Automobile part assembly precision intelligent compensation method and self-adaptive regulation and control system

Abstract

The invention discloses an automobile part assembly precision intelligent compensation method and a self-adaptive regulation and control system, and relates to the field of intelligent compensation, and the method comprises the steps: synchronously collecting part geometric parameters, tool poses and environment data, and constructing an associated data matrix; analyzing data by using an improved random forest-attention model, and outputting a deviation factor contribution degree sequence; a compensation calculation model is designed accordingly, initial compensation amounts are generated in a segmented mode in combination with a precision margin threshold value, and correction is conducted through historical data similarity matching; selecting an execution path according to the compensation amount and the core deviation type; an actual precision value is obtained through laser detection after assembly, and a compensation error is calculated; based on an error triggering model optimization mechanism, model parameters are iteratively updated by using a gradient descent algorithm, and the deviation identification and compensation precision is improved. The method has the advantages that the model attribution deviation is improved, the precise compensation amount is calculated in combination with historical data, flexible execution, real-time monitoring and model self-optimization are matched, and the assembly precision and the production efficiency are efficiently improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent compensation, and in particular to an intelligent compensation method and adaptive control system for the assembly precision of automotive parts. Background Technology

[0002] As the automotive industry rapidly advances towards electrification, intelligence, and lightweighting, the requirements for vehicle performance, quality, and safety are becoming increasingly stringent, making component assembly precision a core constraint. Traditional assembly methods heavily rely on manual experience and offline debugging. Static compensation strategies often fail when faced with unavoidable geometric errors, thermal deformation, and external disturbances in the assembly of complex systems such as the body, powertrain, and battery pack. This leads to excessive gaps, abnormal noises, sealing failures, and even frequent functional malfunctions. Especially in the field of new energy vehicles, scenarios such as high-precision stacking of battery modules, alignment of motor shafts, and multi-material body connections present unprecedented challenges to micron-level dynamic precision control.

[0003] Current automotive parts assembly precision compensation methods and adaptive control systems on the market often rely on single sensors or a few dimensional parameters for data acquisition. For example, they may only monitor part dimensions while ignoring key influencing factors such as environmental vibration and tooling pressure. Furthermore, data from different sources lacks temporal alignment and correlation processing, making analysis prone to bias due to incomplete data. In the deviation attribution stage, traditional methods often employ basic statistical models or ordinary machine learning algorithms, lacking both feature selection and optimization mechanisms and precise weight allocation capabilities. This makes it difficult to quantify the contribution of each factor, often resulting in ambiguous deviation source location. Compensation calculations are mostly based on fixed formulas, failing to incorporate dynamic adjustments using precision margin thresholds and ignoring the reference value of historical data. They cannot correct compensation parameters through similar cases, leading to insufficient accuracy and stability. At the execution and optimization level, compensation paths are singular and lack real-time feedback control. Some methods rely on manual adjustments, resulting in low efficiency and poor adaptability. Simultaneously, most models are static, lacking iterative update mechanisms based on compensation errors, making it difficult to cope with dynamic changes during assembly. In addition, most systems lack real-time interaction with MES and data blockchain traceability, resulting in weak collaboration and traceability. Summary of the Invention

[0004] To improve existing methods and systems, this paper provides an intelligent compensation method and adaptive control system for the assembly accuracy of automotive parts. This method collects multi-dimensional data through distributed sensing to improve model attribution bias, calculates accurate compensation amount by combining historical data, and combines flexible execution, real-time monitoring and model self-optimization. It also integrates system collaboration functions to efficiently improve assembly accuracy and production efficiency.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] Intelligent compensation methods for assembly precision of automotive parts include:

[0007] Based on a distributed sensor network, the geometric parameters of the target parts, the pose parameters of the assembly tooling, and the assembly environment parameters during the assembly process of automotive parts are collected synchronously and preprocessed to obtain an assembly data matrix containing correlation relationships.

[0008] A bias attribution model based on an improved random forest-attention mechanism is constructed. The obtained assembly data matrix is ​​input into the model to obtain the ranking of bias contributions.

[0009] Based on the ranking results of deviation contribution, a compensation amount calculation model is constructed. By introducing an assembly accuracy margin threshold, a piecewise function is used to calculate the initial compensation amount. Historical assembly data is matched with similar data in the historical assembly data through a cosine similarity algorithm to correct the initial compensation amount and obtain the final compensation amount.

[0010] Based on the final compensation amount and the type of core deviation factors, the corresponding compensation execution path is selected. During the compensation execution process, the adjustment amount is monitored and fed back in real time through displacement sensors.

[0011] After the compensation is completed, the assembled parts are inspected using a laser profilometer to obtain the actual assembly accuracy value. The actual assembly accuracy value is then compared with the preset assembly accuracy standard to calculate the compensation error.

[0012] Based on the calculated compensation error, a model optimization triggering mechanism is constructed. The model parameters are iteratively updated using the gradient descent algorithm. The latest assembly data, compensation parameters, and effect evaluation results are used as training samples to optimize the model's deviation identification accuracy and compensation calculation precision.

[0013] Preferably, the step of synchronously collecting and preprocessing the geometric parameters of the target parts, the pose parameters of the assembly tooling, and the assembly environment parameters during the automotive parts assembly process based on a distributed sensor network to obtain an assembly data matrix containing correlation relationships specifically includes:

[0014] The geometric parameters of the target component include the dimensional deviations and geometric tolerances of the key features of the component, which are collected by laser displacement sensors and industrial CT scanning equipment.

[0015] The assembly tooling posture parameters include the tooling positioning pin coordinates and clamping mechanism pressure values, which are collected by a six-dimensional force sensor and a visual positioning camera.

[0016] The assembly environment parameters include the temperature, humidity and vibration frequency of the assembly area, which are collected by temperature and humidity sensors and vibration sensors.

[0017] Data outside the normal range is filtered out, abnormal data is marked and replaced with smoothed data from adjacent time points, high-frequency interference data is removed for geometric parameters, and data smoothing is performed for tooling pose parameters.

