A method and system for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames.

By deploying sensors and gradient boosting tree models in the electrophoresis process, combined with PID controllers, real-time prediction and stable control of paint film thickness were achieved, solving the problems of poor paint film thickness uniformity and high rework rate in existing technologies, and improving production efficiency and material utilization.

CN123108702APending Publication Date: 2026-11-17TIANJIN GOLDEN ARROW IND CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611416035.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-14
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the thickness of the paint film cannot be predicted online in real time. The adjustment of electrophoresis parameters depends on human experience and the coupling of multiple factors makes it difficult to model accurately, resulting in poor uniformity of paint film thickness and high rework rate.

Method used

By deploying sensors to collect electrophoresis process parameters in real time, using a gradient boosting tree prediction model to predict the coating thickness, and combining this with a PID controller to adjust process parameters in real time, data-driven global optimal control is achieved.

Benefits of technology

It enables real-time prediction and stable control of paint film thickness, reduces rework rate, improves production efficiency and material utilization, adapts to working condition drift, and ensures the uniformity and consistency of paint film thickness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This invention discloses a method and system for predicting and controlling the coating thickness in an electrophoresis process for vehicle frames. The method utilizes multiple sensors deployed on the electrophoresis production line to collect real-time data on process parameters such as electrophoresis voltage, current, bath temperature, solids content, and electrophoresis time. The collected data is preprocessed and feature-engineered to obtain a normalized multidimensional feature vector. This multidimensional feature vector is then input into a pre-trained gradient boosting tree prediction model, which outputs a predicted coating thickness value for the current batch of vehicle frames. The predicted value is compared with the target thickness value to calculate the prediction deviation and determine an adjustment strategy. Based on the adjustment strategy, a PID controller generates a smooth adjustment amount, which is then driven by a PLC controller to adjust the electrophoresis process parameters in real time, maintaining the coating thickness within the target range. This invention enables online real-time prediction and automatic control of coating thickness, significantly improving coating thickness consistency, reducing rework rates, and is suitable for mixed-model production lines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of automotive frame electrophoresis technology, specifically relating to a method and system for predicting and controlling the coating thickness in automotive frame electrophoresis. Background Technology

[0002] Electrophoretic coating is a core process for corrosion protection of large metal structural components such as automobile frames. It utilizes a direct current electric field to drive charged paint particles to migrate directionally and deposit on the workpiece surface. After high-temperature baking and curing, an anti-corrosion protective layer is formed. The thickness of the paint film is the core indicator for measuring the quality of electrophoresis. Insufficient thickness will lead to a decrease in protective ability, while excessive thickness will result in paint waste and increase the risk of defects such as sagging and orange peel.

[0003] Currently, paint film thickness testing mainly relies on offline sampling in laboratories. Operators extract samples from the production line and measure them using a magnetic thickness gauge. From sampling to report generation, it typically takes 2 to 4 hours, resulting in a significant time lag. By the time the test results confirm that the product does not meet the standards, hundreds of workpieces in the same batch have already completed the electrophoresis process, leading to extremely serious losses due to rework or scrapping.

[0004] Electrophoretic film formation is a complex electrochemical process involving the coupling of multiple physical fields, with highly nonlinear coupling relationships among process parameters such as voltage, current, and bath temperature. While increasing the voltage can increase the deposition rate, the resulting increase in current and bath temperature will alter the coating's conductivity and penetration, thus negatively impacting deposition efficiency. Operators manually adjusting parameters based solely on experience lack quantitative decision-making support, making it difficult to precisely control the timing and magnitude of adjustments, leading to frequent fluctuations in product quality.

[0005] In recent years, some studies have attempted to introduce machine learning methods such as neural networks and support vector machines into coating quality prediction. However, neural networks require a large training sample size and are prone to overfitting when historical data is limited. The models operate in a "black box" mode, lacking interpretability; process engineers cannot understand the prediction logic and find it difficult to use them as a basis for operation. Most existing research stops at prediction, failing to establish a link from prediction results to parameter control, thus failing to fundamentally change the post-event remedial quality control model. After deployment, the models lack adaptive update mechanisms, making them unable to adapt to changes in operating conditions such as bath aging, paint batch differences, and environmental changes, resulting in a gradual decrease in prediction accuracy after long-term operation. In terms of control, existing technologies either employ rule-based logic control or use PID controllers to independently control a single parameter, failing to organically combine multi-parameter fusion modeling with control, and thus failing to achieve globally optimal control based on real-time quality prediction. Summary of the Invention

[0006] To address this, the present invention provides a method and system for predicting and controlling the coating thickness in the electrophoresis process of vehicle frames, which solves the problems in the prior art where the coating thickness cannot be predicted online in real time, the adjustment of electrophoresis parameters depends on manual experience and multiple factors are coupled, making it difficult to accurately model, resulting in poor coating thickness uniformity and high rework rate.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames, comprising the following steps:

[0008] Sensors deployed on the electrophoresis production line are used to collect electrophoresis process parameter data in real time. The electrophoresis process parameter data includes at least electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time.

[0009] The collected electrophoresis process parameter data are preprocessed and feature-engineered to obtain normalized multidimensional feature vectors.

[0010] The multidimensional feature vector is input into a pre-trained gradient boosting tree prediction model. The gradient boosting tree prediction model is trained with electrophoresis process parameter data as input features and measured paint film thickness as labels, and outputs the predicted paint film thickness of the current batch of frames.

[0011] The predicted film thickness is compared with the target thickness value set in the process specification, the prediction deviation is calculated, and the adjustment strategy of the process parameters is determined based on the magnitude and direction of the prediction deviation.

[0012] According to the aforementioned adjustment strategy, a smooth process parameter adjustment amount is generated by the PID controller, and the PLC controller drives the actuator to adjust the electrophoresis process parameters in real time, so that the coating thickness is kept within the target range.

[0013] As a preferred embodiment of the coating thickness prediction and control method for vehicle frame electrophoresis process, the sensors include a voltage sensor, a current sensor, a temperature sensor, a solids analyzer, and a timer, used to collect the electrophoresis voltage, the electrophoresis current, the bath temperature, the solids content, and the electrophoresis time; the sensors also include a conductivity sensor and a workpiece model identification device, and the electrophoresis process parameter data also include the bath conductivity and workpiece model code; all sensor data are aggregated to the data acquisition gateway via the industrial Ethernet protocol, and synchronously transmitted to the industrial control computer after being unified with a timestamp.

[0014] As a preferred method for predicting and controlling the coating thickness in the electrophoretic coating process for vehicle frames, the data preprocessing includes:

[0015] Outlier data points exceeding the physically reasonable range are removed using the 3σ criterion, and missing data caused by communication interruptions or sensor failures are filled using linear interpolation. After outlier removal and missing data filling, a moving average filtering algorithm is used to smooth the time-series data. The moving average filtering algorithm involves taking the arithmetic mean of continuous sampled values ​​within a preset window length W, where W ranges from 5 to 10. The filtered value after the moving average filtering is expressed as:

[0016] ,

[0017] In the formula, For a moment The original sampled values ​​of the specified electrophoresis process parameters are used. For a moment The corresponding filter value for the process parameters. This is the size of the sliding window.

