Vehicle intelligent detection method and system based on vehicle body gap and non-vulnerable part production date

Abstract

The invention provides a vehicle intelligent detection method and system based on vehicle body gaps and production dates of non-vulnerable parts, and relates to the technical field of vehicle detection. According to the method, displacement and environment data of the surface of a vehicle are collected through a laser array sensor, preprocessing and feature extraction are carried out, a feature matrix is constructed, outliers are recognized through a mahalanobis distance algorithm, a production date and a registration date are compared, and the comprehensive health index of the vehicle is calculated. The method provides an efficient and reliable full-dimensional analysis tool for second-hand car evaluation, insurance fraud prevention and manufacturing quality inspection, and has remarkable industrial application value.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle detection, and in particular to a vehicle intelligent detection method and system based on vehicle body gaps and production dates of non-wearing parts. Background Art

[0002] Currently, vehicle structure inspection has important applications in the fields of used car evaluation, accident vehicle identification, and automobile manufacturing quality control. Traditional inspection methods mainly rely on manual visual inspection and single-point dimension measurement using tools such as vernier calipers. The core focus is on whether the width of the gaps in vehicle covering parts is within a reasonable range (for example, if it is too small, it may cause the parts to be squeezed and paint peeled off; if it is too large or asymmetrical, it may reflect assembly process defects or potential repair traces). However, existing technologies generally have problems such as low automation and single data dimension. For example, they only record the gap width values at discrete positions, lack continuous analysis of the gap morphological characteristics (such as curvature and gradient), and fail to establish a three-dimensional spatial coordinate association, resulting in the inspection results being difficult to fully reflect the health status of the vehicle structure.

[0003] Existing detection technologies face the following key bottlenecks: (1) Significant limitations of manual inspection: The error rate of empirical judgment exceeds 40%, a single-door inspection takes more than 5 minutes, and the naked eye recognition threshold (>0.5mm) cannot capture micron-level deformation; (2) Serious lack of data dimension: Existing methods only obtain the static dimensions of a single point, lack continuous trajectory data and three-dimensional spatial position association (position error>3mm), and cannot analyze the dynamic change characteristics of the gap; (3) Ambiguous judgment standards: The industry lacks a unified quantitative tolerance standard (different brands allow tolerance differences of up to ±0.3mm), and the misjudgment rate between manual judgment and actual vehicle conditions is as high as 32%. These problems lead to low efficiency and insufficient reliability in vehicle damage judgment, especially in accident vehicle identification and repair trace detection. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a vehicle intelligent detection method and system based on vehicle body gaps and the production date of non-wearing parts. Through high-precision sensing, multi-dimensional feature fusion and dynamic intelligent analysis, it realizes full-dimensional accurate detection and health assessment of vehicle operating conditions, significantly improves detection efficiency, accuracy and industrial applicability, and provides a revolutionary solution for automotive aftermarket services and manufacturing quality control.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts, comprising:

[0007] Obtain the production date data of the target vehicle's non-wearing parts, and use a three-axis linkage slide to drive the laser array sensor to move along the target vehicle's surface to simultaneously collect vehicle gap displacement data, linear encoder positioning data, and nanometer-level environmental parameters of the grating scale;

[0008] Preprocessing the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data;

[0009] Performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features;

[0010] Calculate outliers in the multidimensional feature matrix based on the Mahalanobis distance algorithm, and determine abnormal areas using a dynamic threshold;

[0011] The production date data is compared with the registration date of the target vehicle to obtain a comparison result, and a comprehensive vehicle health index is determined based on the comparison result, the distribution density and severity of the outliers.

[0012] Preferably, it also includes:

[0013] The determination result of the abnormal area is associated with the corresponding three-dimensional coordinates according to the second preprocessing data, and a three-dimensional gap heat map of the vehicle surface is generated based on the coordinates of the associated abnormal area and the severity of the outlier.

[0014] Preferably, it also includes:

[0015] Based on a linear regression model, the temperature drift error of the laser array sensor is corrected according to the nanoscale environmental parameters.

[0016] Preferably, preprocessing the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data includes:

[0017] Considering the light and shadow effect of the gap between vehicles, the wavelet noise reduction algorithm is used to remove the high-frequency noise of the displacement data to obtain denoised data. The calculation formula of the denoised data is: in, is the denoised data; W j,k is the kth detail coefficient of the original displacement data under the jth layer wavelet decomposition; ψ j,k (x, y) is the wavelet basis function, corresponding to the decomposition scale j and position k; V J (x, y) is the approximate coefficient under the maximum decomposition scale J, I(x, y) is the local light intensity parameter, which is normalized to the range of 0, 1 by the light sensor or image grayscale value. is the gradient amplitude of the vehicle surface reflectivity, which reflects the degree of light and shadow mutation; α is the light sensitivity weight coefficient, α∈[0.2,0.5]; β is the reflectivity gradient suppression coefficient, β∈[1.0,2.0]; Thresh(·) is the threshold function, Among them, λ is the dynamic threshold, and the calculation formula of λ is: σ is the estimated value of the noise standard deviation, N is the signal length of the displacement data, and γ is the preset illumination dependency coefficient;

[0018] Smoothing the denoised data using a Savitzky-Golay filter to obtain smoothed data; the Savitzky-Golay filter has a window length of 21 and a polynomial order of 3;

[0019] The positioning data of the linear encoder and the axial displacement data of the grating ruler are fused through the Kalman filter algorithm to generate three-dimensional space coordinate data to obtain second pre-processed data.

[0020] Preferably, the state equation and observation equation of the Kalman filter algorithm are expressed as follows:

[0021]

[0022] Among them, x k is the state vector, Among them, X enc 、Y enc are the plane coordinates measured by the linear encoder; Z grating is the axial displacement measured by the grating ruler; are the velocity components of each axis respectively; C k is the compensation coefficient matrix, which is related to the thermal expansion coefficient of the grating scale material; f(T) is the temperature drift function, f(T)=α'(T-T0)+β'(T-T0) 2 ; Wherein, T is the real-time ambient temperature; T0 is the calibration reference temperature; α' and β' are the first-order and second-order temperature coefficients, respectively, which are determined by calibration experiments;

[0023] R k is the dynamic weight observation noise covariance, Among them, σ enc is the inherent noise of the linear encoder, σ enc =1μm;σ grating is the inherent noise of the grating scale, σ grating =1nm; γ is the temperature sensitivity coefficient; ΔT is the temperature deviation, ΔT=T-T0; S / N is the real-time estimated value of the grating scale signal-to-noise ratio; κ is the signal-to-noise ratio suppression coefficient, κ=10.1)

[0024] Kk is the multi-scale fusion gain, is the dynamic weight matrix, W k =diag(w X ,w Y ,w Z ),in, is the axial displacement gradient, w X 、w Y and w Z are the plane coordinate weights of each axis, w X =w Y =1-w Z ;P k|k-1 is the uncertainty of the state estimation based on historical data before time k, Among them, A k is the state transfer matrix, which is used to describe the system dynamic model, P k-1 is the posterior estimated covariance matrix of the previous moment, reflecting the uncertainty of the historical state; Q k is the process noise covariance matrix, which represents the impact of external interference on the system, H k To convert the state vector x k Mapping to observation space z k The matrix of .

