Three-dimensional assembly tolerance prediction method for automobile assembly
By constructing a joint Gaussian process regression model and dynamic correction coefficients, the simulation lag problem of dynamic production processes in existing technologies has been solved, enabling accurate tolerance prediction and risk warning in the automobile final assembly process, and reducing production costs and assembly failure risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-09
- Publication Date
- 2026-04-07
AI Technical Summary
Existing technologies cannot accurately reflect the dynamic production process in automobile assembly manufacturing, resulting in delayed predictions and an inability to handle non-uniform noise, leading to wasted process costs or assembly failure risks.
By constructing a joint Gaussian process regression model, using the residual vector between measured and ideal coordinates, combined with the spatiotemporal coupling covariance function and dynamic correction coefficient, the simulation model is updated to predict deviations and heteroscedasticity, and a comprehensive risk assessment index is calculated to achieve tolerance prediction and risk warning for the automobile assembly process.
It accurately captures systematic deviations and random noises that drift over time on the production line and fluctuate operating conditions, improves the overlap between the confidence interval of simulation predictions and actual data, reduces rework rates and manufacturing costs, and enhances the system's noise robustness and risk sensitivity.
Smart Images
Figure CN121809289A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent manufacturing and data processing for automobiles. More specifically, the present invention relates to a three-dimensional assembly tolerance prediction method for automobile general assembly. Background Art
[0002] In the field of automobile general assembly manufacturing, dimensional engineering is a key link in determining the aesthetics of the vehicle body appearance (such as door gap and surface flatness) and functional sealing. Currently, the industrial community usually uses computer-aided tolerance design (CAT) software to estimate in the product design stage. Such technologies are generally based on the principles of rigid body kinematics and Monte Carlo simulation methods, setting part size tolerances and fixture positioning errors as fixed standard normal distributions, and calculating the size deviation distribution of the final assembly to evaluate the feasibility of the design scheme.
[0003] However, the production environment in the actual mass production stage is highly dynamic and complex, and the existing static simulation technologies have obvious limitations. First, it is difficult for the existing technologies to characterize systematic time-varying drifts. During the production process, affected by factors such as the day-night alternation of environmental temperature, gradual wear of welding fixtures, and thermal deformation of stamping dies, the mean value of part size deviations will drift non-linearly over time, and the simulation model with fixed mean parameters cannot follow this change in time, resulting in serious lag in prediction results. Second, the existing technologies often ignore the heteroscedastic characteristics of the error distribution, that is, most existing models assume that the error fluctuation amplitude (variance) is constant, but actual measurements show that the error dispersion degrees of assembly feature points with different stiffnesses (such as positioning holes and free edges) or different batches of raw materials are significantly different. This input-dependent noise variance makes the existing technologies overestimate the tolerance range during the production stable period, causing waste of process costs; while underestimating the tolerance range during the production fluctuation period, increasing the risk of assembly failure.
[0004] Therefore, how to adaptively and dynamically correct the simulation model by integrating online measurement data has become a technical problem to be solved urgently. [[ID=IS]] Summary of the Invention
[0005] The purpose of the present invention is to propose a three-dimensional assembly tolerance prediction method for automobile general assembly, aiming to solve the problem that the existing static simulation model cannot truly reflect the dynamic production process, resulting in prediction lag and inability to handle non-uniform noise; for this purpose, the present invention provides a solution in one of the following aspects.
[0006] A three-dimensional assembly tolerance prediction method for automobile general assembly provided by the present invention includes: The ideal and measured three-dimensional coordinates of each key measuring point under standard working conditions on the automobile assembly line are obtained, and an input feature vector is constructed based on the ideal three-dimensional coordinates. The residual vector between the measured coordinates and the mean of the ideal three-dimensional coordinates is used as the prediction target vector. The ideal three-dimensional coordinates are the three-dimensional coordinates of the key measuring points under ideal tolerance settings derived from the simulation model. The input feature vector and the predicted target vector are combined to form a historical dataset to train the constructed joint Gaussian process regression model. The trained joint Gaussian process regression model is used to predict the deviation correction and heteroscedasticity under new operating conditions, and the dynamic correction coefficient is calculated in combination with the set maximum allowable fluctuation variance. The input parameters in the simulation model are then updated using the dynamic correction coefficient. The updated input parameters are substituted into the simulation model to regenerate a new tolerance distribution map for key measurement points. The comprehensive risk assessment index is then calculated based on the new tolerance distribution map. Based on the assessment results, tolerance prediction and risk warning for the automobile assembly process are achieved.