[0018] As a preferred embodiment of the method for predicting and controlling the coating thickness in the electrophoretic coating process for vehicle frames, the feature engineering includes:

[0019] Based on the collected basic features of electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time, six derived features are constructed: electrophoresis voltage change rate, current density, bath temperature change rate, solids content change rate, cumulative electrochemical effect and bath conductivity, resulting in an 11-dimensional feature vector.

[0020] The cumulative electrochemical action Using approximate formulas Calculation, where The electrophoresis voltage is... For electrophoresis current, The electrophoresis time is used; the Min-Max normalization method is used to map all features to the [0,1] interval, and the normalization formula is:

[0021] ,

[0022] In the formula, These are the normalized eigenvalues. This represents the original characteristic value of a certain electrophoresis process parameter. and These are the minimum and maximum values ​​of the electrophoresis process parameter in the training dataset, respectively.

[0023] As a preferred method for predicting and controlling the coating thickness in the electrophoretic coating process of vehicle frames, the gradient boosting tree prediction model is composed of a weighted sum of multiple regression trees, expressed as:

[0024] ,

[0025] In the formula, The predicted value of paint film thickness output by the model. For the first Regression trees for input feature vectors The output paint film thickness prediction component, The total number of regression trees; the gradient boosting tree prediction model is constructed through an iterative process, with the output of the new tree in each round fitting the negative gradient of the previous round's model, and the total number of regression trees. This was determined through cross-validation.

[0026] As a preferred method for predicting and controlling the coating thickness in the vehicle frame electrophoresis process, the gradient boosting tree prediction model is trained based on minimizing the objective function, which includes a mean squared error loss term and a regularization penalty term.

[0027] ,

[0028] In the formula, For the first Measured values ​​of paint film thickness for each sample. For the first Predicted film thickness for each sample The total number of training samples; the regularization penalty term for:

[0029] ,

[0030] In the formula, For the first The number of leaf nodes in the regression tree. For the first Tree No. The output weights of the leaf nodes The regularization hyperparameter that penalizes the number of leaf nodes. The L2 regularization hyperparameter is used to determine the weights of the leaf nodes; the regularization hyperparameter is determined by Bayesian optimization or grid search methods.

[0031] As a preferred method for predicting and controlling the coating thickness in the electrophoretic coating process for vehicle frames, the prediction deviation The difference between the predicted paint film thickness and the target thickness is expressed as:

[0032] ,

[0033] In the formula, This is the predicted coating thickness of the electrocoated batch of vehicle frames at the current moment. The target thickness value set for the process specifications;

[0034] When the absolute value of the prediction deviation is not greater than the preset deviation tolerance At this time, the current process parameters remain unchanged;

[0035] When the predicted deviation is greater than the deviation tolerance and the predicted coating thickness is greater than the target thickness, reduce the electrophoresis voltage or shorten the electrophoresis time.

[0036] When the predicted deviation is less than the negative value of the deviation tolerance and the predicted value of the coating thickness is less than the target thickness value, the electrophoresis voltage is increased or the electrophoresis time is extended.

[0037] As a preferred embodiment of the paint film thickness prediction and control method for automotive frame electrophoresis, the PID controller calculates the adjustment amount based on the weighted sum of the proportional, integral, and derivative terms of the control error, and the output of the PID controller... Represented as:

[0038] ,

[0039] In the formula, This represents the current error in controlling the paint film thickness of the electrophoretic batch of vehicle frames. This is the proportionality coefficient. The integral coefficient is... The differential coefficients are: the proportional term responds to the current error, the integral term eliminates the steady-state error, and the differential term predicts the error change trend to suppress overshoot; the actuator includes a rectifier, a variable frequency pump, and a heat exchange system, which are used to adjust the electrophoresis voltage, electrophoresis time, and bath temperature, respectively.

[0040] As a preferred method for predicting and controlling the coating thickness in the electrophoretic coating process for vehicle frames, it also includes an online self-learning update step for the model:

[0041] After each batch of electrophoresis production is completed, the measured values ​​of the coating thickness obtained from random sampling and the corresponding batch of process parameter data are added to the training dataset; every predetermined number of batches, incremental learning updates are performed on the current model using the new samples; when the prediction error continues to exceed the preset threshold in several consecutive batches, the model is triggered to be fully retrained.

[0042] This invention also provides a paint film thickness prediction and control system for vehicle frame electrophoresis process, employing the above-mentioned paint film thickness prediction and control method for vehicle frame electrophoresis process, comprising:

[0043] The data acquisition module is used to collect electrophoresis process parameter data in real time through sensors deployed on the electrophoresis production line. The electrophoresis process parameter data includes at least electrophoresis voltage, electrophoresis current, tank temperature, solids content, and electrophoresis time.

[0044] The data preprocessing and feature construction module is used to perform data preprocessing and feature engineering on the collected electrophoresis process parameter data to obtain normalized multidimensional feature vectors.

[0045] The paint film thickness prediction module is used to input the multi-dimensional feature vector into a pre-trained gradient boosting tree prediction model. The gradient boosting tree prediction model is trained with electrophoresis process parameter data as input features and measured paint film thickness as labels, and outputs the predicted paint film thickness of the current batch of frames.

[0046] The deviation calculation and control decision module is used to compare the predicted value of the paint film thickness with the target thickness value set by the process specification, calculate the prediction deviation, and determine the adjustment strategy of the process parameters based on the magnitude and direction of the prediction deviation.

[0047] The PID control and execution module is used to generate smooth process parameter adjustment amounts through the PID controller according to the control strategy, and drive the actuator through the PLC controller to adjust the electrophoresis process parameters in real time so that the coating thickness is maintained within the target range.

[0048] The present invention has the following advantages:

[0049] This invention outputs the predicted value of the coating film thickness in real time during the electrophoresis process. The prediction period is synchronized with the data acquisition frequency, eliminating the need to wait for offline laboratory test results and providing instant quality status information for process control.

[0050] This invention calculates the adjustment amount based on the prediction results, and smoothly outputs it to the actuator via a PID controller, eliminating the subjectivity and lag of manual experience-based adjustment. It transforms the adjustment of electrophoresis process parameters from experience-driven to data-driven, achieving global optimal control based on real-time quality prediction.

[0051] This invention uses a control mechanism to keep the paint film thickness consistently within the allowable tolerance range near the target value, effectively suppressing the problem of large fluctuations in paint film thickness in traditional processes and ensuring uniform paint film thickness across the entire batch of products.