[0025] Preferably, feature extraction is performed based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and a multidimensional feature matrix is constructed based on the multidimensional gap features, including:

[0026] The first pre-processed data is determined as the gap width, and the width change rate of adjacent measurement points is determined according to the gap width to obtain a width mutation gradient; the width mutation gradient G i The calculation formula is: Among them, W i is the i-th measurement point (X i ,Y i ) corresponds to the gap width, X i and Y i are the horizontal and vertical coordinate values of the i-th measurement point respectively;

[0027] Determine the symmetry axis equation based on the vehicle design model or point cloud fitting;

[0028] Based on the symmetry axis equation, the left measurement point (X L ,Y L ) is mapped to the right symmetric coordinate (X R ,Y R );

[0029] According to the left measurement point (X L ,Y L ) and the right symmetrical coordinate (X R ,Y R ) Determine the left width W L and right width W R ;

[0030] According to the left width W L and the right side width W R Calculate the symmetry deviation S i ;S i The calculation formula is: λ' is the penalty for coordinate asymmetry, λ' = 0.1;

[0031] Based on the three-dimensional coordinates of the second preprocessed data, a cubic spline curve C(t) is fitted along the gap trajectory; the parameterized variable t of each cubic spline interpolation curve is i , calculate the curvature K i ;in,

[0032] Calculate the absolute value of the deviation between the curvature and the mean to obtain the curvature anomaly value; the curvature anomaly value K dev The calculation formula is: K dev =|K i -μ K |; where μ K is the global mean curvature;

[0033] Determining the width mutation gradient, the curvature anomaly value, and the symmetry deviation as the multidimensional gap features;

[0034] The multidimensional gap features are subjected to feature normalization and matrix construction to obtain the multidimensional feature matrix.

[0035] Preferably, calculating the outliers in the multidimensional feature matrix based on the Mahalanobis distance algorithm and determining the abnormal area by a dynamic threshold include:

[0036] Performing Z-score normalization on each feature column in the multidimensional feature matrix to obtain a normalized matrix;

[0037] Calculating a covariance matrix ∑ in the standardized matrix;

[0038] For each sample x in the normalized matrix i , calculate the Mahalanobis distance; the Mahalanobis distance D Mahalanobis (x i ) is calculated as: Among them, μ is the mean vector, N is the total number of samples in the multidimensional feature matrix;

[0039] The mean μ based on the Mahalanobis distance D and standard deviation σ D Set the dynamic threshold Threshold; where Threshold = μ D +3σ D ,

[0040] If D Mahalanobis (x i )>Threshold, the sample is marked as an outlier; the label of the outlier is:

[0041] Bind the outlier labels to the three-dimensional coordinates to generate a set of abnormal regions; the set of abnormal regions for: X i 、Y i , Z i They are respectively the X-axis coordinate, Y-axis ordinate and Z-axis coordinate of the three-dimensional coordinate.

[0042] Preferably, the distribution density is the number of outliers per unit area; and the severity is the mean of the Mahalanobis distances of all outliers.

[0043] Preferably, the manufacturing date data is compared with the registration date of the target vehicle to obtain a comparison result, and a comprehensive vehicle health index is determined based on the comparison result, the distribution density and severity of the outliers, including:

[0044] Comparing the obtained production date with the preset registration date to calculate the useful life of the component;

[0045] Calculating the deviation between the service life of the component and the overall service life of the target vehicle to obtain the comparison result CR;

[0046] The vehicle comprehensive health index is calculated based on the distribution density of the outliers, the severity of the abnormal area and the comparison result CR; the calculation formula of the vehicle comprehensive health index CHI is: CHI=∑(w i ·f i )+α”·distribution density of outliers·severity+β”·CR; where w i is the weight of the multidimensional gap feature, where the weight of the width mutation gradient is 0.35, the weight of the symmetry deviation is 0.28, and the weight of the curvature anomaly is 0.22), f iis the eigenvalue after feature normalization of the multidimensional gap feature, α″ is the outlier density weight, α″=0.15, β is the component age deviation weight, β=0.1.

[0047] A vehicle intelligent detection system based on vehicle body gaps and production dates of non-wearing parts, comprising:

[0048] The data acquisition unit is used to obtain the production date data of the non-wearing parts of the target vehicle and drive the laser array sensor to move along the surface of the target vehicle through a three-axis linkage slide to synchronously collect the displacement data of the vehicle gap, the positioning data of the linear encoder, and the nanometer-level environmental parameters of the grating ruler;

[0049] a data preprocessing unit, configured to preprocess the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data;

[0050] a feature matrix construction unit, performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features;

[0051] An abnormality determination unit, configured to calculate outliers in the multidimensional feature matrix based on a Mahalanobis distance algorithm and determine abnormal areas using a dynamic threshold;

[0052] The detection index calculation unit is used to compare the production date data with the registration date of the target vehicle to obtain a comparison result, and determine the vehicle comprehensive health index based on the comparison result, the distribution density and severity of the outliers.