[0007] The above scheme constructs a feature space and uses measured coordinates to calculate residuals, which can correlate real fluctuations with ideal states, providing a data foundation for subsequent accurate corrections.
[0008] Preferably, the input feature vector includes the ideal three-dimensional coordinates of each key measuring point in the vehicle body coordinate system, the production batch serial number, the current workstation temperature, and the tooling fixture ID.
[0009] Preferably, the kernel function of the joint Gaussian process regression model Satisfying the expression: ; In the formula, Indicates the vertical scale amplitude. Indicates key measurement points and key measurement points Euclidean distance in the vehicle body geometry Indicates key measurement points and key measurement points Batch difference in the production sequence, Represents geometric feature scale, Represents the scale of sequence features. Let i represent the shape parameter of the rational quadratic kernel, and i and j be the serial numbers of the key measurement points, respectively.
[0010] The aforementioned spatiotemporal coupling covariance function uses a Gaussian kernel to capture the positive correlation between adjacent measurement points in space due to similar forces, and uses a rational quadratic kernel to simulate the multi-scale error fluctuation characteristics in the production process, thereby accurately characterizing the complex spatiotemporal error structure in the automobile assembly process.
[0011] Preferably, the posterior distribution of the joint Gaussian process regression model is solved using variational inference methods, and the predicted bias correction and heteroscedasticity are output; and the dynamic correction coefficient is obtained based on the predicted bias correction and heteroscedasticity. The dynamic correction coefficient for: ; In the formula, For any input parameter, the nominal value specified in the design specification is given. This indicates the adjustment of the learning rate. This represents the maximum allowable variance, where x is the new operating condition characteristic. This is the deviation correction amount. For heteroscedasticity, any of the input parameters is the locating pin, the locating surface tolerance of the tooling fixture, or the surface profile of the mating surface of the part.
[0012] The above scheme introduces a mechanism that includes uncertainty penalty, that is, the square root of the variance ratio is introduced into the denominator as a damping factor. When the heteroscedasticity predicted by the model is high, the denominator is significantly increased, automatically reducing the correction magnitude, avoiding model oscillation caused by abnormal data or data sparsity, and ensuring the robustness and safety of the system under complex working conditions.
[0013] Preferably, the comprehensive risk assessment index satisfies the expression: ; In the formula, As a comprehensive risk assessment indicator under new operating conditions, This indicates the total number of key measuring points. Indicates the first The weighting coefficients of each key measurement point The cumulative distribution function represents the standard normal distribution. This indicates the upper limit of the tolerance required by the design specifications. This represents the mean of the theoretical tolerance distribution plot output by the simulation model. This represents the variance of the theoretical tolerance distribution plot output by the simulation model. Indicates the first Deviation correction amount for each key measuring point Indicates the first Heteroscedasticity of key measurement points.
[0014] The comprehensive risk assessment index constructed above uniformly assesses the erosion of safety margin by mean drift and the superposition of volatility by environmental noise. It can keenly detect the nonlinear decline in the pass rate, thereby achieving early warning before defects occur.
[0015] Preferably, the risk warning includes: when the comprehensive risk assessment index exceeds a preset threshold, triggering an alarm signal and outputting the location of the corresponding key measuring point and the corresponding prediction deviation value.
[0016] The above scheme can accurately detect anomalies at key measuring points.
[0017] Preferably, the training of the pre-constructed joint Gaussian process regression model specifically includes: using variational inference methods to approximate the true posterior distribution of the heteroscedastic Gaussian process by minimizing the KL divergence, and determining the model hyperparameters.
[0018] Preferably, the simulation model is a three-dimensional tolerance simulation software.
