics.

[0102] In step 703, the deformation intensity refers to a quantitative index of the surface damage degree. The correlation degree refers to the correlation degree between different damage characteristics. The comprehensive damage map refers to a complete model integrating surface and internal damage characteristics.

[0103] In the embodiments of the present application, first, the microscopic deformation characteristics (such as deformation depth, contour gradient) are fused with the internal crack propagation path characteristics (such as crack extension direction, bifurcation density) repaired in step 702: the surface deformation intensity is associated with the internal path propagation speed through a weight distribution algorithm. For example, a deep depression corresponds to rapid crack extension. Secondly, the two types of feature data are superimposed in three-dimensional space, and a heat map is used to map the surface deformation intensity distribution, while vector arrows are used to mark the internal crack propagation direction. Then, the coupling effect between the surface and internal damages (such as when the deformation intensity increases by 10%, the crack extension speed increases by 15%) is quantified through a correlation degree calculation model to generate a comprehensive damage map including the correlation degree between the deformation intensity of the surface damage and the internal crack propagation path characteristics. Finally, the map is input into the fraud detection module as the core basis for abnormal damage determination.

[0104] The following is a specific example: In the intelligent evaluation system for engineering vehicle insurance, the multi-modal damage correlation system accurately analyzes structural damage. After a concrete pump truck overturned at a construction site, the system initiated high-precision scanning: through microscopic imaging, it captured the fish-scale-like wrinkles on the surface of the outrigger cylinder, and at the same time, millimeter-wave penetration scanning detected intermittent stress cracks inside the load-bearing beam. The algorithm correlated the maximum deformation area on the surface with the anchor points of the star-shaped cracks that first appeared inside in terms of time and space, revealing that both originated from the stress overload point of the second hydraulic support. The system screened the annular areas with excessive surface wrinkle coverage, marked them as the key damage cores, and repaired the continuity of the internal intermittent cracks to restore the tree-like diffusion path extending to the chassis girder. By integrating the surface wrinkle density gradient and the internal crack diffusion topology, a three-dimensional map showing the correlation between damage intensity and crack depth was generated. The intelligent platform compared the historical operation data and found that the hydraulic pressure curve of the vehicle's support had abnormal fluctuations before the accident. Combining the spatial overlap characteristics of the damage and the hydraulic pipeline in the map, it was determined that there was cumulative damage caused by long-term overloading operation. Based on this, the insurance company attributed some of the damage to improper equipment maintenance and updated the new composite damage mode to the engineering vehicle claim settlement model, forming a multi-level loss assessment ability from microscopic deformation to macroscopic structure.

[0105] In summary, steps 701 to 705 achieved the spatio-temporal correlation analysis and comprehensive modeling of surface damage and internal cracks. By establishing the correlation between the deformation contour features and the crack diffusion path, the system can identify the key damage areas and repair the continuity features of the internal cracks. The integrated analysis of the surface deformation intensity and the internal crack diffusion path reveals the internal and external correlation law of damage propagation and constructs a comprehensive damage map covering damage intensity and diffusion path. This technical solution breaks through the limitation of the separation between the surface and the interior of traditional damage detection, forms a multi-dimensional and linked damage evolution cognitive system, and provides a more comprehensive analysis dimension for fraud risk determination.

[0106] In some embodiments, as described in step 106, when the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, it triggers the collaborative verification of multi-source data. By verifying the feature consistency of the damage data in the abnormal damage area, a claim settlement determination result including a maintenance fraud mark is output, including: 801. When the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, generate a multi-source data verification request, and the multi-source data verification request is used to trigger the data extraction of the comprehensive damage map and the historical maintenance records; In step 801, the multi-source data verification request refers to an instruction signal that triggers the cross-verification of multi-dimensional data. The determination threshold refers to the fraud determination critical value dynamically calculated according to the component type and collision mechanical parameters. Data extraction refers to the operation process of retrieving specific information from the database.