[0053] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0054] The present invention provides a vehicle intelligent detection method and system based on vehicle body gaps and the production date of non-wearing parts. Through laser array sensors, linear encoders, grating rulers and multi-dimensional data fusion technology, micron-level precise measurement of vehicle gaps is achieved, and combined with the intelligent comparison of the production date of non-wearing parts and the vehicle registration date, the detection efficiency and comprehensiveness are significantly improved. Through the Mahalanobis distance algorithm determined by dynamic thresholds, the system can identify abnormal areas in the multi-dimensional feature matrix, quantify the density and severity of outliers, and finally generate a comprehensive health index, which can not only reflect the damage to the vehicle body structure in real time, such as assembly errors and collision deformations, but also trace the historical abnormalities of non-wearing parts, such as tampering or replacement records. The present invention can greatly improve the accuracy of accident vehicle identification, and the detection speed is greatly improved compared with traditional manual methods. At the same time, it supports three-dimensional heat map visualization, providing an efficient and reliable full-dimensional analysis tool for used car evaluation, insurance anti-fraud and manufacturing quality inspection, and has significant industrial application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0057] Figure 2 A flowchart for determining a comprehensive vehicle health index according to an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0060] The purpose of this invention is to provide a vehicle intelligent detection method and system based on vehicle body gaps and production dates of non-wearing parts, providing an efficient and reliable full-dimensional analysis tool for used car evaluation, insurance anti-fraud and manufacturing quality inspection, and has significant industrial application value.

[0061] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0062] Figure 1 A flow chart of the method provided in the embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides

[0063] A vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts, comprising:

[0064] Step 100: Obtain the production date data of the non-wearing parts of the target vehicle, and drive the laser array sensor to move along the surface of the target vehicle via a three-axis linkage slide to synchronously collect displacement data of the vehicle gap, positioning data of the linear encoder, and nanometer-level environmental parameters of the grating ruler;

[0065] Step 200: Preprocess the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data;

[0066] Step 300: performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features;

[0067] Step 400: Calculate outliers in the multidimensional feature matrix based on the Mahalanobis distance algorithm, and determine abnormal areas using a dynamic threshold;

[0068] Step 500: Compare the production date data with the registration date of the target vehicle to obtain a comparison result, and determine the vehicle comprehensive health index based on the comparison result, the distribution density and severity of the outliers.

[0069] Specifically, in step 100 of this embodiment, the production date data of non-consumable parts is obtained in two ways: first, by parsing the vehicle identification number (VIN) and connecting to a cloud-based manufacturer database to query production information for the vehicle frame, engine block, and transmission housing; and second, by reading data from RFID tags embedded in non-consumable parts using near-field communication technology. During data collection, a three-axis linkage slide system drives a Keyence LJ-V7080 laser displacement sensor along the vehicle surface. The sensor collects gap displacement data in real time at a sampling rate of 50kHz. The linear encoder uses a HIWIN HGH25CA slide with a planar positioning resolution of 1 micron, and the grating scale uses the Renishaw RESOLUTE series with an axial displacement measurement resolution of 1 nanometer. Temperature and mechanical vibration parameters are also collected simultaneously. A multi-channel data synchronization acquisition card ensures that the acquisition time synchronization error of displacement, positioning, and environmental parameters is less than 1 microsecond.

[0070] During data acquisition, the three-axis slide is controlled by a servo motor, achieving a repeatable positioning accuracy of ±0.005 mm, ensuring precise movement of the laser sensor along the preset path. A linear scale monitors the slide's axial displacement error in real time and compensates for temperature drift using a linear regression model. The compensation coefficient is determined based on calibration experiments. The two-dimensional coordinates output by the linear encoder are combined with the axial data from the linear scale using a Kalman filter algorithm to generate three-dimensional spatial coordinate data. The raw displacement data collected by the laser sensor undergoes wavelet noise reduction preprocessing to retain the signal in the valid frequency band. All data is stored in real time in a local database by the Xilinx Zynq-7000 SoC control unit and bound to the production date to form a complete inspection data set.

[0071] This implementation solves the problems of traditional inspection, such as strong reliance on manual experience and the single-dimensionality of single-point measurement data, through the collaborative operation of high-precision sensors and the fusion of multi-source data. VIN parsing and RFID dual-channel data acquisition mechanisms ensure the authenticity and tamper-resistance of production information for non-consumable parts. The nanometer-level error compensation design of the three-axis slide rail and grating scale improves displacement measurement accuracy to the micron level, overcoming interference from ambient temperature fluctuations and mechanical vibration. Synchronous data acquisition and real-time processing technology significantly shorten inspection time, allowing a full vehicle scan to be completed in under three minutes, providing a highly reliable data foundation for subsequent feature extraction and health assessment.

[0072] Preferably, it also includes:

[0073] The determination result of the abnormal area is associated with the corresponding three-dimensional coordinates according to the second preprocessing data, and a three-dimensional gap heat map of the vehicle surface is generated based on the coordinates of the associated abnormal area and the severity of the outlier.

[0074] Specifically, this embodiment binds the determination result of the abnormal area with the three-dimensional spatial coordinates in the second preprocessed data to form an abnormal area data set containing the coordinates of the abnormal points and their corresponding Mahalanobis distance values. Based on the three-dimensional model of the vehicle surface, the detection area is divided into grid units with a resolution of 0.1 mm by 0.1 mm, and the center point coordinates of each grid unit are determined by a three-dimensional space interpolation algorithm. For grid points that are not directly measured, an inverse distance weighted interpolation algorithm is used to calculate the interpolation weight based on the Mahalanobis distance value and spatial position relationship of adjacent abnormal points, where the weight of areas with high outlier density and high severity is significantly increased. During the interpolation calculation, the axial accuracy data of the grating ruler is introduced as a constraint to ensure the consistency of the interpolation result with the physical space.

[0075] Based on the Mahalanobis distance interpolation results of the grid cells, the degree of vehicle surface anomaly is mapped to a color gradient. The color mapping rules are set as follows: when the grid cell's comprehensive anomaly score is greater than or equal to 0.8, it is marked red to indicate a severe anomaly; scores between 0.5 and 0.8 are marked yellow to indicate a warning; and scores less than 0.5 are marked green to indicate normality. The comprehensive anomaly score is calculated by multiplying the outlier density and the average Mahalanobis distance and is normalized to fit the color threshold range. The resulting 3D heat map is displayed overlaid with the vehicle structure model, supporting multi-view rotation and local zoom operations. A test report containing a list of anomaly coordinates, score details, and repair recommendations is also output.

[0076] This implementation achieves high-precision spatial mapping of detection results through coordinate binding and gridded interpolation of abnormal regions, resolving the issue of ambiguous abnormality location in traditional methods. The inverse distance weighted interpolation algorithm, combined with grating scale data constraints, effectively avoids interpolation distortion caused by sparse measurement points. Color gradient rules are dynamically set based on quantitative scoring, intuitively reflecting the distribution of abnormality severity and supporting rapid decision-making. The combination of 3D visualization output and structured reporting significantly improves the interpretability and industrial applicability of detection results.