[0018] Based on the start time of the assembly process, the timestamps of data from different sources are aligned, and heterogeneous data are mapped to the same assembly timeline coordinate system to form a data matrix that includes the relationship between parts, tooling, and environment.

[0019] Preferably, the construction of the bias attribution model based on the improved random forest-attention mechanism, which involves inputting the acquired assembly data matrix into the model to obtain the bias contribution ranking, specifically includes:

[0020] The acquired assembly data matrix is ​​transformed into a set of feature variables that can be identified by the attribution model. By improving the random forest algorithm and introducing an out-of-bag data error correction mechanism, the decision tree splitting nodes are optimized. The importance of the feature variables is evaluated, variables with minimal impact on assembly accuracy are removed, and key feature variables are retained.

[0021] The key characteristic variables include component size deviation, tooling positioning error, and ambient temperature fluctuation.

[0022] Based on the attention mechanism, the Pearson correlation coefficient between key feature variables and assembly accuracy deviation is calculated, and weights are assigned according to the absolute value of the correlation coefficient.

[0023] Integrate the feature selection results and weight allocation results, quantify the deviation contribution of each key feature variable, sort them from high to low according to their contribution, and output the deviation contribution ranking results.

[0024] Preferably, the step of constructing a compensation amount calculation model based on the deviation contribution ranking results, introducing an assembly accuracy margin threshold, calculating the initial compensation amount using a piecewise function, and correcting the initial compensation amount by matching similar data in historical assembly data using a cosine similarity algorithm to obtain the final compensation amount specifically includes:

[0025] Based on the obtained deviation contribution ranking, the core deviation factors and their corresponding influence coefficients are determined, and the core factors are used as the main input variables for calculating the compensation amount.

[0026] Based on the assembly of components, an assembly accuracy margin threshold is set, and the initial compensation amount is obtained by calculating the influence coefficient of the core deviation factors through a piecewise function.

[0027] The system calls upon compensation data for similar parts from the historical assembly database, determines the fit with the current scenario through similarity matching, and corrects the initial compensation amount based on similar case parameters.

[0028] The compensation amount is iteratively optimized using the assembly accuracy compliance rate as the reward function. The step size is adjusted by decreasing as the number of iterations increases until the compensation amount meets the accuracy expectation, and the final compensation amount is obtained.

[0029] Preferably, the step of selecting the corresponding compensation execution path based on the final compensation amount and the type of core deviation factors, and the real-time monitoring and feedback of the adjustment amount through displacement sensors during the compensation execution process, specifically includes:

[0030] Based on the final compensation amount and the type of deviation factor obtained, determine whether to adopt a passive compensation path or an active compensation path.

[0031] Passive compensation is based on the final compensation amount to calculate the required thickness of the compensation shim, and then the compensation shim of the required thickness is embedded into the assembly gap of the parts to fill the assembly gap caused by dimensional deviation, thus completing the passive compensation.

[0032] Active compensation is based on the final compensation amount and gradually corrects the position of the tooling positioning pin according to the adjustment range to achieve tooling posture correction.

[0033] Throughout the compensation process, the deviation between the actual adjustment results and the preset compensation amount is checked in real time to confirm that the compensation operation meets the standards.

[0034] Preferably, after the compensation is completed, the assembled parts are subjected to precision testing using a laser profilometer to obtain the actual assembly precision value. The actual assembly precision value is then compared with a preset assembly precision standard to calculate the compensation error. Specifically, this includes:

[0035] Based on the compensated assembled parts, a laser profilometer is used to perform accuracy detection and obtain the actual accuracy value.

[0036] The actual assembly accuracy value is compared with the preset assembly accuracy standard, and the deviation value between the two is used as the compensation error to quantify the compensation effect.

[0037] If the compensation error is ≤0.01mm, the compensation is deemed qualified; if the compensation error is >0.01mm, a secondary compensation warning is triggered, and the compensation parameters are recorded to the historical database.

[0038] Preferably, the step of constructing a model optimization triggering mechanism based on the calculated compensation error, iteratively updating the model parameters using the gradient descent algorithm, and using the latest assembly data, compensation parameters, and effect evaluation results as training samples to optimize the model's deviation identification accuracy and compensation calculation precision specifically includes:

[0039] Set model optimization trigger conditions, including the average compensation error of three consecutive assembly batches being greater than 0.008 mm and the occurrence rate of secondary compensation in a single assembly batch exceeding 10%, and monitor in real time whether the trigger conditions are met.

[0040] If the optimization conditions are triggered, the model to be optimized is locked, namely the deviation attribution model and the compensation amount calculation model. The original data, actual compensation parameters and compensation effect evaluation results of the latest assembly process are integrated to form a model training sample set.

[0041] The gradient descent algorithm is used to iteratively update the model parameters using data from the sample set as input, thereby reducing the model's prediction error.

[0042] Preferably, the servo motor used during compensation execution employs an absolute encoder to provide real-time feedback on the motor's rotational accuracy; the blockchain database for data storage and interaction adopts a consortium blockchain architecture, with each assembly data block containing a timestamp, equipment number, operator information, and digital signature to ensure the effectiveness of data traceability; the MES system interface supports the OPC UA protocol, enabling real-time bidirectional data interaction between the system and the MES system, and automatically sending a production pause request to the MES system when assembly accuracy is abnormal.

[0043] Furthermore, an adaptive control system for the assembly precision of automotive parts is proposed, including:

[0044] The data acquisition module consists of a laser displacement sensor, an industrial CT scanning device, a six-dimensional force sensor, a visual positioning camera, a temperature and humidity sensor, and a vibration sensor. It is used to collect multi-dimensional data during the assembly process and transmit the data to the data preprocessing module via the EtherCAT bus.