[0107] In the embodiment of the present application, first, when the system detects that the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of accident collisions, a multi-source data verification request is automatically generated, and this request calls the data interface module through a trigger mechanism. Secondly, after the verification request is started, a data extraction instruction is sent to the comprehensive damage atlas database to obtain the damage model of the current accident; at the same time, a query instruction is sent to the historical maintenance record library to retrieve detailed information such as the maintenance time and replaced components of the vehicle component in past accidents. Then, the extracted damage atlas and maintenance records are classified and stored according to the time stamp and component type, providing input for subsequent feature comparison. Finally, the status of the verification request is updated to "pending", triggering the next stage of the feature extraction process.

[0108] 802. According to the multi-source data verification request, extract the deformation contour features of the surface damage and the internal crack propagation path features of the abnormal damage area from the comprehensive damage atlas, and extract the damage repair features of the vehicle component in past accidents from the historical maintenance records; In step 802, the damage repair feature refers to the component repair morphological characteristics recorded in the historical maintenance records. Past accidents refer to the collision event records that the vehicle has previously experienced. The deformation contour feature refers to the geometric shape data of the surface damage edge. The internal crack propagation path feature refers to the development trajectory information of the internal crack in the material.

[0109] In the embodiment of the present application, first, according to the instruction of the multi-source data verification request, locate the abnormal damage area in the comprehensive damage atlas, and extract the deformation contour features (such as depression depth gradient, crack edge sharpness) of the surface damage and the internal crack propagation path features (such as main crack extension direction, bifurcation density) of this area through an image segmentation algorithm. Secondly, analyze the maintenance details of the same component in past accidents from the historical maintenance records, and extract the damage repair features (such as the change in the coating thickness of the component after repair, the distribution of welding points). Then, align the three types of features (deformation contour, crack path, repair record) according to the spatial coordinates and time nodes to generate a feature comparison matrix to ensure the consistency of data dimensions. Finally, input the matrix into the multi-source verification engine to start the consistency analysis process.

[0110] 803. Perform multi-source feature consistency comparison on the deformation contour features, internal crack propagation path features, and damage repair features, and calculate the spatial coincidence degree parameter between the deformation contour features and the internal crack path features in the abnormal damage area, and the morphological difference parameter between the current damage feature and the repair feature; In step 803, the multi-source feature consistency comparison refers to the cross-verification process of features from different data sources. The spatial coincidence degree parameter refers to the quantitative index of the matching degree between the surface and internal damage positions. The morphological difference parameter refers to the quantitative value of the difference degree between the current damage and the historical repair features.

[0111] In the embodiment of the present application, first, the deformation contour features and the internal crack propagation path features extracted in step 702 are spatially aligned, and the spatial coincidence degree parameter of the two (such as the coordinate deviation value between the deepest point of the surface depression and the starting point of the internal crack) is calculated through the three-dimensional coordinate mapping technology. Secondly, the deformation contour features of the current damage (such as the depth of the new depression) are morphologically compared with the historical damage repair features (such as the average thickness of the repaired area), and a morphological difference parameter (such as the depth difference ratio between the new and old depressions) is generated through the difference degree algorithm. Then, the two types of parameters (spatial coincidence degree, morphological difference) are initially compared with the preset thresholds (such as the coordinate deviation ≤ 2 cm, the depth difference ≤ 10%), and potential conflict points are marked. Finally, the comparison results are classified and stored by region to prepare for the final determination.

[0112] 804. Compare the spatial coincidence degree parameter and the morphological difference parameter with a preset coincidence threshold and a preset difference threshold. If the spatial coincidence degree parameter is lower than the preset coincidence threshold or the morphological difference parameter exceeds the preset difference threshold, it is determined that there is multi-source data inconsistency in the abnormal damage area, a maintenance fraud mark is generated, and a claim settlement determination result including the maintenance fraud mark is output according to the maintenance fraud mark.