[0077] Preferably, it also includes:

[0078] Based on a linear regression model, the temperature drift error of the laser array sensor is corrected according to the nanoscale environmental parameters.

[0079] Specifically, the temperature drift error of the laser array sensor is corrected by a linear regression model. First, in a constant temperature laboratory, the sensor is placed in a temperature-controlled environment, and the sensor zero offset data at different temperatures in the range of -10 degrees Celsius to 50 degrees Celsius is collected by a high-precision temperature control device at intervals of 1 degree Celsius. The grating ruler synchronously records the ambient temperature and axial displacement error to establish a corresponding relationship data set between temperature and displacement error. The least squares method is used to fit the linear regression equation, and the model expression is that the sensor error is equal to the temperature coefficient multiplied by the current temperature plus the reference offset. The temperature coefficient and the reference offset are determined by calibration of experimental data and stored in the flash memory of the control unit. After calibration, the model can receive the temperature parameters provided by the grating ruler in real time and dynamically calculate the drift compensation value at the current temperature.

[0080] During the actual detection process, the grating ruler continuously monitors the ambient temperature data and updates the linear regression model with a period of 10 milliseconds. The control unit calls the pre-stored model parameters based on the current temperature value and calculates the corresponding displacement error compensation. The raw displacement data collected by the laser sensor is synchronously superimposed with this compensation value during the preprocessing stage to eliminate systematic deviations caused by temperature changes. The compensated data undergoes wavelet noise reduction processing to retain the effective signal frequency band. To verify the compensation effect, zero-point calibration is regularly performed using standard gauge blocks to ensure that the measurement error within the full temperature range does not exceed plus or minus 0.005 mm. This correction mechanism enables the system to maintain micron-level measurement accuracy even under ambient temperature fluctuations of -5 degrees Celsius to 45 degrees Celsius.

[0081] This implementation effectively mitigates the impact of temperature changes on laser sensor accuracy by combining experimental calibration with real-time compensation. A linear regression model simplifies complex thermodynamic deformations into quantifiable linear relationships, significantly reducing computational complexity while ensuring compensation efficiency. Dynamic temperature monitoring and periodic compensation mechanisms enable the system to adapt to rapid temperature changes found in workshop environments. Verification using standard gauge blocks has reduced the sensor's zero-point drift error to less than 5% of its original value after compensation, providing reliable assurance for accurate vehicle gap detection.

[0082] Preferably, preprocessing the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data includes:

[0083] Considering the light and shadow effect of the gap between vehicles, the wavelet noise reduction algorithm is used to remove the high-frequency noise of the displacement data to obtain denoised data. The calculation formula of the denoised data is: in, is the denoised data; W j,k is the kth detail coefficient of the original displacement data under the jth layer wavelet decomposition; ψ j,k (x, y) is the wavelet basis function, corresponding to the decomposition scale j and position k; V J (x, y) is the approximate coefficient under the maximum decomposition scale J, I(x, y) is the local light intensity parameter, which is normalized to the range of 0, 1 by the light sensor or image grayscale value. is the gradient amplitude of the vehicle surface reflectivity, which reflects the degree of light and shadow mutation; α is the light sensitivity weight coefficient, α∈[0.2,0.5]; β is the reflectivity gradient suppression coefficient, β∈[1.0,2.0]; Thresh(·) is the threshold function, Among them, λ is the dynamic threshold, and the calculation formula of λ is: σ is the estimated value of the noise standard deviation, N is the signal length of the displacement data, and γ is the preset illumination dependency coefficient;

[0084] Smoothing the denoised data using a Savitzky-Golay filter to obtain smoothed data; the Savitzky-Golay filter has a window length of 21 and a polynomial order of 3;

[0085] The positioning data of the linear encoder and the axial displacement data of the grating ruler are fused through the Kalman filter algorithm to generate three-dimensional space coordinate data to obtain second pre-processed data.

[0086] Specifically, to address the light and shadow interference in vehicle gaps, the system uses an adaptive filtering algorithm based on wavelet denoising to process the raw displacement data. In implementation, the displacement signal is decomposed into five layers using the Daubechies 5 wavelet basis, extracting detail coefficients and approximation coefficients at each scale. Local illumination intensity parameters are measured in real time by photosensors integrated on the sides of the laser sensor and normalized to a range of 0 to 1, reflecting the illumination intensity in the detection area. The vehicle surface reflectivity gradient amplitude is calculated from grayscale images captured by a high-resolution industrial camera, and edge gradient information is extracted using the Sobel operator. The illumination sensitivity weight coefficient is set to 0.3, and the reflectivity gradient suppression coefficient is set to 1.5, determined through experimental calibration to balance noise reduction and edge preservation. When calculating the dynamic threshold, the noise standard deviation estimate is taken from the highest-frequency layer coefficient after wavelet decomposition. The signal length N is the number of sampling points in the current scanning area, and the illumination dependency coefficient γ is fixed at 0.2. The threshold function uses hard thresholding to filter out noise components with an absolute value less than the threshold. The denoised signal is smoothed by a Savitzky-Golay filter with a window length of 21 points covering more than three times the actual gap width, and the polynomial order is 3 to ensure that the curvature characteristics are preserved.

[0087] The plane positioning data of the linear encoder and the axial displacement data of the grating ruler are fused through an improved Kalman filter algorithm. The process noise covariance matrix is calibrated based on the slide rail repeatability accuracy (±0.005 mm), and the observation noise covariance matrix is dynamically adjusted according to the inherent errors of the linear encoder (1 micron) and the grating ruler (1 nanometer). The state vector contains plane coordinates, axial displacement and its first-order derivative. The observation matrix is designed as a 3×6 matrix to match the three-dimensional coordinate output requirements. The axial displacement gradient constraint is introduced when calculating the Kalman gain. When the grating ruler detects a sudden gradient caused by mechanical impact, its weight coefficient is automatically reduced to below 0.5 to avoid outliers affecting the fusion accuracy. The fused three-dimensional coordinate data is mapped to the vehicle surface mesh model through the Bresenham algorithm. The mesh resolution is 0.1 mm, and the uncovered areas are filled using bilinear interpolation.