[0019] Preferably, the measured coordinates are the measured three-dimensional coordinate values of each key measuring point collected in real time by a lidar or blue light scanner installed on the assembly line.
[0020] Preferably, it also includes: performing time alignment processing on the ideal three-dimensional coordinates and the measured coordinates.
[0021] The beneficial effects of this invention are as follows: This invention establishes a residual mapping relationship between measured coordinates and simulation models through a physical-guided heteroscedastic Gaussian process regression model. It can not only accurately capture the systematic deviations that drift over time on the production line, but also simultaneously predict random noise that fluctuates drastically with the operating conditions, thus greatly improving the overlap between the confidence interval of the simulation prediction and the actual measurement data.
[0022] Furthermore, through correction mechanisms and comprehensive risk assessment indicators, this invention endows the system with extremely strong noise resistance and risk sensitivity. It can not only automatically identify and cool down high-uncertainty data to prevent erroneous corrections, but also amplify potential risk signals. Before assembly defects actually occur, it can identify the tendency of tolerances to run out of control due to equipment aging or environmental changes, effectively reducing rework rates and manufacturing costs. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating a three-dimensional assembly tolerance prediction method for automobile assembly according to the present invention. Figure 2 This is a schematic diagram illustrating the comparison of the effects of the present invention and the prior art. Detailed Implementation
[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0025] like Figure 1As shown in this embodiment, a three-dimensional assembly tolerance prediction method for automobile final assembly includes the following steps: S1. Construct historical datasets.
[0026] In this embodiment, the historical dataset includes the input feature vectors of key measurement points on the vehicle assembly line and the prediction target vector.
[0027] The input feature vector includes not only the geometric spatial features of the key measuring points, but also the working condition features representing the production status. For example, the input feature vector of a key measuring point for a door mounting hole can be represented as follows: ,in, The ideal three-dimensional coordinates are in the vehicle body coordinate system, batch_id is the production batch serial number, temp is the current workstation temperature, and fixture_id is the tooling fixture number.
[0028] The aforementioned ideal three-dimensional coordinates were obtained by acquiring simulation data from the automobile assembly line; the simulation data is the theoretical distribution data of key measuring points under ideal tolerance settings derived from the simulation model, including ideal three-dimensional coordinates and theoretical tolerance distribution maps.
[0029] The simulation model in this embodiment can be a three-dimensional tolerance simulation software, such as 3DCS.
[0030] The process of obtaining the target vector for prediction is as follows: First, obtain the ideal three-dimensional coordinates and measured coordinates of all key measuring points.
[0031] Secondly, the residual vector between the measured coordinates and the mean of the ideal three-dimensional coordinates of each key measuring point is used as the prediction target vector. Specifically, the residual is the difference between any measured coordinate value and the corresponding mean of the ideal three-dimensional coordinates.
[0032] The measured coordinates are the actual three-dimensional coordinate data of the corresponding key measurement points collected by online vision measurement systems (such as lidar and blue light scanners) installed on the automobile assembly line.
[0033] In this embodiment, the simulation data and measured coordinates are preprocessed with timestamp synchronization and coordinate system alignment to ensure that the simulation data and measured coordinates are compared under the same reference.
[0034] For example, if the ideal mean coordinate of a key measuring point is 0.0 mm and the measured coordinate is 0.8 mm, then the residual is 0.8 mm.
[0035] In this embodiment, by aligning multi-source data and constructing a feature space, discrete measurements can be linked together, providing rich data support for subsequently capturing the spatiotemporal evolution of errors.
[0036] S2. Train a pre-built joint Gaussian process regression model using the historical dataset.
[0037] Specifically, to address the issue that traditional regression models cannot handle non-uniform noise, this embodiment constructs a Joint Gaussian Process Regression Model (PI-HGPR) to capture the spatial correlation between key measurement points and the temporal correlation of the production sequence using a spatiotemporal coupled covariance function. This Joint Gaussian Process Regression Model comprises two coupled sub-processes: a mean process and a noise process. The mean process is used to fit the systematic drift trend of dimensional deviations; the noise process is used to fit the logarithmic variance of the dimensional deviations, representing the processing stability under the current operating conditions.