[0113] In step 804, the preset coincidence threshold refers to the critical parameter value for judging position consistency. The preset difference threshold refers to the critical parameter value for judging morphological difference. Multi-source data inconsistency refers to the contradictory state between the features of different data sources. The maintenance fraud mark refers to the labeled data indicating fraud behavior. The claim settlement determination result refers to the final processing conclusion including the fraud determination.

[0114] In the embodiment of the present application, first, the calculated spatial coincidence degree parameter is compared with the preset coincidence threshold: if the parameter value is lower than the threshold (such as the actual deviation 3 cm > the preset 2 cm), it is determined as "spatially inconsistent"; at the same time, the morphological difference parameter is compared with the preset difference threshold: if the parameter value exceeds the threshold (such as the depth difference 15% > the preset 10%), it is determined as "morphologically abnormal". Secondly, according to the two types of comparison results, if any condition is satisfied (spatially inconsistent or morphologically abnormal), it is marked that there is multi-source data inconsistency in the abnormal damage area, and a maintenance fraud mark (such as "forged damage" or "over-maintenance") is generated through the decision tree model. Finally, the determination results of all marked areas are integrated to generate a claim settlement determination result including the maintenance fraud mark, with specific contradictory evidence (such as the coordinate deviation value, the conflict time point of the repair record) attached, and output to the claim settlement review system to complete the closed loop.

[0115] In summary, steps 801 to 804 implement a multi-source data collaborative verification mechanism for claim fraud determination. By triggering a multi-source data verification request, the system can synchronously extract damage features and historical repair records for consistency comparison. The dual verification mechanism of the spatial coincidence parameter and the morphological difference parameter effectively identifies the behavior patterns of artificially tampering with damage features. This technical solution constructs a multi-dimensional verification system covering the current damage state and historical repair features, significantly improving the anti-interference ability and result credibility of fraud determination, and forming a closed-loop claim risk prevention and control mechanism.

[0116] Figure 2 FIG. is a schematic structural diagram of a vehicle mutual assistance claim system based on multi-modal data fusion provided by an embodiment of the present application. As Figure 2 shown, the system includes: An integration module 21, configured to integrate damage data composed of a visual sensor, a depth sensing unit, and a repair record in a vehicle accident scenario, extract the types of vehicle components from the repair record, and obtain the mechanical parameters of the accident collision through a depth detector; A capture module 22, configured to perform dual-band scanning on damage features of different dimensions in the damage data to capture microscopic deformation features of surface damage and hidden crack distributions of internal damage; A correlation module 23, configured to perform spatio-temporal correlation on the microscopic deformation features and the hidden crack distributions to form a comprehensive damage map including the correlation degree between surface damage and internal damage; A generation module 24, obtaining the normal damage evolution law of vehicle components in a non-fraud scenario, and generating a standard damage sequence corresponding to the comprehensive damage map in combination with a generative adversarial network; A comparison module 25, configured to perform layer-by-layer comparison between the standard damage sequence and the damage data, extract abnormal damage regions in the damage data that deviate from the evolution law of the standard damage sequence, and calculate a fraud risk index based on the spatial density of the abnormal damage regions on the vehicle components; A verification module 26, configured to trigger collaborative verification of multi-source data when the fraud risk index exceeds a determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, and output a claim determination result including a repair fraud mark by collaboratively verifying the feature consistency of the damage data in the abnormal damage region.

[0117] Figure 2 The vehicle mutual assistance claim system based on multi-modal data fusion can execute Figure 1A vehicle mutual assistance claim settlement method based on multi-modal data fusion described in the illustrated embodiment, its implementation principle and technical effects will not be elaborated further. For a vehicle mutual assistance claim settlement system based on multi-modal data fusion in the above embodiment, the specific ways in which each module and unit perform operations have been described in detail in the embodiment related to this method, and will not be elaborated here.