[0088] This implementation significantly reduces high-frequency noise interference in reflective areas while preserving the micron-scale deformation characteristics of the gap edges through a wavelet denoising algorithm coupled with light and shadow parameters. The window and order selection of the Savitzky-Golay filter has been experimentally verified to smooth noise while avoiding excessive signal distortion. The Kalman filter fusion mechanism effectively coordinates the robustness of planar positioning and the nanometer-level precision of axial displacement through dynamic weight adjustment, and the generated three-dimensional coordinate space error is less than 0.01 mm. Experimental calibration methods for light sensitivity and gradient suppression coefficient ensure that the algorithm has stable performance in complex scenarios such as strong outdoor light and workshop shadows.

[0089] Preferably, the state equation and observation equation of the Kalman filter algorithm are expressed as follows:

[0090]

[0091] Among them, x k is the state vector, Among them, X enc 、Y enc are the plane coordinates measured by the linear encoder; Z grating is the axial displacement measured by the grating ruler; are the velocity components of each axis respectively; C k is the compensation coefficient matrix, which is related to the thermal expansion coefficient of the grating scale material; f(T) is the temperature drift function, f(T)=α'(T-T0)+β'(T-T0) 2 ; Wherein, T is the real-time ambient temperature; T0 is the calibration reference temperature; α' and β' are the first-order and second-order temperature coefficients, respectively, which are determined by calibration experiments;

[0092] R k is the dynamic weight observation noise covariance, Among them, σ enc is the inherent noise of the linear encoder, σ enc =1μm;σ grating is the inherent noise of the grating scale, σ grating =1nm; γ is the temperature sensitivity coefficient; ΔT is the temperature deviation, ΔT=T-T0; S / N is the real-time estimated value of the grating scale signal-to-noise ratio; κ is the signal-to-noise ratio suppression coefficient, κ=10.1)

[0093] K k is the multi-scale fusion gain, is the dynamic weight matrix, W k =diag(w X ,w Y ,w Z ),in, is the axial displacement gradient, w X 、w Y and w Z are the plane coordinate weights of each axis, w X =w Y =1-w Z ;P k|k-1 is the uncertainty of the state estimation based on historical data before time k, Among them, A k is the state transfer matrix, which is used to describe the system dynamic model, P k-1is the posterior estimated covariance matrix of the previous moment, reflecting the uncertainty of the historical state; Q k is the process noise covariance matrix, which represents the impact of external interference on the system, H k To convert the state vector x k Mapping to observation space z k The matrix of .

[0094] Specifically, in the state equation of the Kalman filter algorithm, the compensation coefficient matrix is determined based on the experimental calibration of the thermal expansion coefficient of the grating scale material. In a constant temperature laboratory, the grating scale is placed in a temperature gradient environment, the axial expansion at different temperatures is measured, and the first-order temperature coefficient α and the second-order temperature coefficient β are obtained by least square fitting. The calibration reference temperature T0 is set to 25 degrees Celsius as the reference reference for the temperature drift function. The process noise covariance matrix Q k The calculation is generated by the slide rail repeatability accuracy of ±0.005 mm, including the statistical characteristics of mechanical vibration and slide rail wear. The observation noise covariance matrix R k In the linear encoder, the inherent noise σ enc The value is 1 micron, the inherent noise of the grating scale σ grating The value is 1 nanometer, the temperature sensitivity coefficient γ is set to 0.05 per degree Celsius, and the signal-to-noise ratio suppression coefficient κ is calibrated to 0.1 through the grating scale signal quality test.

[0095] In the actual detection process, the plane coordinate X in the state vector enc 、Y enc and axial displacement Z grating The velocity component is obtained by real-time acquisition by sensors through displacement difference calculation at adjacent moments. k The design is a 3-row 6-column matrix, the first three columns correspond to the observation mapping of plane coordinates and axial displacement, and the last three columns of velocity components are set to zero. k The axial weight w Z Determined by the grating scale resolution and the linear encoder noise ratio, the calculation formula is the reciprocal of the square of the grating scale noise divided by the sum of the reciprocals of the squares of the two noises, multiplied by the reciprocal of the axial displacement gradient plus 1. When the axial displacement gradient exceeds the preset threshold, w is automatically Z When calculating the Kalman gain, the state estimation covariance P is updated every 10 milliseconds. k|k-1 and multiplied with the dynamic weight matrix to ensure that the fusion result takes into account both planar robustness and axial accuracy.

[0096] This implementation method achieves efficient fusion of multi-source sensor data through experimental calibration and dynamic weight adjustment. The temperature compensation model accurately quantifies the thermal expansion effect of the grating scale and suppresses the temperature drift error to the nanometer level. The collaborative design of the dynamic noise covariance matrix and the weight matrix enables the system to maintain a three-dimensional coordinate fusion accuracy better than 0.01 mm under the temperature fluctuations and mechanical vibration environment of the workshop. The axial gradient constraint mechanism effectively avoids the interference of sudden displacement of the slide rail on the fusion result, and improves the reliability of anomaly detection. The structured design of the observation matrix ensures the consistency of the state vector and the physical space, providing a high-reliability data foundation for subsequent health assessments.

[0097] Preferably, feature extraction is performed based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and a multidimensional feature matrix is constructed based on the multidimensional gap features, including:

[0098] The first pre-processed data is determined as the gap width, and the width change rate of adjacent measurement points is determined according to the gap width to obtain a width mutation gradient; the width mutation gradient G i The calculation formula is: Among them, W i is the i-th measurement point (X i ,Y i ) corresponds to the gap width, X i and Y i are the horizontal and vertical coordinate values of the i-th measurement point respectively;

[0099] Determine the symmetry axis equation based on the vehicle design model or point cloud fitting;

[0100] Based on the symmetry axis equation, the left measurement point (X L ,Y L ) is mapped to the right symmetric coordinate (X R ,Y R );

[0101] According to the left measurement point (X L ,Y L ) and the right symmetrical coordinate (X R ,Y R ) Determine the left width W L and right width W R ;

[0102] According to the left width W L and the right side width W R Calculate the symmetry deviation S i ;S i The calculation formula is: λ' is the penalty for coordinate asymmetry, λ' = 0.1;

[0103] Fitting a cubic spline curve C(t) along the gap trajectory based on the three-dimensional coordinates of the second preprocessed data;

[0104] For each parameterized variable t of the cubic spline interpolation curve i , calculate the curvature K i ;in,

[0105] Calculate the absolute value of the deviation between the curvature and the mean to obtain the curvature anomaly value; the curvature anomaly value K dev The calculation formula is: K dev =|K i -μ K |; where μ K is the global mean curvature;

[0106] Determining the width mutation gradient, the curvature anomaly value, and the symmetry deviation as the multidimensional gap features;

[0107] The multidimensional gap features are subjected to feature normalization and matrix construction to obtain the multidimensional feature matrix.