[0038] To accurately capture the physical characteristics of spatially adjacent key measurement point error correlation and temporally adjacent batch error correlation in automobile assembly, this embodiment also designs a spatiotemporally coupled covariance function as the kernel function of the joint Gaussian process regression model, which satisfies the expression: ; in, Indicates key measurement points and key measurement points The kernel function values between Indicates the vertical scale range; Indicates key measurement points and key measurement points Euclidean distance in the vehicle body geometry Indicates key measurement points and key measurement points Time interval (batch difference) in the production sequence and These are geometric feature scales and sequence feature scales, respectively. Let i be the shape parameter of the rational quadratic kernel, and i and j be the serial numbers of the key measurement points, respectively.
[0039] Among them, in the formula The Gaussian kernel form is used because, in physical space, two assembly points that are closer together experience similar tooling constraints, resulting in a high positive correlation in their errors; in the formula... Using a rational quadratic kernel, it is able to simulate the temporal correlation of coexisting long and short memories.
[0040] For example, suppose the key measurement points Located in the upper left corner of the car door, batch number 100; key measuring point Located at the lower left corner of the car door, batch number 105, the geometric distance between the two measuring points. Batch difference Set parameters , , , Therefore, the spatial correlation terms to be calculated are: .
[0041] The time-related terms calculated are: .
[0042] Then the kernel function values of the two key measurement points .
[0043] The kernel function value mentioned above is essentially a comprehensive correlation strength, which indicates that the two key measurement points have a strong correlation in error performance. Therefore, the joint Gaussian process regression model mentioned above can use the information of one key measurement point to infer the state of the other measurement point.
[0044] Thus, by introducing a spatiotemporal coupled covariance function, the joint Gaussian process regression model can make full use of the physical constraints and temporal patterns in the automotive assembly process, significantly improving the accuracy of error prediction for unmeasured points or future batches.
[0045] After constructing the joint Gaussian process regression model, the model is trained using historical datasets. During training, the input feature vector is used as input, the predicted target vector is used as output, and variational inference is used to approximate the true posterior distribution of the heteroscedastic Gaussian process by minimizing the KL divergence, thereby determining the hyperparameters of the joint Gaussian process regression model.
[0046] Since the training process of the joint Gaussian process regression model is an existing technique, it will not be described in detail here.
[0047] It should be noted that since the posterior distribution of a heteroscedastic Gaussian process contains non-Gaussian terms, variational inference is used to solve it.
[0048] S3. Using the trained joint Gaussian process regression model, predict the systematic deviation correction amount and heteroscedasticity for the new operating condition characteristics, and calculate the dynamic correction coefficient in combination with the set maximum allowable fluctuation variance, and use the dynamic correction coefficient to update the input parameters in the simulation model.
[0049] In this embodiment, when new operating condition characteristics After inputting the trained model, the model outputs the prediction bias correction amount. and predicted heteroscedasticity .
[0050] The above input parameters include, but are not limited to, the locating pin, the locating surface tolerance of the tooling fixture, and the surface profile of the mating surfaces of the parts.
[0051] Based on these two predicted values, the dynamic correction coefficient is calculated. This is used to update input parameters in the simulation model, such as the position tolerance of the locating pin.
[0052] Among them, the dynamic correction coefficient for: ; in, This is the nominal value of the locating pin. To adjust the learning rate, To allow for the maximum variance in volatility, For the predicted heteroscedasticity, This is the deviation correction amount.
[0053] The above formula introduces an uncertainty penalty mechanism, with the denominator acting as an adaptive damper. When the model predicts heteroscedasticity... When the value is very large (meaning that current production is extremely unstable or data is sparse), the denominator increases significantly, and the correction amount is compressed, reflecting a safety strategy of making fewer corrections as uncertainty increases.
[0054] For example, suppose the nominal value of a certain locating pin is... Correct learning rate Maximum allowable variance .
[0055] Case A (Data is reliable): Correction amount for bias in model predictions heteroscedasticity ,at this time, Dynamic correction factor: .
