Plastic dipping coating thickness control method based on multi-dimensional environment parameter real-time compensation

Abstract

The invention relates to a plastic dipping coating thickness control method based on multi-dimensional environmental parameter real-time compensation, and aims to solve the problems that the coating thickness is easily influenced by environmental fluctuation and the control precision is low in a traditional plastic dipping process. The method comprises the following steps: acquiring environment temperature, relative humidity, viscosity of plastic dipping liquid and surface temperature of a workpiece in real time through a plurality of groups of sensors, carrying out standardization processing on data, and constructing an environment parameter data set with time sequence characteristics; based on the data set, establishing a nonlinear coupling prediction model of the coating thickness and the environmental factors, and generating a prediction coefficient matrix for real-time calculation; the matrix and the real-time environment parameters are used for calculating the thickness of a target coating and the change trend of the target coating, a thickness deviation compensation signal is generated, the workpiece preheating temperature, the plastic dipping heat preservation time and the lifting speed are adjusted in a linkage mode according to the thickness deviation compensation signal, and an optimized technological parameter scheme is formed; the plastic dipping process is controlled by executing the scheme, a feedback result is used for model updating, and accurate control and continuous optimization of the coating thickness are achieved.

Description

Technical Field

[0001] The present invention relates to the field of control technology, and in particular to a method for controlling the thickness of a plastic dip coating based on real-time compensation of multi-dimensional environmental parameters. Background Art

[0002] In the existing dipping process, the workpiece is usually immersed in the thermoplastic dipping liquid after preheating, and lifted after a specific time to form a uniform plastic coating. In order to achieve a relatively stable coating thickness, some production lines have been equipped with temperature control devices and timing systems to control the preheating temperature, immersion time and lifting speed of the workpiece, and monitor the ambient temperature and liquid viscosity manually or through simple sensors to assist in operation adjustments.

[0003] However, since coating thickness is highly sensitive to multiple variables such as ambient temperature and humidity, viscosity of the dipping liquid, and surface temperature of the workpiece, existing methods mostly rely on empirical adjustments or static parameter configurations, lacking the ability to respond to real-time changing factors, resulting in large fluctuations in coating thickness and difficulty in maintaining consistency in multi-batch and multi-period production. In addition, existing control methods generally lack feedback mechanisms and cannot adaptively optimize subsequent processes based on actual coating effects.

[0004] In view of the above problems, there is an urgent need for a new dipping control method that can respond to multi-dimensional environmental changes in real time and realize dynamic regulation of coating thickness. Summary of the invention

[0005] The present application provides a method for controlling the thickness of a dip coating based on real-time compensation of multi-dimensional environmental parameters to improve the control accuracy of the thickness of the dip coating.

[0006] The present application provides a method for controlling the thickness of a dip coating based on real-time compensation of multi-dimensional environmental parameters, comprising: The ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece in the dipping environment are collected through multiple sets of sensors, and the collected data are standardized to construct an environmental parameter data set containing time series characteristics; Based on the environmental parameter data set, a nonlinear coupling prediction model between coating thickness and multidimensional environmental factors is constructed, and a coating thickness prediction coefficient matrix for real-time reasoning is generated during the training phase; Using the current real-time environmental parameters as input, combined with the coating thickness prediction coefficient matrix, dynamically calculate the target coating thickness range and its change trend of the workpiece under the current conditions, and generate a thickness deviation compensation signal including the deviation direction and degree; Adjust the dipping control parameters including the workpiece preheating temperature, dipping insulation time and lifting speed according to the thickness deviation compensation signal to form a process parameter optimization plan; The process parameter optimization scheme is used to control the execution process of the dipping equipment so that the coating thickness dynamically approaches the target value, and the execution result is used as feedback input to continuously optimize the nonlinear coupling prediction model.