[0108] Specifically, this embodiment first determines the gap width of each measuring point based on the preprocessed displacement data, and calculates the width change rate through the coordinate difference of adjacent points, and defines the change rate as the width mutation gradient. The horizontal and vertical coordinate values of adjacent points are directly obtained from the plane positioning data of the linear encoder to ensure the spatial accuracy of the calculation. The symmetry axis equation is determined by importing the center axis parameters in the vehicle CAD design model. If the model is not available, the least squares method is used to fit the laser scanning point cloud data to generate an approximate symmetry axis. The left measurement point is mirrored according to the symmetry axis equation to obtain the theoretical right coordinate, and the corresponding gap width is obtained by matching the nearest neighbor data of the actual right measurement point. The symmetry deviation is obtained by calculating the weighted sum of the left and right width differences and the coordinate offset, where the coordinate offset penalty coefficient is calibrated to 0.1 through experiments to balance width sensitivity and position tolerance.

[0109] Based on the fused 3D coordinate data, a cubic spline curve is fitted along the gap trajectory. The curve is parameterized using the chord-length method to ensure consistency in physical space. The local curvature is calculated for each interpolation point on the curve, and the curvature value is analytically solved using the first- and second-order derivatives of the curve. The global mean curvature is statistically derived from the curvature data of all measurement points, and the curvature outlier is defined as the absolute deviation of the curvature at each point from the mean. The final extracted width mutation gradient, symmetry deviation, and curvature outlier are scaled to the range of 0 to 1 using the minimum-maximum normalization method to eliminate dimensional differences. The normalized feature data are arranged in the order of the measurement points and combined with the 3D coordinates to construct a multidimensional feature matrix. Each row in the matrix corresponds to the feature vector of a measurement point, which is used for subsequent Mahalanobis distance analysis and health assessment.

[0110] This implementation method comprehensively quantifies the morphological anomalies of vehicle gaps through multi-dimensional extraction of geometric and statistical features. The experimental calibration of the penalty coefficient in the calculation of symmetry deviation effectively distinguishes assembly errors from measurement noise. The chord-length parameterized cubic spline fitting ensures the physical authenticity of the curvature calculation and avoids the spatial distortion caused by uniform parameterization. Normalization processing and matrix structured design make different features comparable in subsequent algorithms, significantly improving the robustness of anomaly detection. This solution converts abstract gap data into a resolvable numerical matrix, providing high-information-density input for the machine learning model, supporting an accident vehicle identification accuracy rate of 98.7%.

[0111] Preferably, calculating the outliers in the multidimensional feature matrix based on the Mahalanobis distance algorithm and determining the abnormal area by a dynamic threshold include:

[0112] Performing Z-score normalization on each feature column in the multidimensional feature matrix to obtain a normalized matrix;

[0113] Calculating a covariance matrix ∑ in the standardized matrix;

[0114] For each sample x in the normalized matrix i , calculate the Mahalanobis distance; the Mahalanobis distance D Mahalanobis (x i ) is calculated as: Among them, μ is the mean vector, N is the total number of samples in the multidimensional feature matrix;

[0115] The mean μ based on the Mahalanobis distance D and standard deviation σ D Set the dynamic threshold Threshold; where Threshold = μ D +3σ D ,

[0116] If DMahalanobis (x i )>Threshold, the sample is marked as an outlier; the label of the outlier is:

[0117] Bind the outlier labels to the three-dimensional coordinates to generate a set of abnormal regions; the set of abnormal regions for: X i 、Y i , Z i They are respectively the X-axis coordinate, Y-axis ordinate and Z-axis coordinate of the three-dimensional coordinate.

[0118] Preferably, the distribution density is the number of outliers per unit area; and the severity is the mean of the Mahalanobis distances of all outliers.

[0119] Preferably, the comparison result is obtained by comparing the production date data with the registration date of the target vehicle, including:

[0120] Comparing the obtained production date with the preset registration date to calculate the useful life of the component;

[0121] Calculating the deviation between the service life of the component and the overall service life of the target vehicle to obtain the comparison result CR;

[0122] The vehicle comprehensive health index is calculated based on the distribution density of the outliers, the severity of the abnormal area and the comparison result CR; the calculation formula of the vehicle comprehensive health index CHI is: CHI=∑(w i ·f i )+α”·distribution density of outliers·severity+β”·CR; where w i is the weight of the multidimensional gap feature, where the weight of the width mutation gradient is 0.35, the weight of the symmetry deviation is 0.28, and the weight of the curvature anomaly is 0.22), f i is the eigenvalue after feature normalization of the multidimensional gap feature, α″ is the outlier density weight, α″=0.15, β is the component age deviation weight, β=0.1.

[0123] Specifically, this embodiment first performs Z-score normalization on each feature column in the multidimensional feature matrix to eliminate the dimensional differences of different features. During the normalization process, the mean and standard deviation of each feature column are calculated in real time based on the current detection data set to ensure the centralization of the data distribution. The covariance matrix is obtained by multiplying the transpose of the normalization matrix and dividing it by the number of samples minus one, reflecting the correlation between features. When calculating the Mahalanobis distance, the mean vector is composed of the feature means of all samples, and the inverse matrix of the covariance matrix is efficiently solved using the Cholesky decomposition method. The dynamic threshold is set to 3 times the standard deviation of the Mahalanobis distance mean, and the standard deviation is obtained by full sample statistics before outlier detection. When the sample Mahalanobis distance exceeds the threshold, the system automatically marks it as an outlier and binds the three-dimensional coordinates corresponding to the point to the abnormal area set.

[0124] The calculation of the outlier distribution density is based on the vehicle surface grid division, with each grid cell area of 10 square centimeters. The density value is the number of outliers in the cell divided by the cell area. The severity is quantified by the arithmetic mean of the Mahalanobis distance of all outliers, reflecting the overall deviation intensity of the abnormal area. The three-dimensional coordinate data comes from the positioning results after fusion in the preprocessing stage, including the X-axis horizontal coordinate, the Y-axis longitudinal coordinate, and the Z-axis vertical coordinate. After the abnormal area set is generated, the system maps the discrete points into a continuous probability distribution through the kernel density estimation algorithm, and combines the gradient color rendering technology to generate a three-dimensional heat map. The red area in the heat map corresponds to high-density and high-severity anomalies, yellow is medium anomalies, and green is normal areas. Multi-perspective interactive analysis is supported.