[0056] The above data has been corrected to be close to the predicted values.
[0057] Case B (Data in Doubt): Correction amount for bias in model predictions heteroscedasticity (e.g., sensor jitter).
[0058] at this time, Dynamic correction factor: .
[0059] The above data has only undergone minor corrections, effectively suppressing noise interference.
[0060] In this embodiment, the dynamic correction mechanism ensures that the updated input parameters in the simulation model not only reflect the actual deviation trend, but also do not fluctuate violently due to individual abnormal data, thus guaranteeing the stable operation of the system.
[0061] S4. Substitute the updated input parameters into the simulation model, regenerate a new tolerance distribution map of the key measurement points, and calculate the comprehensive risk assessment index based on the new tolerance distribution map. Based on the comprehensive risk assessment index, achieve tolerance prediction and risk warning for the automobile assembly process.
[0062] Specifically, the corrected parameters Substitute the Monte Carlo simulation engine into the simulation model to regenerate a new tolerance distribution map for the key measurement points.
[0063] To quantify the risk of assembly failure, a comprehensive risk assessment index was constructed. Specifically: ; in, This indicates the total number of key measuring points. Indicates the first The weighting coefficients of each key measurement point The cumulative distribution function represents the standard normal distribution. This indicates the upper limit of the tolerance required by the design specifications. This represents the mean of the theoretical tolerance distribution plot output by the simulation model. This represents the variance of the theoretical tolerance distribution plot output by the simulation model. Indicates the first Deviation correction amount for each key measuring point Indicates the first Heteroscedasticity of key measurement points.
[0064] In the formula The calculation is based on the measured value being less than the upper tolerance limit. The probability, i.e., the pass rate. Taking the reciprocal of the pass rate makes the comprehensive risk assessment index inversely proportional to the pass rate; when the pass rate decreases, the risk value increases sharply and non-linearly.
[0065] In one embodiment, the weighting coefficient The weighting can be set based on human experience. For example, if the flatness of the door handle is of interest, the weight of the measurement point at that point is set to 0.8, and the weight of the others is 0.2.
[0066] In another embodiment, the weighting is set based on the degree of influence of each key measuring point on the final assembly quality, and the sum of the weighting coefficients of all measuring points is 1. Specifically, the degree of influence mentioned above can be obtained by sensitivity analysis or expert scoring, and the degree of influence is normalized to obtain the weighting coefficient of the corresponding key measuring point.
[0067] For example, suppose we only consider a single point ( Design tolerance upper limit Original simulation: Simulation mean ,variance .
[0068] Model prediction: bias correction (Mean shift), heteroscedasticity (Increased fluctuations).
[0069] The new overall mean is The new total standard deviation is Z-score is At this point, consult the standard normal distribution table. (That is, a pass rate of approximately 84.13%), then, the risk value... .
[0070] Compared to the ideal situation (pass rate 99.7%), Risk If the risk value increases significantly (1.003), the system will trigger an early warning.
[0071] Figure 2 This is a comparison chart of the effects of the present invention and existing technologies. The chart shows the distribution of key measurement point dimensional deviations (Y-axis) along a production batch timeline (X-axis). The chart includes the static 3σ range of the existing technology, which is a horizontal band of constant width. When the real-time measurement data exhibits a wavy drift with gradually increasing fluctuation amplitude (as shown by the circular data points in the chart), then from... Figure 2 As can be seen from the image, at the lower left arrow, the static model's prediction range is too wide, indicating overestimation. At the upper right arrow, a specific circular data point exceeds the static 3σ range of existing technologies, suggesting that the existing technology missed this anomaly. In contrast, the dynamic 3σ confidence interval and dynamic trend prediction curve in the figure closely follow the mean trend of the circular data points. Furthermore, the width of the dynamic 3σ confidence interval expands naturally and smoothly as data fluctuations increase, perfectly encompassing all data points, including the risk point that falls outside the static 3σ range.