[0045] Data preprocessing module: Built-in wavelet transform denoising algorithm, Kalman filter algorithm and time series correlation processing program, used to remove outliers, denoise and align time series of the collected data, and output the fused assembly data matrix;

[0046] Bias Attribution Module: Equipped with a bias attribution model based on an improved random forest-attention mechanism, it receives the assembly data matrix output by the data preprocessing module and outputs the ranking results of bias contribution.

[0047] Compensation calculation module: Stores assembly accuracy margin threshold parameters and historical assembly database, has built-in compensation amount calculation model and reinforcement learning optimization program, calculates and optimizes the final compensation amount based on deviation attribution results;

[0048] Precision monitoring module: Uses an online laser profilometer to detect the assembly precision after compensation, calculate the compensation error, and transmit the evaluation results to the model self-optimization module;

[0049] Model self-optimization module: It has a built-in gradient descent optimization algorithm and model parameter update program. Based on the evaluation results of the accuracy monitoring module, it optimizes the parameters of the bias attribution model and the compensation calculation model.

[0050] Processor: The processor is used to handle the calculation process of each formula and the construction calculation process of each model.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] During the data acquisition phase, a distributed sensor network is used to simultaneously acquire multi-dimensional parameters of component geometry, tooling pose, and environment. These parameters are preprocessed to form a correlated data matrix, providing a comprehensive and reliable data foundation for subsequent analysis. Deviation attribution employs an improved random forest-attention mechanism model, which accurately filters key deviation factors and quantifies their contribution, resolving the ambiguity issue in traditional attribution methods. Compensation calculation combines accuracy margin thresholds with historical data cosine similarity matching, coupled with iterative optimization, to ensure scientific and accurate compensation. Compensation execution can flexibly select active or passive paths depending on the deviation type, with real-time monitoring ensuring adjustments meet standards. Furthermore, after compensation, a laser profilometer accurately detects errors and triggers warnings. The gradient descent algorithm enables model self-optimization, continuously improving accuracy. Combined with integrated blockchain traceability and MES interaction functions, the system balances process traceability and production collaboration, effectively improving assembly accuracy stability and production efficiency. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the method proposed in this invention;

[0054] Figure 2 This is a schematic diagram of the assembly data matrix acquisition method proposed in this invention;

[0055] Figure 3 This is a schematic diagram illustrating the ranking of deviation contribution proposed in this invention;

[0056] Figure 4 This is a schematic diagram illustrating the method for obtaining the final compensation amount as proposed in this invention.

[0057] Figure 5 This is a schematic diagram of the selected compensation execution path proposed in this invention;

[0058] Figure 6 This is a schematic diagram illustrating the calculation of compensation error proposed in this invention;

[0059] Figure 7 This is a schematic diagram of the optimization model parameters proposed in this invention;

[0060] Figure 8 This is a schematic diagram of the synchronously acquired assembly data matrix proposed in this invention;

[0061] Figure 9 This is a schematic diagram illustrating the ranking of deviation contributions proposed in this invention;

[0062] Figure 10 This is a schematic diagram of the compensation calculation and optimization process proposed in this invention;

[0063] Figure 11 This is a schematic diagram of the gradient descent algorithm and model optimization process proposed in this invention;

[0064] Figure 12 This is a schematic diagram illustrating the compensation error calculation and effect evaluation proposed in this invention. Detailed Implementation

[0065] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0066] An adaptive control system for the assembly precision of automotive parts includes:

[0067] The data acquisition module consists of a laser displacement sensor, an industrial CT scanning device, a six-dimensional force sensor, a visual positioning camera, a temperature and humidity sensor, and a vibration sensor. It is used to collect multi-dimensional data during the assembly process and transmit the data to the data preprocessing module via the EtherCAT bus.

[0068] Data preprocessing module: Built-in wavelet transform denoising algorithm, Kalman filter algorithm and time series correlation processing program, used to remove outliers, denoise and align time series of the collected data, and output the fused assembly data matrix;

[0069] Bias Attribution Module: Equipped with a bias attribution model based on an improved random forest-attention mechanism, it receives the assembly data matrix output by the data preprocessing module and outputs the ranking results of bias contribution.

[0070] Compensation calculation module: Stores assembly accuracy margin threshold parameters and historical assembly database, has built-in compensation amount calculation model and reinforcement learning optimization program, calculates and optimizes the final compensation amount based on deviation attribution results;

[0071] Precision monitoring module: Uses an online laser profilometer to detect the assembly precision after compensation, calculate the compensation error, and transmit the evaluation results to the model self-optimization module;

[0072] Model self-optimization module: It has a built-in gradient descent optimization algorithm and model parameter update program. Based on the evaluation results of the accuracy monitoring module, it optimizes the parameters of the bias attribution model and the compensation calculation model.

[0073] Processor: The processor is used to handle the calculation process of each formula and the construction calculation process of each model.

[0074] See Figure 1 As shown, the intelligent compensation method for assembly precision of automotive parts includes:

[0075] Step 1: Based on a distributed sensor network, synchronously collect the geometric parameters of the target parts, the pose parameters of the assembly tooling, and the assembly environment parameters during the automotive parts assembly process, and perform preprocessing to obtain an assembly data matrix containing correlation relationships.

[0076] Step 2: Construct a bias attribution model based on an improved random forest-attention mechanism. Input the acquired assembly data matrix into the model and obtain the ranking of bias contributions.

[0077] Step 3: Based on the ranking results of deviation contribution, construct a compensation amount calculation model. By introducing an assembly accuracy margin threshold, use a piecewise function to calculate the initial compensation amount. Use a cosine similarity algorithm to match similar data in historical assembly data to correct the initial compensation amount and obtain the final compensation amount.

[0078] Step 4: Based on the final compensation amount and the type of core deviation factors, select the corresponding compensation execution path. During the compensation execution process, the adjustment amount is monitored and fed back in real time through displacement sensors.

[0079] Step 5: After the compensation is completed, the assembled parts are inspected using a laser profilometer to obtain the actual assembly accuracy value. The actual assembly accuracy value is then compared with the preset assembly accuracy standard to calculate the compensation error.