[0118] In a possible design, Figure 2 A vehicle mutual assistance claim settlement system based on multi-modal data fusion in the illustrated embodiment can be implemented as a computing device, such as Figure 3 As shown, this computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, where the one or more computer instructions are called and executed by the processing component 32.

[0119] The processing component 32 is used for the above Figure 1 A vehicle mutual assistance claim settlement method based on multi-modal data fusion in the illustrated embodiment.

[0120] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0121] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0122] Of course, the computing device may also necessarily include other components, such as input / output interfaces, display components, communication components, etc.

[0123] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc.

[0124] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0125] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server. The above processing components, storage components, etc. can be basic server resources rented or purchased from a cloud computing platform.

[0126] The embodiments of the present application also provide a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above Figure 1 a vehicle mutual claim settlement method based on multi-modal data fusion shown in the embodiments.

[0127] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0128] The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0129] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A vehicle mutual compensation method based on multimodal data fusion, characterized in that: include: Integrate the damage data composed of visual sensors, depth sensing units and maintenance records in the vehicle accident scene, extract the type of vehicle parts from the maintenance records, and obtain the mechanical parameters of the accident collision through the depth detector; Performing dual-band scanning on damage features of different dimensions in the damage data to capture microscopic deformation features of surface damage and hidden crack distribution of internal damage; Correlating the microscopic deformation characteristics with the hidden crack distribution in time and space to form a comprehensive damage map including the correlation between surface damage and internal damage; Obtain the normal damage evolution law of vehicle parts in a non-fraud scenario, and generate a standard damage sequence corresponding to the comprehensive damage map in combination with a generative adversarial network; Comparing the standard damage sequence with the damage data layer by layer, extracting abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculating a fraud risk index based on the spatial density of the abnormal damage areas on the vehicle components; When the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, the collaborative verification of multi-source data is triggered. By collaboratively verifying the feature consistency of the damage data in the abnormal damage area, a claim determination result containing a repair fraud mark is output.

2. The method according to claim 1, characterized in that Also includes: Establish a user behavior baseline based on members’ payment records, driving behavior data, and historical accident data during the vehicle mutual assistance period; Extract key features from the collected driving behavior data of member vehicles and the vehicle status data when the accident is triggered, and match the key features with the user behavior baseline for deviation to generate a behavior abnormality coefficient. The key features include sudden acceleration, high-frequency sudden braking and night-time accident tendency features; When receiving a damage assessment form from a cooperative repair shop, the multi-dimensional coupling weight of the abnormal behavior coefficient and the fraud risk index is calculated based on the member's vehicle accident data, vehicle injury data, damage assessment amount and the fraud risk index, so as to adjust the ratio threshold of the transfer from the mutual aid pool to the repair shop according to the multi-dimensional coupling weight; If at least two of the abnormal behavior coefficient, fraud risk index and fund flow compliance mark exceed the ratio threshold, the mutual aid rights of the member will be frozen and manual review will be initiated. At the same time, the fraud behavior characteristics confirmed by the review will be reversely injected into the adversarial generation network to optimize the generation strategy of the standard damage sequence.

3. The method according to claim 1, characterized in that: Comparing the standard damage sequence with the damage data layer by layer, extracting abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculating the fraud risk index based on the spatial density of the abnormal damage areas on the vehicle parts, including: Performing a hierarchical comparison on the damage morphology change path of the vehicle component in the damage data and the damage morphology change path of the standard damage sequence, wherein the hierarchical comparison divides the damage area of ​​the vehicle into a plurality of sub-areas, and compares the damage diffusion direction and damage morphology similarity layer by layer in each sub-area; Based on the result of the hierarchical comparison, extract the sub-regions in the damage data where the damage diffusion direction deviates from the standard damage sequence by more than a preset angle threshold, mark the sub-regions as deviation regions, and calculate the abnormal damage features in each deviation region where the damage morphology similarity is lower than a preset similarity threshold; Counting the distribution positions and coverage of the deviation areas on the vehicle components, and merging adjacent deviation areas according to the continuity of the damage diffusion direction to obtain a set of abnormal damage areas; Calculate a spatial density parameter of the abnormal damage area according to the coverage area of ​​each deviation area in the abnormal damage area set on the vehicle component and the total area of ​​the corresponding component, wherein the spatial density parameter is proportional to the coverage area; The spatial density parameter and the deviation amplitude of the damage morphology similarity in the deviation area are weightedly superimposed to obtain a superimposed value as a fraud risk index.