[0125] This implementation ensures the accurate modeling of the correlation between multi-dimensional features by the Mahalanobis distance through the dynamic calculation of the normalization and covariance matrix, and solves the misjudgment problem of the traditional Euclidean distance under non-independent features. The dynamic threshold is adaptively adjusted based on the statistical 3σ principle, taking into account both detection sensitivity and false alarm rate control. Quantitative indicators of distribution density and severity transform abstract abnormal data into actionable maintenance priority scores, significantly improving the interpretability of detection results. The three-dimensional heat map visualization mapping technology, combined with the spatial reference of the vehicle structure model, enables technicians to quickly locate high-risk areas and shorten the decision-making cycle. In an experimental environment, this solution achieved an accident vehicle identification accuracy rate of 98.7% and a false detection rate of less than 1.5%, with industrial-grade reliability.

[0126] Corresponding to the above method, such as Figure 3 As shown, this embodiment also provides a vehicle intelligent detection system based on vehicle body gaps and production dates of non-wearing parts, including:

[0127] The data acquisition unit is used to obtain the production date data of the non-wearing parts of the target vehicle and drive the laser array sensor to move along the surface of the target vehicle through a three-axis linkage slide to synchronously collect the displacement data of the vehicle gap, the positioning data of the linear encoder, and the nanometer-level environmental parameters of the grating ruler;

[0128] a data preprocessing unit, configured to preprocess the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data;

[0129] a feature matrix construction unit, performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features;

[0130] An abnormality determination unit, configured to calculate outliers in the multidimensional feature matrix based on a Mahalanobis distance algorithm and determine abnormal areas using a dynamic threshold;

[0131] The detection index calculation unit is used to compare the production date data with the registration date of the target vehicle to obtain a comparison result, and determine the vehicle comprehensive health index based on the comparison result, the distribution density and severity of the outliers.

[0132] The beneficial effects of the present invention are as follows:

[0133] (1) The present invention uses a three-axis linkage slide to drive a laser array sensor, enabling precise acquisition of displacement data and environmental parameters of vehicle body gaps. It also utilizes a variety of data preprocessing techniques (such as wavelet noise reduction, Savitzky-Golay filtering, and Kalman filtering) to effectively remove noise, improving data accuracy and reliability. This comprehensive data processing capability ensures high-precision vehicle detection and adapts to diverse application scenarios.

[0134] (2) The present invention constructs units through a feature matrix. This method can extract multidimensional gap features, calculate outliers using the Mahalanobis distance algorithm, and dynamically determine abnormal areas. This process provides data support for further analysis of the health status of the vehicle body structure and a scientific basis for subsequent dynamic threshold setting, thereby improving the effectiveness of detection.

[0135] (3) The present invention accurately calculates the vehicle's comprehensive health index by comparing the production date of non-wearing parts with the vehicle registration date and combining the distribution density and severity of outliers. This method ensures that the vehicle's historical usage status is effectively reflected, making the test results practical and providing users with a more intuitive assessment of vehicle health.

[0136] (4) The vehicle intelligent detection system of the present invention integrates multiple modules, including data acquisition, preprocessing, feature extraction, anomaly determination, and health index calculation, forming a complete detection process. Through an optimized combination of hardware and software, the system enhances the intelligent level of vehicle detection, making detection work in the fields of automobile manufacturing and maintenance more efficient and convenient.

[0137] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0138] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts, characterized in that: include: Obtain the production date data of the target vehicle's non-wearing parts, and use a three-axis linkage slide to drive the laser array sensor to move along the target vehicle's surface to simultaneously collect vehicle gap displacement data, linear encoder positioning data, and nanometer-level environmental parameters of the grating scale; Preprocessing the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data; Performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features; Calculate outliers in the multidimensional feature matrix based on the Mahalanobis distance algorithm, and determine abnormal areas using a dynamic threshold; The production date data is compared with the registration date of the target vehicle to obtain a comparison result, and a comprehensive vehicle health index is determined based on the comparison result, the distribution density and severity of the outliers.

2. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 1 is characterized in that: Also includes: The determination result of the abnormal area is associated with the corresponding three-dimensional coordinates according to the second preprocessing data, and a three-dimensional gap heat map of the vehicle surface is generated based on the coordinates of the associated abnormal area and the severity of the outlier.

3. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 1 is characterized in that: Also includes: Based on a linear regression model, the temperature drift error of the laser array sensor is corrected according to the nanoscale environmental parameters.

4. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 1 is characterized in that: Preprocessing the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data includes: Considering the light and shadow effect of the gap between vehicles, the wavelet noise reduction algorithm is used to remove the high-frequency noise of the displacement data to obtain denoised data. The calculation formula of the denoised data is: in, is the denoised data; W j,k is the kth detail coefficient of the original displacement data under the jth layer wavelet decomposition; ψ j,k (x, y) is the wavelet basis function, corresponding to the decomposition scale j and position k; V J (x, y) is the approximate coefficient under the maximum decomposition scale J, I(x, y) is the local light intensity parameter, which is normalized to the range of 0, 1 by the light sensor or image grayscale value. is the gradient amplitude of the vehicle surface reflectivity, which reflects the degree of light and shadow mutation; α is the light sensitivity weight coefficient, α∈[0.2,0.5]; β is the reflectivity gradient suppression coefficient, β∈[1.0,2.0]; Thresh(·) is the threshold function, Among them, λ is the dynamic threshold, and the calculation formula of λ is: σ is the estimated value of the noise standard deviation, N is the signal length of the displacement data, and γ is the preset illumination dependency coefficient; Smoothing the denoised data using a Savitzky-Golay filter to obtain smoothed data; the Savitzky-Golay filter has a window length of 21 and a polynomial order of 3; The positioning data of the linear encoder and the axial displacement data of the grating ruler are fused through the Kalman filter algorithm to generate three-dimensional space coordinate data to obtain second pre-processed data.

5. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 4 is characterized in that: The expressions of the state equation and observation equation of the Kalman filter algorithm are: Among them, x k is the state vector, Among them, X enc 、Y enc are the plane coordinates measured by the linear encoder; Z grating is the axial displacement measured by the grating ruler; are the velocity components of each axis respectively; C k is the compensation coefficient matrix, which is related to the thermal expansion coefficient of the grating scale material; f(T) is the temperature drift function, f(T)=α'(T-T0)+β'(T-T0) 2 ; Wherein, T is the real-time ambient temperature; T0 is the calibration reference temperature; α' and β' are the first-order and second-order temperature coefficients, respectively, which are determined by calibration experiments; R k is the dynamic weight observation noise covariance, Among them, σ enc is the inherent noise of the linear encoder, σ enc =1μm;σ grating is the inherent noise of the grating scale, σ grating =1nm; γ is the temperature sensitivity coefficient; ΔT is the temperature deviation, ΔT=T-T0; S / N is the real-time estimated value of the grating scale signal-to-noise ratio; κ is the signal-to-noise ratio suppression coefficient, κ=10.1) K k is the multi-scale fusion gain, is the dynamic weight matrix, W k =diag(w X ,w Y ,w Z ),in, is the axial displacement gradient, w X 、w Y and w Z are the plane coordinate weights of each axis, w X =w Y =1-w Z ;P k|k-1 is the uncertainty of the state estimation based on historical data before time k, Among them, A k is the state transfer matrix, which is used to describe the system dynamic model, P k-1 is the posterior estimated covariance matrix of the previous moment, reflecting the uncertainty of the historical state; Q k is the process noise covariance matrix, which represents the impact of external interference on the system, H k To convert the state vector x k Mapping to observation space z k The matrix of .

6. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 1 is characterized in that: Performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features, including: The first pre-processed data is determined as the gap width, and the width change rate of adjacent measurement points is determined according to the gap width to obtain a width mutation gradient; the width mutation gradient G i The calculation formula is: Among them, W i is the i-th measurement point (X i ,Y i ) corresponds to the gap width, X i and Y i are the horizontal and vertical coordinate values of the i-th measurement point respectively; Determine the symmetry axis equation based on the vehicle design model or point cloud fitting; Based on the symmetry axis equation, the left measurement point (X L ,Y L ) is mapped to the right symmetric coordinate (X R ,Y R ); According to the left measurement point (X L ,Y L ) and the right symmetrical coordinate (X R ,Y R ) Determine the left width W L and right width W R ; According to the left width W L and the right side width W R Calculate the symmetry deviation S i ;S i The calculation formula is: λ' is the penalty for coordinate asymmetry, λ' = 0.1; Fitting a cubic spline curve C(t) along the gap trajectory based on the three-dimensional coordinates of the second preprocessed data; For each parameterized variable t of the cubic spline interpolation curve i , calculate the curvature K i ;in, Calculate the absolute value of the deviation between the curvature and the mean to obtain the curvature anomaly value; the curvature anomaly value K dev The calculation formula is: K dev =|K i -μ K |; where μ K is the global mean curvature; Determining the width mutation gradient, the curvature anomaly value, and the symmetry deviation as the multidimensional gap features; The multidimensional gap features are subjected to feature normalization and matrix construction to obtain the multidimensional feature matrix.

7. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 6 is characterized in that: Calculating outliers in the multidimensional feature matrix based on the Mahalanobis distance algorithm and determining abnormal areas using a dynamic threshold include: Performing Z-score normalization on each feature column in the multidimensional feature matrix to obtain a normalized matrix; Calculating the covariance matrix Σ in the standardized matrix; For each sample x in the normalized matrix i , calculate the Mahalanobis distance; the Mahalanobis distance D Mahalanobis (x i ) is calculated as: Among them, μ is the mean vector, N is the total number of samples in the multidimensional feature matrix; The mean μ based on the Mahalanobis distance D and standard deviation σ D Set the dynamic threshold Threshold; where Threshold = μ D +3σ D , If D Mahalanobis (x i )>Threshold, the sample is marked as an outlier; the label of the outlier is: Bind the outlier labels to the three-dimensional coordinates to generate a set of abnormal regions; the set of abnormal regions for: ;X i 、Y i , Z i They are respectively the X-axis coordinate, Y-axis ordinate and Z-axis coordinate of the three-dimensional coordinate.

8. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 7 is characterized in that: The distribution density is the number of outliers per unit area; the severity is the mean of the Mahalanobis distances of all outliers.

9. The vehicle intelligent detection method based on vehicle body gaps and production dates of non-wearing parts according to claim 8 is characterized in that: Comparing the production date data with the registration date of the target vehicle to obtain a comparison result, and determining a comprehensive vehicle health index based on the comparison result, the distribution density and severity of the outliers, including: Comparing the obtained production date with the preset registration date to calculate the useful life of the component; Calculating the deviation between the service life of the component and the overall service life of the target vehicle to obtain the comparison result CR; The vehicle comprehensive health index is calculated based on the distribution density of the outliers, the severity of the abnormal area and the comparison result CR; the calculation formula of the vehicle comprehensive health index is: CHI=∑(w i ·f i )+α”·distribution density of outliers·severity+β”·CR; where w i is the weight of the multidimensional gap feature, where the weight of the width mutation gradient is 0.35, the weight of the symmetry deviation is 0.28, and the weight of the curvature anomaly is 0.22), f i is the eigenvalue after feature normalization of the multidimensional gap feature, α″ is the outlier density weight, α″=0.15, β is the component age deviation weight, β=0.

1.

10. A vehicle intelligent detection system based on vehicle body gaps and production dates of non-wearing parts, characterized in that: include: The data acquisition unit is used to obtain the production date data of the non-wearing parts of the target vehicle and drive the laser array sensor to move along the surface of the target vehicle through a three-axis linkage slide to synchronously collect the displacement data of the vehicle gap, the positioning data of the linear encoder, and the nanometer-level environmental parameters of the grating ruler; a data preprocessing unit, configured to preprocess the displacement data and the positioning data respectively to obtain corresponding first preprocessed data and second preprocessed data; a feature matrix construction unit, performing feature extraction based on the first preprocessed data and the second preprocessed data to obtain multidimensional gap features, and constructing a multidimensional feature matrix based on the multidimensional gap features; An abnormality determination unit, configured to calculate outliers in the multidimensional feature matrix based on a Mahalanobis distance algorithm and determine abnormal areas using a dynamic threshold; The detection index calculation unit is used to compare the production date data with the registration date of the target vehicle to obtain a comparison result, and determine the vehicle comprehensive health index based on the comparison result, the distribution density and severity of the outliers.

Images