[0080] Step Six: Based on the calculated compensation error, construct a model optimization trigger mechanism, iteratively update the model parameters using the gradient descent algorithm, and use the latest assembly data, compensation parameters, and effect evaluation results as training samples to optimize the model's deviation identification accuracy and compensation calculation precision.

[0081] See Figure 2 As shown, based on a distributed sensor network, the geometric parameters of the target parts, the pose parameters of the assembly tooling, and the assembly environment parameters during the automotive parts assembly process are synchronously collected and preprocessed to obtain an assembly data matrix containing correlation relationships. Specifically, this includes:

[0082] The geometric parameters of the target component include the dimensional deviations and geometric tolerances of the key features of the component, which are collected by laser displacement sensors and industrial CT scanning equipment.

[0083] The assembly tooling posture parameters include the tooling positioning pin coordinates and clamping mechanism pressure values, which are collected by a six-dimensional force sensor and a visual positioning camera.

[0084] The assembly environment parameters include the temperature, humidity and vibration frequency of the assembly area, which are collected by temperature and humidity sensors and vibration sensors.

[0085] Data outside the normal range is filtered out, abnormal data is marked and replaced with smoothed data from adjacent time points, high-frequency interference data is removed for geometric parameters, and data smoothing is performed for tooling pose parameters.

[0086] Based on the start time of the assembly process, the timestamps of data from different sources are aligned, and heterogeneous data are mapped to the same assembly timeline coordinate system to form a data matrix that includes the relationship between parts, tooling, and environment.

[0087] See Figure 8 As shown, this illustrates the assembly data matrix synchronously collected through a distributed sensor network during the assembly of automotive parts. Figure 8 In the graph, the vertical axis represents data values, and the horizontal axis represents time (seconds). This graph illustrates changes in three key data sources: geometric parameters, tooling pose parameters, and environmental parameters. These data are collected using various sensors, such as laser displacement sensors, industrial CT scanners, six-dimensional force sensors, and temperature and humidity sensors, to monitor key factors in the assembly process in real time.

[0088] Figure 8 The blue curves represent changes in geometric parameters, reflecting deviations in the shape and size of components during assembly. This data helps assess whether components conform to preset geometric standards during assembly and provides the raw data needed for compensation. The orange curves represent tooling pose parameters, showing tooling positioning errors and pressure variations. These parameters directly affect assembly accuracy and provide important inputs for the compensation model. The green curves represent environmental parameters, reflecting the potential impact of environmental changes during assembly, such as temperature, humidity, and vibration, on assembly accuracy. This invention specifically emphasizes that monitoring environmental changes is crucial for reducing errors and improving accuracy.

[0089] The synchronously collected data is preprocessed to form a data matrix containing correlations, which provides a basis for subsequent deviation attribution analysis and compensation calculation.

[0090] See Figure 3 As shown, a bias attribution model based on an improved random forest-attention mechanism is constructed. The acquired assembly data matrix is ​​input into the model to obtain the ranking of bias contributions. Specifically, this includes:

[0091] The acquired assembly data matrix is ​​transformed into a set of feature variables that can be identified by the attribution model. By improving the random forest algorithm and introducing an out-of-bag data error correction mechanism, the decision tree splitting nodes are optimized. The importance of the feature variables is evaluated, variables with minimal impact on assembly accuracy are removed, and key feature variables are retained.

[0092] The key characteristic variables include component size deviation, tooling positioning error, and ambient temperature fluctuation.

[0093] Based on the attention mechanism, the Pearson correlation coefficient between key feature variables and assembly accuracy deviation is calculated, and weights are assigned according to the absolute value of the correlation coefficient.

[0094] Integrate the feature selection results and weight allocation results, quantify the deviation contribution of each key feature variable, sort them from high to low according to their contribution, and output the deviation contribution ranking results.

[0095] See Figure 9 As shown, the ranking of the contributions of three key characteristic variables (dimensional deviation, tooling positioning error, and ambient temperature fluctuation) to deviation during the assembly process is illustrated. These three characteristics are displayed in the figure using bars of different colors, with the height of the bars representing their degree of influence on the overall assembly accuracy deviation.

[0096] Dimensional deviation (left column) represents the maximum impact of dimensional errors on deviation during component assembly. As mentioned above, dimensional deviations of key component features are collected using laser displacement sensors or industrial CT scanning equipment. The high contribution of this feature means that dimensional error is the most important influencing factor in the assembly process, and therefore, this factor must be given priority consideration in the compensation calculation model.

[0097] The tooling positioning error (intermediate pillar) illustrates the impact of tooling positioning errors on deviations during assembly. Since it involves the coordinates of the tooling positioning pins and the pressure values ​​of the clamping mechanism, these are directly related to the tooling positioning accuracy. Because the positioning accuracy of the tooling affects whether parts are assembled correctly, the contribution of tooling positioning errors is very important and often plays a key role in assembly accuracy compensation models.

[0098] The ambient temperature fluctuations (right bar) reflect the impact of environmental factors on deviations during assembly. Changes in environmental parameters such as temperature, humidity, and vibration frequency in the assembly area can cause thermal expansion or contraction of components, thus affecting assembly accuracy. Although the contribution of environmental factors to this figure is relatively small, they still affect deviations, especially in high-precision assembly tasks, where this factor cannot be ignored.

[0099] Specifically, a basic random forest model is first constructed, and the preprocessed assembly data matrix is ​​split into a training set and a validation set. The model is trained using the feature variable-assembly accuracy deviation as a mapping relationship. Next, an out-of-bag (OOB) error correction mechanism is introduced. For each decision tree in the model, its prediction error is calculated using OOB data that did not participate in the training of that tree. The selection of split nodes of the decision tree is then adjusted according to the error value. If the OOB error corresponding to a certain split node exceeds a preset threshold, the split feature is reselected to avoid the excessive influence of a single feature on the split result. The importance of each feature variable to the assembly accuracy deviation prediction result is calculated through the validation set. An importance threshold is set, and variables with importance below the threshold are removed, finally obtaining a preliminary set of key feature variables.