[0072] The solution of this invention, through comprehensive risk assessment indicators and dynamic distribution generation, can accurately perceive the trend of increased fluctuations in the later stage of production and automatically relax the prediction boundary to include potential extreme fluctuations in the control scope, thereby avoiding false alarms and missed detections and achieving accurate capture of assembly failure risks.
[0073] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.
[0074] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.
Claims
1. A three-dimensional assembly tolerance prediction method for automobile final assembly, characterized in that, include: Obtain the ideal three-dimensional coordinates and measured coordinates of each key measuring point under standard working conditions on the automobile assembly line, and construct the input feature vector based on the ideal three-dimensional coordinates; The residual vector between the measured coordinates and the mean of the ideal three-dimensional coordinates is used as the prediction target vector; the ideal three-dimensional coordinates are the three-dimensional coordinates of key measuring points derived from the simulation model under the ideal tolerance setting. The input feature vector and the predicted target vector are combined to form a historical dataset to train the constructed joint Gaussian process regression model. The trained joint Gaussian process regression model is used to predict the deviation correction and heteroscedasticity under new operating conditions, and the dynamic correction coefficient is calculated in combination with the set maximum allowable fluctuation variance. The input parameters in the simulation model are then updated using the dynamic correction coefficient. The updated input parameters are substituted into the simulation model to regenerate a new tolerance distribution map for key measurement points. The comprehensive risk assessment index is then calculated based on the new tolerance distribution map. Based on the assessment results, tolerance prediction and risk warning for the automobile assembly process are achieved.
2. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The input feature vector includes the ideal three-dimensional coordinates of each key measuring point in the vehicle body coordinate system, the production batch serial number, the current workstation temperature, and the tooling fixture ID.
3. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The kernel function of the joint Gaussian process regression model for: ； In the formula, Indicates the vertical scale amplitude. Indicates key measurement points and key measurement points Euclidean distance in the vehicle body geometry Indicates key measurement points and key measurement points Batch difference in the production sequence, Represents geometric feature scale, Represents the scale of sequence features. Let i represent the shape parameter of the rational quadratic kernel, and i and j be the serial numbers of the key measurement points, respectively.
4. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The dynamic correction coefficient includes: The posterior distribution of the joint Gaussian process regression model is solved using variational inference methods, and the predicted bias correction and heteroscedasticity are output; and the dynamic correction coefficient is obtained based on the predicted bias correction and heteroscedasticity. The dynamic correction coefficient for: ； In the formula, For any input parameter, the nominal value specified in the design specification is given. This indicates the adjustment of the learning rate. This represents the maximum allowable variance, where x is the new operating condition characteristic. This is the deviation correction amount. For heteroscedasticity, any of the input parameters is the locating pin, the locating surface tolerance of the tooling fixture, or the surface profile of the mating surface of the part.
5. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The comprehensive risk assessment index satisfies the expression: ； In the formula, As a comprehensive risk assessment indicator under new operating conditions, This indicates the total number of key measuring points. Indicates the first The weighting coefficients of key measurement points The cumulative distribution function represents the standard normal distribution. This indicates the upper limit of the tolerance required by the design specifications. This represents the mean of the theoretical tolerance distribution plot output by the simulation model. This represents the variance of the theoretical tolerance distribution plot output by the simulation model. Indicates the first Deviation correction amount for each key measuring point Indicates the first Heteroscedasticity of predictions for key measurement points.
6. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The risk warning includes: when the comprehensive risk assessment index exceeds a preset threshold, triggering an alarm signal and outputting the location of the corresponding key measuring point and the corresponding prediction deviation value.
7. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The joint Gaussian process regression model constructed through training specifically includes: using variational inference methods to approximate the true posterior distribution of the heteroscedastic Gaussian process by minimizing the KL divergence, and determining the model hyperparameters.
8. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The simulation model is a three-dimensional tolerance simulation software.
9. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, The measured coordinates are the measured three-dimensional coordinate values of each key measuring point collected in real time by a lidar or blue light scanner installed on the assembly line.
10. The three-dimensional assembly tolerance prediction method for automobile final assembly according to claim 1, characterized in that, Also includes: Time alignment is performed between the ideal 3D coordinates and the measured coordinates.