[0007] The beneficial effects of the technical solution provided by this application include: (1) By constructing a nonlinear coupling prediction model and compensating for thickness deviation in real time, the workpiece can still obtain a stable coating thickness under complex and changeable environmental conditions, reducing the rate of deviation and rework rate. (2) Collecting multidimensional parameters such as ambient temperature, humidity, viscosity, etc. and introducing time series characteristics, the control strategy has dynamic response capabilities and can adapt to environmental fluctuations in different batches and different time periods. (3) According to the compensation signal, the preheating temperature, dipping time and lifting speed are adjusted in a coordinated manner to coordinate the operation of each link, optimize the dipping effect as a whole, and improve the uniformity and adhesion quality of the coating. (4) The execution results are used for iterative optimization of the prediction model, which has self-learning ability, which helps to continuously improve the process parameters and improve the long-term production stability and automation level. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flow chart of a method for controlling the thickness of a dip coating based on real-time compensation of multi-dimensional environmental parameters provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0009] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0010] The first embodiment of the present application provides a method for controlling the thickness of a dip coating based on real-time compensation of multi-dimensional environmental parameters. Figure 1 , which is a schematic diagram of the first embodiment of the present application. Figure 1 The first embodiment of the present application provides a method for controlling the thickness of a dip coating based on real-time compensation of multi-dimensional environmental parameters, which is described in detail.

[0011] Step S101: The ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece in the dipping environment are collected through multiple sets of sensors, and the collected data are standardized to construct an environmental parameter data set containing time series characteristics.

[0012] Step S101 aims to obtain multi-dimensional environmental data that affects the thickness of the dip coating, and standardize and structure it to construct an environmental parameter data set with time series characteristics. First, multiple groups of high-precision sensors are arranged at key positions of the dip coating production line, including ambient temperature sensors, ambient humidity sensors, dip coating viscosity sensors, and workpiece surface temperature sensors. The ambient temperature sensor is preferably a thermistor or an infrared temperature sensor, which is used to collect the air temperature of the dip coating workshop or closed working area. The sampling frequency can be set to once every 5 seconds to capture fluctuations in a short period of time. The ambient humidity sensor can be a capacitive humidity sensor, and the measurement range should cover 20%RH (Relative Humidity) to 90%RH, with a resolution of not less than ±1%RH, which is used to detect the influence of water vapor content in the air on the coating curing process.

[0013] The viscosity sensor of the plastic dipping liquid can be an online rotary or ultrasonic viscometer, which should be placed in a representative area of ​​the plastic dipping liquid tank to ensure that it can reflect the effect of changes in liquid temperature or solution concentration on viscosity. It is preferred to control the sampling interval to within 10 seconds. The workpiece surface temperature sensor should be a contact thermocouple or a non-contact infrared sensor, installed at a specific position on the workpiece movement path, especially in the preheating zone and the buffer zone before immersion, to collect the workpiece surface temperature in real time to determine its heat absorption status and initial conditions.

[0014] The collected raw data must be uniformly connected to the central processing unit through the data acquisition controller and standardized. Standardization methods may include Z-score standardization, minimum-maximum normalization or exponential smoothing filtering to eliminate the dimensional differences between different physical quantities and enhance the time sensitivity of the model. In addition, in order to enhance the model's ability to perceive trends, the system constructs a time series window for each type of parameter, for example, with a time window of 60 seconds, using a sliding window mechanism for continuous update, and extracting statistical features within the window (such as mean, variance, maximum, minimum, slope, etc.) as the input feature dimension of the data set.

[0015] Finally, the system combines all standardized environmental parameters with their time series statistical characteristics to construct a structured environmental parameter dataset. This dataset is used for subsequent prediction model training and real-time reasoning input to ensure that environmental changes can be dynamically reflected during the coating thickness control process, improving the real-time and accuracy of control.

[0016] If implemented, this step can integrate functional modules such as sensor signal acquisition, circuit filtering, data integration and preprocessing through an industrial automation platform (such as PLC or industrial computer), and use the data for model calling through an embedded computing module or a cloud-edge collaborative system, thereby fully realizing the functions of this step.

[0017] Furthermore, the ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece in the dipping environment are collected by multiple sets of sensors, and the collected data are standardized to construct an environmental parameter data set containing time series characteristics, including: An integrated environmental temperature and humidity sensor, an online rotary viscometer and a non-contact infrared temperature sensor are respectively installed in the preheating section, the dipping section and the lifting section of the dipping production line. Each of the sensors is equipped with a unified timestamp synchronization module to associate the collected data with the same time axis to form a raw data stream with time consistency. Data cleaning and trend extraction are performed on each type of collected data through edge processing nodes. The trend extraction includes calculating the first-order derivative and local fluctuation variance of the parameter within a preset sliding time window, and judging whether it exceeds the set threshold. If it exceeds, the time point is marked as a disturbance point and a disturbance level label is assigned. The standardized raw data, first-order derivatives, local fluctuation variance values ​​and disturbance labels are combined to form an extended feature vector of environmental parameters, and the data window is encapsulated in a fixed period to form a time series parameter data set containing dynamic feature dimensions for subsequent thickness prediction model calls.