[0100] The dynamic weight adjustment of key feature variables is combined with changes in the real-time assembly scenario to dynamically optimize the initial weights. When the assembly object is changed to a different model of part, the correlation between the key features and deviation indicators corresponding to that model of part is recalculated. If the correlation is higher than that of the previous model, the weight of that feature is increased accordingly. When a certain type of parameter fluctuates abnormally during the assembly process, the weight of the feature variable corresponding to that type of parameter is temporarily increased to strengthen its impact on deviation attribution. At the same time, the interaction between feature variables is considered. If it is found that the tooling positioning error and the part size deviation are both out of tolerance, and the increase in the deviation indicator is 1.5 times that of a single variable being out of tolerance, the weights of these two variables are increased simultaneously to avoid imbalance in weight allocation due to ignoring the interaction effect and to ensure that the weights can match the actual impact relationship of the assembly process in real time.

[0101] See Figure 4 As shown, based on the ranking results of deviation contribution, a compensation amount calculation model is constructed. By introducing an assembly accuracy margin threshold, a piecewise function is used to calculate the initial compensation amount. Historical assembly data is matched with similar data in the historical assembly data using a cosine similarity algorithm to correct the initial compensation amount. The final compensation amount is obtained by specifically including:

[0102] Based on the obtained deviation contribution ranking, the core deviation factors and their corresponding influence coefficients are determined, and the core factors are used as the main input variables for calculating the compensation amount.

[0103] Based on the assembly of components, an assembly accuracy margin threshold is set, and the initial compensation amount is obtained by calculating the influence coefficient of the core deviation factors through a piecewise function.

[0104] The system calls upon compensation data for similar parts from the historical assembly database, determines the fit with the current scenario through similarity matching, and corrects the initial compensation amount based on similar case parameters.

[0105] The compensation amount is iteratively optimized using the assembly accuracy compliance rate as the reward function. The step size is adjusted by decreasing as the number of iterations increases until the compensation amount meets the accuracy expectation, and the final compensation amount is obtained.

[0106] Specifically, assembly levels are classified according to the functional importance of automotive parts, and the corresponding precision margin thresholds for different levels are clearly defined. Parts that directly affect driving safety, such as brake system valve bodies and steering gears, are classified as critical safety level, and their margin threshold is set at 80% of the preset precision standard. Structural parts that do not directly affect safety, such as body trim parts and chassis brackets, are classified as general structure level, and their margin threshold is set at 90% of the preset precision standard. Core transmission components with extremely high precision requirements, such as engine crankshaft bearings, are set as precision transmission level, and their margin threshold is further tightened to 75% of the preset standard. After the thresholds are set, they are stored in the parameter library for compensation calculation, which can be automatically called when assembling different types of parts.

[0107] The initial compensation amount is determined by segmented logic, which combines the deviation range of the core deviation factors, the comprehensive influence coefficient, and the assembly accuracy margin threshold. When the deviation range of the core factor is in the slight deviation range (e.g., 0.001-0.005mm), the deviation range multiplied by the comprehensive influence coefficient is used as the basic compensation amount. At the same time, it is ensured that the compensation amount does not exceed 80% of the margin threshold. For example, if the dimensional deviation of a certain component is 0.003mm, the comprehensive influence coefficient is 1.2, and the margin threshold is 0.008mm, then the basic compensation amount is 0.003×1.2=0.0036mm, which does not exceed 0.003mm. 0.008 × 80% = 0.0064 mm, so 0.0036 mm is directly used as the initial compensation amount for this range; when the deviation is in the moderate deviation range (0.005-0.01 mm), a margin correction term is added on the basis of the basic compensation amount, that is, an additional compensation amount of 10% of the margin threshold is added. For example, if the deviation is 0.006 mm, then the initial compensation amount is 0.0072 + 0.008 × 10% = 0.008 mm; when the deviation exceeds 0.01 mm (severe deviation range), in addition to the basic compensation amount and the margin correction term, an additional 20% safety redundancy compensation is added to avoid insufficient compensation in a single instance;

[0108] The formula for the basic compensation amount in the slightly out-of-tolerance range is:

[0109]

[0110] in, Basic compensation amount, The deviation margin of the core deviation factor. The comprehensive influence coefficient of the core deviation factors;

[0111] The formula for the initial compensation amount in the moderately out-of-tolerance range is:

[0112]

[0113] in, This represents the initial compensation amount for the moderately out-of-tolerance range. Basic compensation amount, This is the assembly accuracy margin threshold.

[0114] The formula for the initial compensation amount for the severely out-of-tolerance range is:

[0115]

[0116] in, This is the initial compensation amount for the severely out-of-tolerance range. Basic compensation amount, This is the assembly accuracy margin threshold.

[0117] See Figure 10 The diagram illustrates the calculation and optimization process for the compensation amount. Figure 10 It contains two curves:

[0118] Basic compensation amount (solid blue line): This curve represents the calculation process of the basic compensation amount. It is proportional to the deviation range, that is, the compensation amount is 0.5 times the deviation range. This represents the compensation amount initially calculated directly based on the deviation range, which is suitable for simple compensation strategies, but does not consider other factors.

[0119] Optimized Compensation Amount (Green Dashed Line): This curve represents the optimized compensation amount based on the basic compensation amount. The optimization process introduces an adjustment factor to make finer adjustments during the compensation process. This optimized compensation amount considers more factors, such as historical data similarity matching and dynamic adjustment of the assembly precision margin threshold, enabling more accurate correction of the initially calculated compensation amount. In other words, by introducing historical data similarity matching and precision margin adjustment, the compensation amount is optimized to achieve better assembly precision.

[0120] By comparing these two curves, it can be seen that the optimized compensation amount has a smoother adjustment effect than the basic compensation amount. Especially when the deviation is large, the optimized compensation amount can more effectively control the compensation range and avoid the risk of over-compensation or under-compensation that the basic compensation amount may bring.