[0052] This invention replaces the traditional post-processing inspection and rework model with pre-production prediction and process control, significantly saving paint and energy consumption, improving the overall efficiency of the production line, and reducing material waste and lost time. Through online incremental learning and periodic full retraining, it can continuously adapt to the drift of working conditions such as bath aging, paint batch differences, and environmental changes. After long-term operation, the prediction accuracy is stable, and it has strong versatility and robustness.

[0053] This invention incorporates the chassis model code as a classification feature into the model input, enabling adaptive prediction and control under mixed production line conditions for multiple vehicle models. The system automatically loads the prediction model and control parameters for the corresponding vehicle model without manual switching, effectively solving the problem of electric field distribution and deposition efficiency variations caused by differences in the geometric structure of different vehicle models. Attached Figure Description

[0054] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0055] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0056] Figure 1 This is a schematic diagram of the process for predicting and controlling the coating thickness in the electrophoretic coating process of a vehicle frame, provided in an embodiment of the present invention.

[0057] Figure 2 This is an overall flowchart of online prediction and control in the paint film thickness prediction and control method for vehicle frame electrophoresis provided in this embodiment of the invention;

[0058] Figure 3 This is a sensor deployment architecture diagram in the method for predicting and controlling the coating thickness in the vehicle frame electrophoresis process provided in this embodiment of the invention;

[0059] Figure 4 This is a flowchart of data preprocessing and feature engineering in the method for predicting and controlling the coating thickness of the vehicle frame electrophoresis process provided in this embodiment of the invention;

[0060] Figure 5 This is a flowchart of the prediction model training process in the method for predicting and controlling the coating thickness of the vehicle frame electrophoresis process provided in this embodiment of the invention.

[0061] Figure 6 This is a diagram illustrating the online model update mechanism in the method for predicting and controlling the coating thickness in the vehicle frame electrophoresis process provided in this embodiment of the invention.

[0062] Figure 7 This is a comparison and analysis chart of paint film thickness consistency provided in an embodiment of the present invention;

[0063] Figure 8 A comparative analysis chart of pass rate and rework rate provided for embodiments of the present invention;

[0064] Figure 9 A comparative analysis chart of process parameter adjustment frequency and amplitude provided in the embodiments of the present invention;

[0065] Figure 10 This is a performance comparison analysis chart of three prediction models provided in the embodiments of the present invention;

[0066] Figure 11 This is a statistical analysis chart of paint film thickness for multi-model mixed-line production provided in an embodiment of the present invention;

[0067] Figure 12 This is an analysis diagram of the model self-learning adaptive recovery process provided in the embodiments of the present invention;

[0068] Figure 13 This is a diagram illustrating the architecture of a vehicle frame electrophoresis coating thickness prediction and control system provided in an embodiment of the present invention. Detailed Implementation

[0069] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Example 1

[0071] See Figure 1 and Figure 2 This invention provides a method for predicting and controlling the coating thickness in a vehicle frame electrophoresis process, comprising the following steps:

[0072] Sensors deployed on the electrophoresis production line are used to collect electrophoresis process parameter data in real time. The electrophoresis process parameter data includes at least electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time.

[0073] The collected electrophoresis process parameter data are preprocessed and feature-engineered to obtain normalized multidimensional feature vectors.

[0074] The multidimensional feature vector is input into a pre-trained gradient boosting tree prediction model. The gradient boosting tree prediction model is trained with electrophoresis process parameter data as input features and measured paint film thickness as labels, and outputs the predicted paint film thickness of the current batch of frames.

[0075] The predicted film thickness is compared with the target thickness value set in the process specification, the prediction deviation is calculated, and the adjustment strategy of the process parameters is determined based on the magnitude and direction of the prediction deviation.

[0076] According to the aforementioned adjustment strategy, a smooth process parameter adjustment amount is generated by the PID controller, and the PLC controller drives the actuator to adjust the electrophoresis process parameters in real time, so that the coating thickness is kept within the target range.

[0077] In this embodiment, the sensors include a voltage sensor, a current sensor, a temperature sensor, a solids analyzer, and a timer, used to collect the electrophoresis voltage, the electrophoresis current, the bath temperature, the solids content, and the electrophoresis time. The sensors also include a conductivity sensor and a workpiece model identification device. The electrophoresis process parameter data also include the bath conductivity and the workpiece model code. All sensor data are aggregated to the data acquisition gateway via the industrial Ethernet protocol and synchronously transmitted to the industrial control computer after being timestamped.

[0078] See Figure 3 Specifically, multiple sets of sensors are deployed on the electrophoresis production line to achieve comprehensive monitoring of the process status. A DC voltage sensor is installed at the rectifier output, with a sampling frequency set to 5 to 10 Hz and a range covering 0 to 500 volts. This range is compatible with the operating voltage range of most cathodic electrophoresis processes. A Hall current sensor is installed in the main electrophoresis circuit to synchronously acquire the electrophoresis current signal, with a range of 0 to 5000 amperes, accurately reflecting the migration intensity of charged particles in the electrophoresis tank. The tank temperature is measured using a Pt100 platinum resistance temperature sensor. Multiple temperature measurement points are set up in different areas of the electrophoresis tank, and the arithmetic mean of the readings at each point is taken as the tank temperature value, with a range covering 15 to 40 degrees Celsius and a measurement accuracy of ±0.5 degrees Celsius. An online solids analyzer is installed in the circulation pipeline of the electrophoresis tank, continuously monitoring the mass percentage of solid particles in the tank solution based on near-infrared spectroscopy or refractive index principles, with a range of 10% to 30% and a measurement accuracy of ±0.5 percentage points. Electrophoresis time is precisely recorded by a timer to determine the duration of workpiece immersion in the electrophoresis tank. Workpiece model codes are obtained via barcode scanning or RFID identification and used as classification features in the model to differentiate the impact of different vehicle frame geometries on electric field distribution and deposition efficiency. Conductivity sensors are installed in the tank's circulation loop to monitor changes in conductivity. All sensor data is aggregated to a data acquisition gateway via Modbus TCP and synchronously transmitted to the central industrial control computer after being timestamped, ensuring that each set of data strictly corresponds to the same process state at the same moment.

[0079] In this embodiment, the data preprocessing includes:

[0080] Outlier data points exceeding the physically reasonable range are removed using the 3σ criterion, and missing data caused by communication interruptions or sensor failures are filled using linear interpolation. After outlier removal and missing data filling, a moving average filtering algorithm is used to smooth the time-series data. The moving average filtering algorithm involves taking the arithmetic mean of continuous sampled values ​​within a preset window length W, where W ranges from 5 to 10. The filtered value after the moving average filtering is expressed as:

[0081]

[0082] In the formula, For a moment The original sampled values ​​of the specified electrophoresis process parameters are used. For a moment The corresponding filter value for the process parameters. This is the size of the sliding window.

[0083] The feature engineering includes:

[0084] Based on the collected basic features of electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time, six derived features are constructed: electrophoresis voltage change rate, current density, bath temperature change rate, solids content change rate, cumulative electrochemical effect and bath conductivity, resulting in an 11-dimensional feature vector.