[0018] In this embodiment, in order to improve the real-time response capability of coating thickness control to multi-dimensional environmental changes, multiple parameters including ambient temperature, relative humidity, viscosity of the dipping liquid, and surface temperature of the workpiece need to be collected. To ensure the consistency and time series comparability of these data, the present invention optimizes the sensor layout position and data synchronization method on the dipping production line.

[0019] An integrated environmental temperature and humidity sensor is installed in the preheating section of the process flow. The sensor integrates a temperature measurement unit and a humidity measurement unit, and can synchronously obtain the air temperature and relative humidity in the preheating area to avoid measurement distortion caused by time dislocation. In the dipping section, an online rotational viscometer is installed. The viscometer measures the viscosity change of the dipping liquid in real time by sensing the rotational resistance, and can reflect the viscosity change of the dipping liquid caused by temperature fluctuations or solvent volatilization. In the lifting section, a non-contact infrared temperature sensor is installed to measure the surface temperature of the workpiece within a 0.5-second time window after the workpiece is separated from the dipping liquid, ensuring that thermal field data close to the instantaneous state of dipping can be obtained.

[0020] In order to achieve data time alignment across devices, all sensors are equipped with a unified timestamp synchronization module. This module can achieve millisecond-level data sampling time consistency based on the Network Time Protocol (NTP) or the industrial bus master clock signal. Each set of collected data is attached with a corresponding timestamp for subsequent data fusion and dynamic feature extraction.

[0021] After data collection is completed, the raw data is first preprocessed by the edge processing units distributed near each sensor node. This processing includes outlier removal, noise filtering, and missing data interpolation to improve data quality. Subsequently, the system performs trend extraction operations on each type of parameter within a sliding time window. Trend characteristics are mainly realized by calculating the first-order derivative (that is, the difference between the current value and the value at the previous moment) and the local fluctuation variance. If the derivative value or fluctuation variance at a certain moment exceeds the preset threshold, the system will mark the moment as a "disturbance point". For example, when the ambient humidity rises rapidly in a short period of time, it may affect the curing rate of the coating. At this time, it will be identified as a disturbance and assigned a level label, such as slight, moderate or severe, to characterize the degree of its possible impact on process stability.

[0022] Finally, the system fuses the above data to form a high-dimensional extended feature vector of environmental parameters. This feature vector contains not only the standardized values ​​of the original parameters, but also their first-order derivatives, local variances, and disturbance level labels. Data with this structure can fully describe the current environmental state and its changing trends. All feature vectors will be windowed at fixed time intervals (for example, every 30 seconds) to form a continuous time series parameter data set.

[0023] This time series dataset, as a model input, can significantly improve the coating thickness prediction's ability to perceive short-term disturbances, and provide a dynamic and detailed environmental basis for subsequent thickness control strategies, ensuring that the coating thickness control system is stable, adaptable, and real-time.

[0024] Step S102: Based on the environmental parameter data set, a nonlinear coupling prediction model between coating thickness and multidimensional environmental factors is constructed, and a coating thickness prediction coefficient matrix for real-time reasoning is generated during the training phase.

[0025] In step S102, the system builds a coupled prediction model that can describe the nonlinear relationship between coating thickness and environmental variables based on the environmental parameter data set constructed in step S101, and completes the training process of the model to generate a coating thickness prediction coefficient matrix that can be used in real time. To ensure that the model has good generalization ability and real-time response performance, this step needs to comprehensively consider the technical implementation details of the model structure, input feature selection, training method and model output structure.

[0026] First, in terms of model input, the data set used should include the original parameters such as the standardized ambient temperature, relative humidity, viscosity of the plastic dipping liquid, and surface temperature of the workpiece, and a high-dimensional input vector should be constructed in combination with time series characteristics, such as the change value, change rate, trend direction, local mean and standard deviation of each parameter at the last N time points, in order to capture dynamic fluctuation information in the short term. This processing method can significantly improve the model's sensitivity to rapid environmental changes.