[0121] See Figure 5 As shown, based on the final compensation amount and the type of core deviation factors, the corresponding compensation execution path is selected. During the compensation execution process, the adjustment amount is monitored and fed back in real time through displacement sensors, specifically including:

[0122] Based on the final compensation amount and the type of deviation factor obtained, determine whether to adopt a passive compensation path or an active compensation path.

[0123] Passive compensation is based on the final compensation amount to calculate the required thickness of the compensation shim, and then the compensation shim of the required thickness is embedded into the assembly gap of the parts to fill the assembly gap caused by dimensional deviation, thus completing the passive compensation.

[0124] Active compensation is based on the final compensation amount and gradually corrects the position of the tooling positioning pin according to the adjustment range to achieve tooling posture correction.

[0125] Throughout the compensation process, the deviation between the actual adjustment results and the preset compensation amount is checked in real time to confirm that the compensation operation meets the standards.

[0126] Specifically, a real-time deviation early warning mechanism is established throughout the compensation process. During passive compensation, if the measurement finds that the assembly gap suddenly increases, the operation is immediately stopped, and the parts are checked for misalignment or surface damage. After the problem is eliminated, the measurement is repeated. If the thickness of the selected shim exceeds the maximum allowable compensation thickness of this type of part, a compensation over-limit warning is triggered, the core deviation factor judgment result is re-evaluated, and it is confirmed whether there are any unidentified tooling or environmental interference factors.

[0127] During active compensation, if the rate of change of parameters exceeds the preset threshold during the adjustment process, the adjustment range will be automatically slowed down to prevent sudden changes in parameters; if the parameters show a reverse deviation after adjustment, the adjustment will be paused and reversed for correction, while the calibration status of the parameter measuring equipment will be checked to ensure the reliability of the measurement data.

[0128] See Figure 6 As shown, after the compensation is completed, the assembled parts are inspected using a laser profilometer to obtain the actual assembly accuracy value. The actual assembly accuracy value is then compared with the preset assembly accuracy standard to calculate the compensation error, which specifically includes:

[0129] Based on the compensated assembled parts, a laser profilometer is used to perform accuracy detection and obtain the actual accuracy value.

[0130] The actual assembly accuracy value is compared with the preset assembly accuracy standard, and the deviation value between the two is used as the compensation error to quantify the compensation effect.

[0131] If the compensation error is ≤0.01mm, the compensation is deemed qualified; if the compensation error is >0.01mm, a secondary compensation warning is triggered, and the compensation parameters are recorded to the historical database.

[0132] For details, please refer to Figure 7 As shown, based on the calculated compensation error, a model optimization triggering mechanism is constructed. The model parameters are iteratively updated using the gradient descent algorithm. The latest assembly data, compensation parameters, and effect evaluation results are used as training samples to optimize the model's deviation identification accuracy and compensation calculation precision. Specifically, this includes:

[0133] Set model optimization trigger conditions, including the average compensation error of three consecutive assembly batches being greater than 0.008 mm and the occurrence rate of secondary compensation in a single assembly batch exceeding 10%, and monitor in real time whether the trigger conditions are met.

[0134] If the optimization conditions are triggered, the model to be optimized is locked, namely the deviation attribution model and the compensation amount calculation model. The original data, actual compensation parameters and compensation effect evaluation results of the latest assembly process are integrated to form a model training sample set.

[0135] The gradient descent algorithm is used to iteratively update the model parameters using data from the sample set as input, thereby reducing the model's prediction error.

[0136] like Figure 11 The figure illustrates the process of parameter optimization for the bias attribution model and the compensation calculation model based on the gradient descent algorithm. The horizontal axis represents the number of iterations, the vertical axis represents the model's loss value, and the green curve represents the model's performance error in each iteration. Figure 11 It can be clearly seen that the initial loss value of the model is relatively high, about 0.08, but as the number of iterations increases, the loss value gradually decreases in an exponential manner and tends to stabilize, eventually converging to a level close to or below 0.01, indicating that the prediction accuracy of the model has been significantly improved.

[0137] In this application, a triggering condition for model optimization is set, such as if the average compensation error of multiple consecutive batches is too high or the occurrence rate of secondary compensation is too high. Once triggered, the system will use the latest assembly data and compensation effect as training samples and use the gradient descent algorithm to iteratively update the model parameters. Figure 11 The gradual convergence of the green waveform is a direct manifestation of this mechanism. It shows that the model continuously corrects its parameters during the optimization process to reduce errors and improve the accuracy of bias identification and the calculation precision of compensation.

[0138] also, Figure 11 The waveform in the middle has slight fluctuations during the descent process, which simulates the disturbance caused by data changes or noise during the optimization process, and reflects the stability and robustness of model iteration in actual industrial scenarios.

[0139] Specifically, based on the preset assembly accuracy standard, the compensation error for each inspection item is calculated one by one. The compensation error for a certain inspection item is calculated as |actual accuracy value of the inspection item - preset accuracy standard value|. For example, the actual coaxiality of the shaft hole is 0.008mm, and the preset standard value is 0.01mm, so the compensation error is 0.002mm; the actual contact area of ​​the gear tooth surface is 82%, and the preset standard value is 85%, so the compensation error is 3%. After the calculation is completed, the errors are classified and statistically analyzed according to the type of inspection item: dimensional errors, geometric errors, and functional errors are classified separately to form an error statistics list, clearly showing the compensation effect of each type of accuracy.