[0085] The cumulative electrochemical action Using approximate formulas Calculation, where The electrophoresis voltage is... For electrophoresis current, The electrophoresis time is used; the Min-Max normalization method is used to map all features to the [0,1] interval, and the normalization formula is:

[0086]

[0087] In the formula, These are the normalized eigenvalues. This represents the original characteristic value of a certain electrophoresis process parameter. and These are the minimum and maximum values ​​of the electrophoresis process parameter in the training dataset, respectively.

[0088] See Figure 4The raw data collected by the sensors contains noise, outliers, and missing values, and must be preprocessed before it can be used for model training and prediction. Outlier data points exceeding the physically reasonable range are removed using the 3σ criterion, with the feature mean as the center and three times the standard deviation as the boundary; any sampling point exceeding this range is considered an anomaly and removed. For missing data caused by communication interruptions or momentary sensor failures, linear interpolation is used to fill in the gaps to maintain the continuity of the data sequence. After anomaly removal and missing data filling, the original signal still contains high-frequency random noise. To effectively filter out noise and retain the process change trend, a moving average filtering algorithm is used to smooth the time-series signal. This filtering method can effectively attenuate high-frequency noise components while maintaining the integrity of low-frequency process trend signals.

[0089] While the basic features obtained from the initial data acquisition (voltage, current, temperature, solid content, and time) contain the most fundamental process information, they are insufficient to fully reflect the dynamic trends and cumulative effects of parameter changes. Therefore, this scheme constructs the following six derived features based on the basic features:

[0090] (1) The rate of change of voltage over time reflects the stability and regulation strength of voltage regulation;

[0091] (2) Current density, defined as current divided by the effective surface area of ​​the workpiece, is used to eliminate the influence of frame size differences on the absolute current value and to reflect the electric field deposition intensity in a more physical sense.

[0092] (3) Temperature change rate, which characterizes the dynamic trend of the bath temperature;

[0093] (4) The rate of change of solids content characterizes the time-varying properties of the bath solution composition;

[0094] (5) Cumulative electrochemical action, using an approximate formula Calculation (dimensional analysis:) The unit is volts. The unit is ampere. The unit is seconds, and the product unit is volt-ampere-second (joule), which represents the cumulative amount of electrochemical energy and is positively correlated with the total amount of electrodeposition.

[0095] (6) The conductivity of the bath solution reflects the changes in ion concentration and migration ability. The above six derived features, together with the five original features, constitute an 11-dimensional feature vector, which serves as the input variable for the prediction model.

[0096] Because the dimensions and orders of magnitude of the features differ significantly (e.g., voltage is on the order of hundreds of volts, while solid components are only a few percent), without scaling, the larger-order features will dominate the direction of the model gradient update, leading to convergence difficulties. Therefore, the Min-Max normalization method is used to map all features to the [0,1] interval. After normalization, the 11-dimensional features are on the same order of magnitude, providing numerically favorable input data for model training.

[0097] In this embodiment, the gradient boosting tree prediction model is composed of a weighted sum of multiple regression trees, and is represented as follows:

[0098]

[0099] In the formula, The predicted value of paint film thickness output by the model. For the first Regression trees for input feature vectors The output paint film thickness prediction component, The total number of regression trees; the gradient boosting tree prediction model is constructed through an iterative process, with the output of the new tree in each round fitting the negative gradient of the previous round's model, and the total number of regression trees. This was determined through cross-validation.

[0100] The gradient boosting tree prediction model is trained based on minimizing an objective function, which includes a mean squared error loss term and a regularization penalty term.

[0101]

[0102] In the formula, For the first Measured values ​​of paint film thickness for each sample. For the first Predicted film thickness for each sample The total number of training samples; the regularization penalty term for:

[0103]

[0104] In the formula, For the first The number of leaf nodes in the regression tree. For the first Tree No. The output weights of the leaf nodes The regularization hyperparameter that penalizes the number of leaf nodes. The L2 regularization hyperparameter is used to determine the weights of the leaf nodes; the regularization hyperparameter is determined by Bayesian optimization or grid search methods.

[0105] See Figure 5 Specifically, a gradient boosting tree algorithm is used to construct a paint film thickness prediction model. This algorithm belongs to the Boosting framework in ensemble learning methods. It iteratively adds regression trees round by round, with each new tree built in the direction of fitting the negative gradient (i.e., residual) of the previous model. Through multiple rounds of accumulation, it gradually approaches the optimal solution of the objective function. The reason for choosing this algorithm is that it can effectively fit nonlinear relationships, has good robustness to outliers, has high training efficiency, and has a built-in feature importance ranking mechanism, which can provide good model interpretability while ensuring prediction accuracy.

[0106] Let the training dataset be... ,in The total number of training samples, For the first The 11-dimensional input feature vector of each sample (each dimension has a different dimension, which is unified as a normalized dimensionless value). This corresponds to the measured value of the paint film thickness. The prediction model is ultimately expressed as... The weighted sum of regression trees; the objective function for model training consists of two parts: the loss function. To measure the difference between predicted and measured values, this scheme uses mean squared error (MSE) as the loss function, and its expression is: Regularization term Used to control model complexity to prevent overfitting.

[0107] The model is updated iteratively in an additive manner, in the first round... During the iteration, the update formula is:

[0108]

[0109] In the formula, For the first Model predictions after rounds of iterations This is the cumulative forecast value from the previous round. For the first The output of the regression tree (whose construction goal is to fit the previous round loss function with respect to...) (negative gradient) The learning rate (dimensionless, ranging from 0.01 to 0.05) is used to control the proportion of each new tree's contribution to the final prediction.

[0110] At the model training implementation level, the XGBoost or LightGBM open-source frameworks are used for engineering implementation. Hyperparameter optimization employs Bayesian optimization methods, aiming to minimize the root mean square error of 5-fold cross-validation to search for the optimal parameter combination. The search range for the main hyperparameter is set as follows: learning rate... The maximum depth of the tree (Dimensionless), Subsampling Ratio (Dimensionless), Number of regression trees (Dimensionless). During training, the early stopping rounds are set to 50 rounds, meaning that training is terminated early if the validation set error does not improve within 50 rounds, in order to save computational resources and prevent overfitting.

[0111] After training, based on the feature importance ranking results built into the model using split gain accumulation, redundant features with importance scores below 0.01 are removed, thereby simplifying the model structure and reducing the computational load during online inference. The trained and performance-validated prediction model is then deployed to the industrial control computer on the production floor. During each data acquisition cycle, the system inputs the latest acquired, preprocessed, and feature-constructed 11-dimensional feature vector into the trained model. After calculation, the model outputs a predicted value for the workpiece paint film thickness under the current process conditions, denoted as . The prediction output is synchronized with the data acquisition cycle, enabling continuous online prediction of paint film thickness.