4. The method according to claim 1, characterized in that: Obtain the normal damage evolution law of vehicle parts in a non-fraud scenario, and combine the adversarial generative network to generate a standard damage sequence corresponding to the comprehensive damage map, including: Obtain the normal damage evolution law of vehicle parts in a non-fraud scenario, which is obtained by integrating the type of vehicle parts, the mechanical parameters of the accident collision and the corresponding damage morphological change path, which records the starting position, diffusion direction and morphological branch change of the damage diffusion of vehicle parts after the collision; Based on the correspondence between the types of vehicle components and mechanical parameters in the normal damage evolution law, a generator of a generative adversarial network is combined to generate candidate damage diffusion rules for different types of vehicle components under specific mechanical parameters; The candidate damage diffusion rules are screened by using a discriminator of a generative adversarial network, and the rules that meet the spatiotemporal distribution characteristics of the damage morphology change path in a non-fraud scenario are retained and integrated into dynamic generation rules; The starting position and diffusion direction in the comprehensive damage map are input into the dynamic generation rule, and the corresponding mechanical parameters are matched in combination with the type of vehicle components to generate a standard damage sequence consistent with the damage diffusion direction and morphological branch changes in the comprehensive damage map.

5. The method according to claim 4, characterized in that The starting position and diffusion direction in the comprehensive damage map are input into the dynamic generation rule, and the corresponding mechanical parameters are matched in combination with the type of vehicle components to generate a standard damage sequence consistent with the damage diffusion direction and morphological branch changes in the comprehensive damage map, including: Inputting the starting position coordinates of the surface damage and the diffusion direction angle of the internal damage in the comprehensive damage map into the dynamic generation rule, wherein the dynamic generation rule includes the angle tolerance range of the damage diffusion direction and the change threshold of the number of morphological branches of different types of vehicle parts under specific mechanical parameters; According to the type of vehicle components in the comprehensive damage map, a corresponding mechanical parameter combination is matched from the dynamic generation rule, wherein the mechanical parameter combination includes a collision angle, a speed, and a force range, and is used to constrain the damage diffusion direction and the generation path of the morphological branch; Based on the matching mechanical parameter combination, a standard damage sequence is generated that is consistent with the coordinates of the surface damage starting position in the comprehensive damage atlas and consistent with the internal damage diffusion direction angle. The number of morphological branches of the standard damage sequence is limited by the change threshold in the dynamic generation rule.

6. The method according to claim 1, characterized in that Dual-band scanning is performed on the damage features of different dimensions in the damage data to capture the microscopic deformation characteristics of surface damage and the hidden crack distribution of internal damage, including: Performing dual-band scanning on the surface damage area in the damage data to interrupt the first-band scanning, wherein the first-band scanning focuses on the microscopic deformation details of the surface damage by adjusting the resolution parameters of the detection equipment, and highlights the deformation contour characteristics of the surface damage and the texture differences of the deformation area, so as to extract the microscopic deformation characteristics of the surface damage; The internal damage area in the damage data is scanned by the second band in the dual-band scan. The second band scan focuses on the hidden crack diffusion path of the internal damage by adjusting the penetration parameters of the detection equipment, captures the discontinuous connection characteristics of the internal cracks, and extracts the hidden crack distribution of the internal damage.