[0140] Multi-dimensional qualification assessment and result grading are based on the statistical results of compensation errors, and are judged from three dimensions: single-item qualification, full-item qualification, and stability qualification. Single-item qualification assessment: if the compensation error of a certain test item is ≤ the preset qualification threshold, then the item is judged as qualified. Full-item qualification assessment: all key precision test items must meet the single-item qualification standard, and no error of any item must exceed the threshold for the overall compensation to be judged as qualified. Stability qualification assessment: for three consecutively compensated parts in the same batch, the average compensation error of their corresponding test items is calculated. If the average value is ≤ 0.008mm and the error fluctuation range is ≤ 0.003mm, it indicates that the compensation effect is stable, and the batch is judged as stable. According to the judgment results, the compensation effect is divided into three levels: excellent, qualified, and unqualified.

[0141] See Figure 12 As shown in the figure, a heatmap is used to display the calculation results of compensation errors for three key inspection items in three assembly batches. The horizontal axis represents the assembly inspection items, and the vertical axis represents the batch number. The value in each cell represents the compensation error value of the corresponding inspection item in that batch. The darker the color, the closer the error is to the preset threshold upper limit; the lighter the color, the more ideal the compensation effect.

[0142] The dimensional error, form and position error, and functional error in the diagram correspond to the precision inspection dimensions mentioned above. These dimensions are measured using equipment such as laser profilometers during actual assembly precision inspection and compared with preset precision standards to calculate compensation errors. For example, the dimensional error in the first batch is 0.006 mm, far below the acceptable threshold of 0.01 mm, indicating that the compensation precision meets the standard; the dimensional error in the third batch is 0.009 mm, which, although still within the acceptable range, is close to the threshold, indicating that the compensation in this batch has slight fluctuations.

[0143] Furthermore, if the average compensation error of multiple consecutive batches exceeds a certain level, the system will trigger a self-optimization mechanism. Figure 12 It provides an intuitive evaluation basis for determining whether the model's stability meets the standards, and also helps to detect the trend of declining compensation accuracy in advance.

[0144] The servo motor used during compensation execution employs an absolute encoder to provide real-time feedback on motor rotation accuracy. The blockchain database for data storage and interaction adopts a consortium blockchain architecture, with each assembly data block containing a timestamp, equipment number, operator information, and digital signature to ensure the effectiveness of data traceability. The MES system interface supports the OPC UA protocol, enabling real-time bidirectional data interaction between the system and the MES system. When assembly accuracy is abnormal, a production pause request is automatically sent to the MES system.

[0145] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0146] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0147] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent compensation method for assembly precision of automotive parts, characterized in that, include: Based on a distributed sensor network, the geometric parameters of the target parts, the pose parameters of the assembly tooling, and the assembly environment parameters during the assembly process of automotive parts are collected synchronously and preprocessed to obtain an assembly data matrix containing correlation relationships. A deviation attribution model based on an improved random forest-attention mechanism is constructed. The obtained assembly data matrix is ​​input into the model to obtain the ranking of deviation contributions. The deviation attribution model based on the improved random forest-attention mechanism optimizes the decision tree splitting nodes by introducing an out-of-bag data error correction mechanism, evaluates the importance of feature variables, removes variables with minimal impact on assembly accuracy, and retains key feature variables. Based on the ranking results of deviation contribution, a compensation amount calculation model is constructed. By introducing an assembly accuracy margin threshold, a piecewise function is used to calculate the initial compensation amount. Historical assembly data is matched with similar data in the historical assembly data through a cosine similarity algorithm to correct the initial compensation amount and obtain the final compensation amount. Based on the final compensation amount and the type of core deviation factors, the corresponding compensation execution path is selected. During the compensation execution process, the adjustment amount is monitored and fed back in real time through displacement sensors. After the compensation is completed, the assembled parts are inspected using a laser profilometer to obtain the actual assembly accuracy value. The actual assembly accuracy value is then compared with the preset assembly accuracy standard to calculate the compensation error. Based on the calculated compensation error, a model optimization triggering mechanism is constructed. The model parameters are iteratively updated using the gradient descent algorithm. The latest assembly data, compensation parameters, and effect evaluation results are used as training samples to optimize the model's deviation identification accuracy and compensation calculation precision.

2. The intelligent compensation method for assembly accuracy of automotive parts according to claim 1, characterized in that, The process of synchronously collecting and preprocessing the geometric parameters of target components, the pose parameters of assembly fixtures, and the assembly environment parameters during the automotive parts assembly process, based on a distributed sensor network, to obtain an assembly data matrix containing correlations, specifically includes: The geometric parameters of the target component include the dimensional deviations and geometric tolerances of the key features of the component, which are collected by laser displacement sensors and industrial CT scanning equipment. The assembly tooling posture parameters include the tooling positioning pin coordinates and clamping mechanism pressure values, which are collected by a six-dimensional force sensor and a visual positioning camera. The assembly environment parameters include the temperature, humidity and vibration frequency of the assembly area, which are collected by temperature and humidity sensors and vibration sensors. Data outside the normal range is filtered out, abnormal data is marked and replaced with smoothed data from adjacent time points, high-frequency interference data is removed for geometric parameters, and data smoothing is performed for tooling pose parameters. Based on the start time of the assembly process, the timestamps of data from different sources are aligned, and heterogeneous data are mapped to the same assembly timeline coordinate system to form a data matrix that includes the relationship between parts, tooling, and environment.

3. The intelligent compensation method for assembly accuracy of automotive parts according to claim 1, characterized in that, The construction of the bias attribution model based on the improved random forest-attention mechanism involves inputting the acquired assembly data matrix into the model to obtain the ranking of bias contributions. Specifically, this includes: The acquired assembly data matrix is ​​transformed into a set of feature variables that can be identified by the attribution model. By improving the random forest algorithm and introducing an out-of-bag data error correction mechanism, the decision tree splitting nodes are optimized. The importance of the feature variables is evaluated, variables with minimal impact on assembly accuracy are removed, and key feature variables are retained. The key characteristic variables include component size deviation, tooling positioning error, and ambient temperature fluctuation. Based on the attention mechanism, the Pearson correlation coefficient between key feature variables and assembly accuracy deviation is calculated, and weights are assigned according to the absolute value of the correlation coefficient. Integrate the feature selection results and weight allocation results, quantify the deviation contribution of each key feature variable, sort them from high to low according to their contribution, and output the deviation contribution ranking results.