[0112] In this embodiment, the prediction deviation The difference between the predicted paint film thickness and the target thickness is expressed as:

[0113]

[0114] In the formula, This is the predicted coating thickness of the electrocoated batch of vehicle frames at the current moment. The target thickness value set for the process specifications;

[0115] When the absolute value of the prediction deviation is not greater than the preset deviation tolerance At this time, the current process parameters remain unchanged;

[0116] When the predicted deviation is greater than the deviation tolerance and the predicted coating thickness is greater than the target thickness, reduce the electrophoresis voltage or shorten the electrophoresis time.

[0117] When the predicted deviation is less than the negative value of the deviation tolerance and the predicted value of the coating thickness is less than the target thickness value, the electrophoresis voltage is increased or the electrophoresis time is extended.

[0118] Specifically, the ultimate value of the predicted values ​​lies in guiding the real-time adjustment of process parameters. The system outputs the predicted thickness from the model. Target thickness as set by process specifications Perform a comparison. Assume the allowable deviation tolerance for the process is... The current prediction bias Defined as Based on the deviation calculation results, the system executes the following control decision logic:

[0119] (1) If If the prediction deviation is within the allowable tolerance range, then the current process state is considered to meet the quality requirements, and all process parameters are kept unchanged.

[0120] (2) If and If the coating thickness exceeds the target value and is too thick, the system determines that the deposition is excessive and implements a thinning adjustment strategy. Specific measures include: reducing the electrophoresis voltage (with a single adjustment range of 2 to 10 volts), shortening the electrophoresis time (with a single adjustment range of 5 to 30 seconds), or adjusting other parameters according to actual process requirements. The specific adjustment range is determined by the magnitude of the deviation and the absolute value of the current process parameters through a built-in adjustment algorithm.

[0121] (3) If and If the coating thickness is lower than the target value and too thin, the system determines that the deposition amount is insufficient and implements a thickness adjustment strategy. Specific measures include: increasing the electrophoresis voltage (with a single adjustment range of 2 to 10 volts), or extending the electrophoresis time (with a single adjustment range of 5 to 30 seconds), or adjusting other parameters according to actual process requirements.

[0122] In this embodiment, the PID controller calculates the adjustment amount based on the weighted sum of the proportional, integral, and derivative terms of the control error, and the output of the PID controller... Represented as:

[0123]

[0124] In the formula, This represents the current error in controlling the paint film thickness of the electrophoretic batch of vehicle frames. This is the proportionality coefficient. The integral coefficient is... The differential coefficients are: the proportional term responds to the current error, the integral term eliminates the steady-state error, and the differential term predicts the error change trend to suppress overshoot; the actuator includes a rectifier, a variable frequency pump, and a heat exchange system, which are used to adjust the electrophoresis voltage, electrophoresis time, and bath temperature, respectively.

[0125] Specifically, to eliminate the adverse effects of parameter abrupt changes on the appearance quality of the paint film, the output of the adjustment quantity is smoothed using a PID controller. Let... The control error at time is The output of the PID controller (That is, the adjustment command applied to the actuator, the dimension of which is determined by the specific object being adjusted, such as volts when adjusting voltage) is expressed as:

[0126]

[0127] In the formula, This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients, each tuned according to the actual process response characteristics. The proportional term reflects the immediate impact of the current deviation, the integral term eliminates the cumulative steady-state error of the system, and the differential term introduces predictive information about the rate of change of deviation to suppress overshoot. The output signal of the PID controller is converted into corresponding instructions by the PLC controller, which drive the rectifier (controlling voltage and current), the variable frequency pump (controlling electrophoresis time), or the heat exchange system (controlling the bath temperature) to adjust the process parameters.

[0128] This embodiment also includes an online self-learning update step for the model:

[0129] After each batch of electrophoresis production is completed, the measured values ​​of the coating thickness obtained from random sampling and the corresponding batch of process parameter data are added to the training dataset; every predetermined number of batches, incremental learning updates are performed on the current model using the new samples; when the prediction error continues to exceed the preset threshold in several consecutive batches, the model is triggered to be fully retrained.

[0130] See Figure 6 Specifically, during the actual operation of an electrophoresis production line, the properties of the bath solution will slowly age due to long-term use, and factors such as differences in coating formulations between batches and seasonal fluctuations in ambient temperature will also cause drift in process conditions. If the model remains fixed after deployment, its prediction accuracy will gradually decrease as the process conditions drift. To address this issue, this invention designs an online self-learning update mechanism.

[0131] After each batch of electrophoresis production is completed, operators randomly inspect the workpieces according to the standard quality control procedures and obtain the measured values ​​of the coating thickness. These measured values, along with the corresponding batch's process parameter data, are automatically collected by the system and added to the training dataset. Every predetermined number of batches, the system performs incremental learning updates on the current model using the new samples. That is, based on the original model structure, the system fine-tunes the leaf node weights and tree structure using the new data, allowing the model to gradually adapt to the current process conditions. When the system detects that the prediction error continuously exceeds a preset threshold (e.g., root mean square error continuously greater than 2 micrometers) for several consecutive batches, it automatically triggers a full model retraining, retraining the model using all historically accumulated valid samples to ensure that the prediction accuracy is restored to the ideal level.

[0132] To verify the practical effectiveness of the technical solution described in this invention, a comparative experiment was conducted on a cathodic electrophoretic coating production line. The experiment was conducted in two groups: the control group used traditional manual experience-based adjustment, while the experimental group used the predictive control system described in this invention. This production line mainly produces chassis for SUVs and pickup trucks, with an annual capacity of approximately 150,000 sets. The electrophoresis tank has a volume of 120 cubic meters, and the rectifier outputs a rated voltage of 350 volts and a rated current of 3000 amperes.

[0133] I. Test Conditions

[0134] The experiment was conducted on the same production line under the same environmental conditions. The electrophoresis tank had a volume of 120 cubic meters, the rectifier had a rated voltage of 350 volts and a rated current of 3000 amperes, the tank temperature was controlled between 28 and 32 degrees Celsius, the solids content was controlled between 18% and 22%, and the electrophoresis time was 120 seconds. The data acquisition frequency was 5 Hz. A total of 400 batches were tested, with 200 batches in the control group and 200 batches in the experimental group, each batch containing approximately 20 chassis.

[0135] After each batch of electrophoresis is completed, three frames are randomly selected from the conveyor chain. Following GB / T13452.2 standard, three measurement points are taken at each of the five locations: the left longitudinal beam, right longitudinal beam, front crossbeam, rear crossbeam, and middle crossbeam, for a total of 15 measurement points. A magnetic thickness gauge is used to measure the paint film thickness, and the arithmetic mean is taken as the paint film thickness value for that batch. The target value for the electrophoretic paint film thickness is set at 22 micrometers, with a process tolerance of ±2 micrometers. Products exceeding this range are considered unqualified.