7. The method according to claim 1, characterized in that The microscopic deformation characteristics are temporally and spatially correlated with the hidden crack distribution to form a comprehensive damage map including the correlation between surface damage and internal damage, including: The deformation profile feature in the microscopic deformation feature and the discontinuous connection feature in the hidden crack distribution are associated with the damage path to obtain the spatiotemporal correlation between the area with the largest deformation degree in the deformation area of ​​the surface damage and the position where the crack first appears in the discontinuous connection feature of the internal crack; According to the spatiotemporal correlation, the portion of the area with the largest deformation degree exceeding a preset threshold is screened and marked as a key damage area, and the discontinuous connection features of the internal cracks corresponding to the key damage area are repaired continuously in the diffusion direction to form internal crack diffusion path features; The microscopic deformation characteristics and the internal crack diffusion path characteristics are integrated to generate a comprehensive damage map including the correlation between the deformation intensity of surface damage and the internal crack diffusion path characteristics.

8. The method according to claim 1, characterized in that When the fraud risk index exceeds the determination threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision, the collaborative verification of multi-source data is triggered. By collaboratively verifying the feature consistency of the damage data in the abnormal damage area, the claim determination result containing the repair fraud mark is output, including: When the fraud risk index exceeds a determination threshold dynamically adjusted according to the type of vehicle components and mechanical parameters of the accident collision, a multi-source data verification request is generated, wherein the multi-source data verification request is used to trigger data extraction of the comprehensive damage map and historical maintenance records; According to the multi-source data verification request, the deformation contour characteristics of the surface damage and the internal crack diffusion path characteristics of the abnormal damage area are extracted from the comprehensive damage map, and the damage repair characteristics of the vehicle component in the past accidents are extracted from the historical maintenance records; Performing a multi-source feature consistency comparison on the deformation contour feature, the internal crack diffusion path feature and the damage repair feature, and calculating the spatial coincidence parameters of the deformation contour feature and the internal crack path feature in the abnormal damage area, as well as the morphological difference parameters of the current damage feature and the repair feature; The spatial overlap parameter and the morphological difference parameter are compared with the preset overlap threshold and the preset difference threshold. If the spatial overlap parameter is lower than the preset overlap threshold or the morphological difference parameter exceeds the preset difference threshold, it is determined that there is multi-source data inconsistency in the abnormal damage area, and a maintenance fraud mark is generated. According to the maintenance fraud mark, a claim determination result including the maintenance fraud mark is output.

9. A vehicle mutual compensation system based on multimodal data fusion, characterized in that: include: An integration module, for integrating damage data composed of visual sensors, depth sensing units and maintenance records in vehicle accident scenes, extracting the types of vehicle components from the maintenance records, and obtaining mechanical parameters of accident collisions through depth detectors; A capture module, used for performing dual-band scanning on damage features of different dimensions in the damage data, so as to capture microscopic deformation features of surface damage and hidden crack distribution of internal damage; A correlation module, used for temporally and spatially correlating the microscopic deformation characteristics with the hidden crack distribution to form a comprehensive damage map including the correlation between surface damage and internal damage; A generation module, which obtains the normal damage evolution law of vehicle parts in a non-fraud scenario, and generates a standard damage sequence corresponding to the comprehensive damage map in combination with a generative adversarial network; A comparison module, for comparing the standard damage sequence with the damage data layer by layer, extracting abnormal damage areas in the damage data that deviate from the evolution law of the standard damage sequence, and calculating a fraud risk index based on the spatial density of the abnormal damage areas on the vehicle parts; The verification module is used to trigger the collaborative verification of multi-source data when the fraud risk index exceeds the judgment threshold dynamically adjusted according to the type of vehicle components and the mechanical parameters of the accident collision. By collaboratively verifying the feature consistency of the damage data in the abnormal damage area, the claim judgment result containing the repair fraud mark is output.

10. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a vehicle mutual compensation method based on multimodal data fusion as described in any one of claims 1 to 8.

Images