4. The intelligent compensation method for assembly accuracy of automotive parts according to claim 1, characterized in that, The compensation calculation model is constructed based on the deviation contribution ranking results. By introducing an assembly accuracy margin threshold, a piecewise function is used to calculate the initial compensation amount. Historical assembly data is matched with similar data in the historical assembly data using a cosine similarity algorithm to correct the initial compensation amount. The final compensation amount is obtained by specifically including: Based on the obtained deviation contribution ranking, the core deviation factors and their corresponding influence coefficients are determined, and the core factors are used as the main input variables for calculating the compensation amount. Based on the assembly of components, an assembly accuracy margin threshold is set, and the initial compensation amount is obtained by calculating the influence coefficient of the core deviation factors through a piecewise function. The system calls upon compensation data for similar parts from the historical assembly database, determines the fit with the current scenario through similarity matching, and corrects the initial compensation amount based on similar case parameters. The compensation amount is iteratively optimized using the assembly accuracy compliance rate as the reward function. The step size is adjusted by decreasing as the number of iterations increases until the compensation amount meets the accuracy expectation, and the final compensation amount is obtained.

5. The intelligent compensation method for assembly accuracy of automotive parts according to claim 1, characterized in that, The selection of the corresponding compensation execution path based on the final compensation amount and the type of core deviation factors, and the real-time monitoring and feedback of the adjustment amount through displacement sensors during the compensation execution process, specifically include: Based on the final compensation amount and the type of deviation factor obtained, determine whether to adopt a passive compensation path or an active compensation path. Passive compensation is based on the final compensation amount to calculate the required thickness of the compensation shim, and then the compensation shim of the required thickness is embedded into the assembly gap of the parts to fill the assembly gap caused by dimensional deviation, thus completing the passive compensation. Active compensation is based on the final compensation amount and gradually corrects the position of the tooling positioning pin according to the adjustment range to achieve tooling posture correction. Throughout the compensation process, the deviation between the actual adjustment results and the preset compensation amount is checked in real time to confirm that the compensation operation meets the standards.

6. The intelligent compensation method for assembly accuracy of automotive parts according to claim 1, characterized in that, After the compensation is completed, the assembled parts are inspected using a laser profilometer to obtain the actual assembly accuracy value. The actual assembly accuracy value is then compared with the preset assembly accuracy standard to calculate the compensation error, which specifically includes: Based on the compensated assembled parts, a laser profilometer is used to perform accuracy detection and obtain the actual accuracy value. The actual assembly accuracy value is compared with the preset assembly accuracy standard, and the deviation value between the two is used as the compensation error to quantify the compensation effect. If the compensation error is ≤0.01mm, the compensation is deemed qualified; if the compensation error is >0.01mm, a secondary compensation warning is triggered, and the compensation parameters are recorded to the historical database.

7. The intelligent compensation method for assembly accuracy of automotive parts according to claim 1, characterized in that, The aforementioned model optimization triggering mechanism, based on the calculated compensation error, iteratively updates the model parameters using a gradient descent algorithm, and uses the latest assembly data, compensation parameters, and effect evaluation results as training samples to optimize the model's deviation identification accuracy and compensation calculation precision. Specifically, this includes: Set model optimization trigger conditions, including the average compensation error of three consecutive assembly batches being greater than 0.008 mm and the occurrence rate of secondary compensation in a single assembly batch exceeding 10%, and monitor in real time whether the trigger conditions are met. If the optimization conditions are triggered, the model to be optimized is locked, namely the deviation attribution model and the compensation amount calculation model. The original data, actual compensation parameters and compensation effect evaluation results of the latest assembly process are integrated to form a model training sample set. The gradient descent algorithm is used to iteratively update the model parameters using data from the sample set as input, thereby reducing the model's prediction error.

8. An adaptive control system for automotive component assembly precision, used to implement the intelligent compensation method for automotive component assembly precision as described in any one of claims 1-7, characterized in that, include: The data acquisition module consists of a laser displacement sensor, an industrial CT scanning device, a six-dimensional force sensor, a visual positioning camera, a temperature and humidity sensor, and a vibration sensor. It is used to collect multi-dimensional data during the assembly process and transmit the data to the data preprocessing module via the EtherCAT bus. Data preprocessing module: Built-in wavelet transform denoising algorithm, Kalman filter algorithm and time series correlation processing program, used to remove outliers, denoise and align time series of the collected data, and output the fused assembly data matrix; Bias Attribution Module: Equipped with a bias attribution model based on an improved random forest-attention mechanism, it receives the assembly data matrix output by the data preprocessing module and outputs the ranking results of bias contribution. Compensation calculation module: Stores assembly accuracy margin threshold parameters and historical assembly database, has built-in compensation amount calculation model and reinforcement learning optimization program, calculates and optimizes the final compensation amount based on deviation attribution results; Precision monitoring module: Uses an online laser profilometer to detect the assembly precision after compensation, calculate the compensation error, and transmit the evaluation results to the model self-optimization module; Model self-optimization module: It has a built-in gradient descent optimization algorithm and model parameter update program. Based on the evaluation results of the accuracy monitoring module, it optimizes the parameters of the bias attribution model and the compensation calculation model. Processor: The processor is used to handle the calculation process of each formula and the construction calculation process of each model.

9. The adaptive control system for assembly precision of automotive parts according to claim 8, characterized in that, The servo motor used during compensation execution employs an absolute encoder to provide real-time feedback on the motor's rotational accuracy; the blockchain database for data storage and interaction adopts a consortium blockchain architecture, with each assembly data block containing a timestamp, equipment number, operator information, and digital signature to ensure the effectiveness of data traceability; The MES system interface supports the OPC UA protocol, enabling real-time bidirectional data interaction between the system and the MES system. When assembly accuracy is abnormal, it automatically sends a production pause request to the MES system.

Images