[0136] II. Test Plan

[0137] The control group used a traditional production method, where operators manually adjusted process parameters such as electrophoresis voltage and time based on their personal experience. Operators checked laboratory test results every two hours, and when the test results showed that the film thickness exceeded the acceptable range, they adjusted the electrophoresis voltage based on experience, with each adjustment ranging from 5 to 10 volts.

[0138] The experimental group employed the predictive control system described in this invention. For data acquisition, a DC voltage sensor, a Hall current sensor, a Pt100 temperature sensor, and an online solid-state analyzer were deployed, with a sampling frequency of 5 Hz. The prediction model used the GBDT algorithm, with an 11-dimensional feature vector as input. The model was trained on 20,000 historical samples, achieving a root mean square error of 1.38 micrometers. For closed-loop control, a target thickness of 22 micrometers and a deviation tolerance of 1.5 micrometers were set, using a PID controller for smooth adjustment. For model self-learning, incremental learning updates were triggered every 50 batches, and full retraining was triggered when the prediction error consistently exceeded 2 micrometers.

[0139] III. Comparison of Coating Film Thickness Consistency

[0140] Table 1. Comparison Statistics of Coating Film Thickness Consistency

[0141]

[0142] Depend on Figure 7 As shown in Table 1, the standard deviation of the paint film thickness in the experimental group decreased from 2.6 μm to 0.9 μm, a reduction of 65%, and the coefficient of variation decreased from 11.7% to 4.1%, also a reduction of 65%. The range narrowed from 12.3 μm to 4.5 μm, a reduction of over 60%. This indicates that the present invention, by controlling the paint film thickness to remain stable within the allowable tolerance range near the target value, fundamentally solves the problem of large fluctuations in paint film thickness in traditional processes, and significantly improves the consistency of paint film thickness.

[0143] IV. Comparison of paint film thickness pass rate and rework rate

[0144] Table 2 Comparison of Coating Film Thickness Pass Rate and Rework Rate

[0145]

[0146] from Figure 8 As shown in Table 2, the rework rate in the control group was 12.0%, while the rework rate in the experimental group was 2.0%, a decrease of approximately 83%. The pass rate in the control group was only 88.0%, meaning that on average, one batch of defective products was produced every 8.3 batches. In contrast, the pass rate in the experimental group was as high as 98.0%, meaning that only one batch of defective products was produced every 50 batches. Only 4 out of 200 batches in the experimental group were defective, mainly concentrated in the initial stage of production line startup when the model's prediction accuracy was not yet stable. After the model underwent self-learning and updates, the pass rate of subsequent batches approached 100%, indicating that this invention effectively replaces the traditional post-production inspection and rework mode through pre-production prediction and process control, significantly reducing the rework rate and production costs.

[0147] V. Comparison of process parameter adjustment frequency and amplitude

[0148] Table 3 Comparison of process parameter adjustment frequency and amplitude

[0149]

[0150] Depend on Figure 9As shown in Table 3, the control group, due to manual delayed adjustments, averaged approximately 35 adjustments per 100 batches, with an average single voltage adjustment amplitude of 8.2 volts and a total adjustment amount of 287 volts. Furthermore, these "emergency" adjustments were often made only when the thickness had already deviated significantly. This abrupt and large-amplitude adjustment method itself can cause defects in the paint film appearance, creating a vicious cycle. In contrast, the experimental group, with smooth adjustment using a PID controller, only required approximately 12 adjustments per 100 batches, with an average single adjustment amplitude of 3.5 volts and a total adjustment amount of only 42 volts. The experimental group's low adjustment frequency and small amplitude, along with the smooth and continuous adjustment process, effectively avoided the impact of parameter abrupt changes on the paint film appearance quality and fully ensured the continuous stability of the process.

[0151] VI. Model Performance Comparison Test

[0152] 6.1 Comparison with BP Neural Network Model

[0153] Table 4 Comparison with BP neural network model

[0154]

[0155] Depend on Figure 10 As shown in Table 4, the root mean square error of the prediction model (GBDT model) of this invention on the test set is 1.38 micrometers, and the coefficient of determination R0 is... 2 The coefficient of determination (COP) was 0.95, which is superior to the BP neural network's 2.41 micrometers and 0.83. Prediction accuracy improved by approximately 43%, and the COP increased from 0.83 to 0.95, indicating that the GBDT model's ability to fit multi-factor nonlinear coupling is far stronger than that of neural networks. In terms of training efficiency, the GBDT training time is about one-fifth that of the BP network; in terms of inference speed, a single prediction requires only 3 milliseconds, about 40% of that of the BP network. More importantly, the GBDT model can output a ranking of feature importance, providing good interpretability. Process engineers can intuitively determine the contribution of each electrophoresis parameter to the coating thickness, thereby optimizing the production process accordingly.

[0156] 6.2 Comparison with multi-parameter linear regression model

[0157] Table 5 Comparison with the multi-parameter linear regression model

[0158]

[0159] from Figure 10 As shown in Table 5, the root mean square error of the linear regression model is as high as 3.82 micrometers, R0 2With a mean value of only 0.56, meaning the linear regression model can only explain 56% of the variation in paint film thickness, and a maximum absolute error of 7.5 micrometers, it is far from meeting the accuracy requirements of industrial control. This clearly demonstrates that the nonlinear coupling relationship between process parameters such as electrophoresis voltage, current, bath temperature, solids content, and electrophoresis time is extremely significant, and traditional linear models cannot accurately describe it. However, the prediction model of this invention can effectively capture the complex nonlinear relationship between multiple factors by iteratively fitting the residuals.

[0160] VII. Multi-model mixed-line production test

[0161] 7.1 Statistics on paint film thickness for various vehicle models

[0162] Table 6: Statistics on paint film thickness for each vehicle model

[0163]

[0164] Depend on Figure 11 As shown in Table 6, a total of 240 batches of paint film thickness were tested under mixed-line production conditions for the three models. The coefficient of variation of paint film thickness was controlled at around 4.0%, and the overall rework rate was 1.3%. The deviation between the average thickness and the target thickness of each model was within 0.2 micrometers, and the standard deviation was less than 1.0 micrometer. This indicates that after incorporating the chassis model code as a classification feature into the model input, the prediction model can effectively learn the differences in electric field distribution and deposition efficiency caused by the chassis geometry of different models, and achieve high-precision adaptive prediction and control under mixed-line conditions for multiple models.

[0165] 7.2 Vehicle model switching response time

[0166] 7.2 Vehicle model switching response time

[0167]

[0168] As shown in Table 7, the system automatically loads the corresponding prediction model and control parameters when switching vehicle models on the production line. The switching response time is less than 1 second, and the average response time for the three vehicle models is approximately 0.8 seconds. This response speed fully meets the fast-paced switching requirements of the production line, and the entire process requires no manual intervention, significantly improving the automation level and efficiency of mixed-line production.

[0169] VIII. Validation of Model Self-Learning Effect

[0170] 8.1 Adaptability to Drift under Operating Conditions

[0171] Table 8 Results of Drift Adaptability Test under Operating Conditions

[0172]

[0173] Depend on Figure 12As shown in Table 8, in the 100th batch of the experiment, the same specification electrophoretic paint from the paint supplier was artificially replaced (the solids content control range was adjusted from 18%~20% to 19%~21%). After the paint replacement, due to the change in the bath properties, the model prediction error rapidly increased from 1.38 micrometers to 2.63 micrometers, an increase of 90%. After the incremental learning update was triggered in the 111th batch, the error gradually decreased to 1.92 micrometers; after the second incremental learning, the error further decreased to 1.45 micrometers; after the third incremental learning, the error recovered to 1.40 micrometers, which is close to the level before the paint replacement. This experiment fully demonstrates that the model can quickly adapt to new process conditions through online self-learning after the occurrence of operating condition drift, and has strong adaptive capabilities.

[0174] 8.2 Incremental learning versus full retraining effects

[0175] Table 9. Effects of Incremental Learning and Full Retraining

[0176]

[0177] As shown in Table 9, incremental learning updates were triggered four times and full retraining was triggered once during the experiment. Two of these were triggered according to a predetermined cycle, in batches 51 and 101, respectively; the other two were triggered due to changes in operating conditions, in batches 111 and 151, respectively. Full retraining, triggered at the end of the quarter, further optimized the error from 1.40 micrometers to 1.30 micrometers. Incremental learning can quickly respond to periodic fluctuations and sudden changes in operating conditions, while full retraining systematically updates the model structure. The two work together to achieve continuous model optimization and stable long-term prediction accuracy.

[0178] IX. System Response Speed ​​Verification

[0179] Table 10 Response Time of Each Component of the System

[0180]

[0181] Table 10 lists the response times of each component of the system. The entire system takes approximately 1 second from sensor data acquisition to actuator adjustment, meeting the real-time requirements of the electrophoresis production line. The sensor acquisition cycle is 0.2 seconds (5 Hz sampling frequency), data preprocessing and feature construction takes only 0.05 seconds, model inference takes only 0.003 seconds, PID calculation takes 0.002 seconds, PLC command issuance takes 0.02 seconds, and actuator response takes 0.5 to 1.0 seconds. This latter part accounts for the majority of the total response time. The PID controller in the experimental group was tuned with a proportional gain of 2.0, an integral gain of 0.1, and a derivative gain of 0.5. During the experiment, the voltage regulation curve was smooth and continuous, without any parameter abrupt changes or oscillations. All adjustments were performed within acceptable process ranges, verifying the stability of the control strategy.

[0182] 10. Long-term operational stability verification

[0183] Table 11 Long-term operational stability statistics

[0184]

[0185] As shown in Table 11, the system ran continuously for 6 months, covering approximately 1500 batches. The root mean square error of the prediction accuracy fluctuated between 1.36 and 1.45 micrometers each month, the paint film thickness pass rate remained stable between 98.0% and 99.2%, and the rework rate was consistently kept below 2.0%. In the fifth month, due to changes in summer ambient temperature and aging of the bath solution, the model error slightly increased to 1.45 micrometers. The system responded and adjusted promptly through incremental learning, and by the sixth month, it had returned to the normal level of 1.39 micrometers. Throughout the 6-month verification period, the model prediction accuracy and paint film thickness pass rate remained stable, fully verifying the engineering reliability and stability of this invention under long-term industrial operating conditions.

[0186] In summary, this experiment verifies that the technical solution proposed in this invention is superior to existing technologies in the following aspects: Regarding film thickness consistency, the coefficient of variation is reduced from 11.7% to 4.1%; regarding rework rate, it is reduced from 12.0% to 2.0%; regarding prediction accuracy, the root mean square error is 1.38 micrometers, and the coefficient of determination R² is 0.95, which is superior to BP neural networks and linear regression models; regarding prediction speed, a single inference takes only 3 milliseconds; regarding model interpretability, it can output feature importance ranking, facilitating understanding and optimization by process engineers; regarding multi-vehicle adaptation, the coefficient of variation for all three vehicle types is controlled at around 4.0%, and the switching response time is less than 1 second; regarding model self-learning, accuracy can be restored after 2 to 3 incremental learning iterations following operating condition drift; regarding system response, fully automatic closed-loop control is completed within 1 second; regarding long-term operation, accuracy and pass rate remain stable during 6 months of operation. The above experimental results demonstrate that this invention has significant technical advancements and practical value.

[0187] Example 2

[0188] See Figure 13 Embodiment 2 of the present invention also provides a paint film thickness prediction and control system for vehicle frame electrophoresis process, employing a paint film thickness prediction and control method for vehicle frame electrophoresis process described in the above embodiments, comprising:

[0189] The data acquisition module 100 is used to collect electrophoresis process parameter data in real time through sensors deployed on the electrophoresis production line. The electrophoresis process parameter data includes at least electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time.

[0190] The data preprocessing and feature construction module 200 is used to perform data preprocessing and feature engineering on the collected electrophoresis process parameter data to obtain a normalized multidimensional feature vector.

[0191] The paint film thickness prediction module 300 is used to input the multidimensional feature vector into a pre-trained gradient boosting tree prediction model. The gradient boosting tree prediction model is trained with electrophoresis process parameter data as input features and measured paint film thickness as labels, and outputs the predicted paint film thickness of the current batch of frames.

[0192] The deviation calculation and control decision module 400 is used to compare the predicted value of the paint film thickness with the target thickness value set by the process specification, calculate the prediction deviation, and determine the adjustment strategy of the process parameters based on the magnitude and direction of the prediction deviation.

[0193] The PID control and execution module 500 is used to generate a smooth process parameter adjustment amount through the PID controller according to the control strategy, and drive the actuator through the PLC controller to adjust the electrophoresis process parameters in real time so that the coating thickness is kept within the target range.

[0194] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0195] Example 3

[0196] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a method of predicting and controlling the coating thickness of a vehicle frame electrophoresis process. The program code includes instructions for executing the method of predicting and controlling the coating thickness of a vehicle frame electrophoresis process according to Embodiment 1 or any possible implementation thereof.

[0197] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0198] Example 4

[0199] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0200] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute the method for predicting and controlling the coating thickness of a vehicle frame electrophoresis process by calling the program instructions.

[0201] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0202] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0203] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0204] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames, characterized in that, Includes the following steps: Sensors deployed on the electrophoresis production line are used to collect electrophoresis process parameter data in real time. The electrophoresis process parameter data includes at least electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time. The collected electrophoresis process parameter data are preprocessed and feature-engineered to obtain normalized multidimensional feature vectors. The multidimensional feature vector is input into a pre-trained gradient boosting tree prediction model. The gradient boosting tree prediction model is trained with electrophoresis process parameter data as input features and measured paint film thickness as labels, and outputs the predicted paint film thickness of the current batch of frames. The predicted film thickness is compared with the target thickness value set in the process specification, the prediction deviation is calculated, and the adjustment strategy of the process parameters is determined based on the magnitude and direction of the prediction deviation. According to the aforementioned adjustment strategy, a smooth process parameter adjustment amount is generated by the PID controller, and the PLC controller drives the actuator to adjust the electrophoresis process parameters in real time, so that the coating thickness is kept within the target range.

2. The method for predicting and controlling the coating thickness in a vehicle frame electrophoresis process according to claim 1, characterized in that, The sensors include a voltage sensor, a current sensor, a temperature sensor, a solids analyzer, and a timer, used to collect the electrophoresis voltage, the electrophoresis current, the bath temperature, the solids content, and the electrophoresis time. The sensors also include a conductivity sensor and a workpiece model identification device. The electrophoresis process parameter data also include the bath conductivity and the workpiece model code. All sensor data are aggregated to the data acquisition gateway via the industrial Ethernet protocol and synchronously transmitted to the industrial control computer after being timestamped.

3. The method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames according to claim 1, characterized in that, The data preprocessing includes: Outlier data points exceeding the physically reasonable range are removed using the 3σ criterion, and missing data caused by communication interruptions or sensor failures are filled using linear interpolation. After outlier removal and missing data filling, a moving average filtering algorithm is used to smooth the time-series data. The moving average filtering algorithm involves taking the arithmetic mean of continuous sampled values ​​within a preset window length W, where W ranges from 5 to 10. The filtered value after the moving average filtering is expressed as: , In the formula, For a moment The original sampled values ​​of the specified electrophoresis process parameters are used. For a moment The corresponding filter value for the process parameters. This is the size of the sliding window.

4. The method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames according to claim 1, characterized in that, The feature engineering includes: Based on the collected basic features of electrophoresis voltage, electrophoresis current, bath temperature, solids content and electrophoresis time, six derived features are constructed: electrophoresis voltage change rate, current density, bath temperature change rate, solids content change rate, cumulative electrochemical effect and bath conductivity, resulting in an 11-dimensional feature vector. The cumulative electrochemical action Using approximate formulas Calculation, where The electrophoresis voltage is... For electrophoresis current, The electrophoresis time is used; the Min-Max normalization method is used to map all features to the [0,1] interval, and the normalization formula is: , In the formula, These are the normalized eigenvalues. This represents the original characteristic value of a certain electrophoresis process parameter. and These are the minimum and maximum values ​​of the electrophoresis process parameter in the training dataset, respectively.

5. The method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames according to claim 1, characterized in that, The gradient boosting tree prediction model is composed of a weighted sum of multiple regression trees, expressed as: , In the formula, The predicted value of paint film thickness output by the model. For the first Regression trees for input feature vectors The output paint film thickness prediction component, The total number of regression trees; the gradient boosting tree prediction model is constructed through an iterative process, with the output of the new tree in each round fitting the negative gradient of the previous round's model, and the total number of regression trees. This was determined through cross-validation.

6. The method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames according to claim 5, characterized in that, The gradient boosting tree prediction model is trained based on minimizing the objective function, which includes a mean squared error loss term and a regularization penalty term. , In the formula, For the first Measured values ​​of paint film thickness for each sample. For the first Predicted film thickness for each sample The total number of training samples; the regularization penalty term for: , In the formula, For the first The number of leaf nodes in the regression tree. For the first Tree No. The output weights of the leaf nodes The regularization hyperparameter that penalizes the number of leaf nodes. The L2 regularization hyperparameter is used to determine the weights of the leaf nodes; the regularization hyperparameter is determined by Bayesian optimization or grid search methods.

7. The method for predicting and controlling the coating thickness in a vehicle frame electrophoresis process according to claim 1, characterized in that, The prediction bias The difference between the predicted paint film thickness and the target thickness is expressed as: , In the formula, This is the predicted coating thickness of the electrocoated batch of vehicle frames at the current moment. The target thickness value set for the process specifications; When the absolute value of the prediction deviation is not greater than the preset deviation tolerance At this time, the current process parameters remain unchanged; When the predicted deviation is greater than the deviation tolerance and the predicted coating thickness is greater than the target thickness, reduce the electrophoresis voltage or shorten the electrophoresis time. When the predicted deviation is less than the negative value of the deviation tolerance and the predicted value of the coating thickness is less than the target thickness value, the electrophoresis voltage is increased or the electrophoresis time is extended.

8. The method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames according to claim 7, characterized in that, The PID controller calculates the adjustment amount based on the weighted sum of the proportional, integral, and derivative terms of the control error, and the output of the PID controller is... Represented as: , In the formula, This represents the current error in controlling the paint film thickness of the electrophoretic batch of vehicle frames. This is the proportionality coefficient. The integral coefficient is... The differential coefficients are: the proportional term responds to the current error, the integral term eliminates the steady-state error, and the differential term predicts the error change trend to suppress overshoot; the actuator includes a rectifier, a variable frequency pump, and a heat exchange system, which are used to adjust the electrophoresis voltage, electrophoresis time, and bath temperature, respectively.

9. The method for predicting and controlling the coating thickness in an electrophoretic coating process for vehicle frames according to claim 1, characterized in that, It also includes the online self-learning update step of the model: After each batch of electrophoresis production is completed, the measured values ​​of the coating thickness obtained from random sampling and the corresponding batch of process parameter data are added to the training dataset; every predetermined number of batches, incremental learning updates are performed on the current model using the new samples; when the prediction error continues to exceed the preset threshold in several consecutive batches, the model is triggered to be fully retrained.

10. A paint film thickness prediction and control system for vehicle frame electrophoresis, employing the paint film thickness prediction and control method for vehicle frame electrophoresis as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect electrophoresis process parameter data in real time through sensors deployed on the electrophoresis production line. The electrophoresis process parameter data includes at least electrophoresis voltage, electrophoresis current, tank temperature, solids content, and electrophoresis time. The data preprocessing and feature construction module is used to perform data preprocessing and feature engineering on the collected electrophoresis process parameter data to obtain normalized multidimensional feature vectors. The paint film thickness prediction module is used to input the multi-dimensional feature vector into a pre-trained gradient boosting tree prediction model. The gradient boosting tree prediction model is trained with electrophoresis process parameter data as input features and measured paint film thickness as labels, and outputs the predicted paint film thickness of the current batch of frames. The deviation calculation and control decision module is used to compare the predicted value of the paint film thickness with the target thickness value set by the process specification, calculate the prediction deviation, and determine the adjustment strategy of the process parameters based on the magnitude and direction of the prediction deviation. The PID control and execution module is used to generate smooth process parameter adjustment amounts through the PID controller according to the control strategy, and drive the actuator through the PLC controller to adjust the electrophoresis process parameters in real time so that the coating thickness is maintained within the target range.