[0027] Next, in terms of model structure selection, it is preferred to use machine learning models that support strong nonlinear mapping capabilities, such as gradient boosting decision trees (GBDT), support vector regression (SVR), artificial neural networks (ANN), or lightweight deep neural networks (such as 1-2 layer fully connected networks) when edge computing capabilities allow. The output of the model is the predicted theoretical coating thickness value under given input environmental conditions, or it can further output a coating thickness range to express the model's uncertainty assessment of the prediction accuracy.

[0028] During the training process, the labeled environmental parameters and corresponding actual coating thickness samples in the historical production batches are used as training data, and the supervised learning method is used for fitting. The training can be performed on a centralized server, using mean square error (MSE), mean absolute error (MAE), etc. as loss functions, combined with the cross-validation method to tune the model hyperparameters to ensure its prediction stability and accuracy under different environmental conditions. In order to improve the model's adaptability to edge scenarios (such as extreme humidity and viscosity fluctuations), augmented samples or ensemble learning strategies can also be used to expand the training data set.

[0029] After the model training is completed, the model parameters are solidified into a set of weight coefficients and bias values, which are saved in the form of a matrix or tensor, namely the "coating thickness prediction coefficient matrix", which can also be called the trained model. This matrix can be directly embedded in the subsequent real-time control system without retraining. It only needs to input the current real-time environmental parameters to quickly complete the coating thickness prediction in a short time, meeting the high-frequency control needs of the industrial site.

[0030] In addition, to ensure the long-term availability and accuracy of the model, this step should also include a regular retraining mechanism for the model. The system can dynamically correct the training sample set based on the feedback information in step S105 (i.e., the deviation between the actual coating thickness and the predicted value), and regularly update the model structure or parameters to maintain the adaptive ability of coating thickness control.

[0031] In summary, step S102 provides core data support for the coating thickness control strategy by establishing and training a nonlinear coupling prediction model and outputting a coating thickness prediction coefficient matrix.

[0032] Furthermore, based on the environmental parameter data set, a nonlinear coupling prediction model between coating thickness and multidimensional environmental factors is constructed, and a coating thickness prediction coefficient matrix for real-time reasoning is generated during the training phase, including: Each parameter of ambient temperature, relative humidity, viscosity of the plastic dipping liquid and workpiece surface temperature is classified according to its change rate in historical data to form two feature sets representing fast-changing parameters and slow-changing parameters respectively. The change amplitude, average trend and disturbance frequency of the parameters in each set within the preset time window are extracted to construct an extended feature vector. Based on the extended feature vector, a support vector regression model based on a combined kernel function is trained, wherein the combined kernel function simultaneously considers the local similarity and nonlinear trend relationship between parameters, and is used to establish a nonlinear mapping relationship between coating thickness and multidimensional environmental factors, and after the training is completed, a coating thickness prediction coefficient matrix for real-time prediction is output; During the model training process, a sample weighting mechanism based on the similarity of environmental characteristics is introduced for historical environmental data samples. The weighting mechanism assigns weights according to the distance between the sample and the current working condition in the multidimensional parameter space, so that historical samples that are closer to the current working condition contribute more to the model training, thereby improving the adaptability and robustness of the prediction results under actual dipping conditions.

[0033] During the implementation process, the first thing to be processed is the environmental parameter data set collected from the sensor, which usually contains a variety of variables that have a significant impact on the coating thickness during the dipping process, including ambient temperature, relative humidity, dipping liquid viscosity, and workpiece surface temperature. Since different parameters show different rates of change and influence patterns in actual working conditions, the present invention first dynamically classifies these environmental parameters. Specifically, by statistically analyzing the change amplitude and fluctuation frequency of each parameter in the historical data within a fixed time window, it is determined whether it belongs to a fast-response parameter (such as ambient temperature or workpiece surface temperature) or a relatively slow-changing parameter (such as viscosity or humidity). Based on the classification results, two feature sets are constructed respectively, and the key statistical features within a specific time window are extracted for the parameters in each set, including the maximum change amplitude, the change trend of the moving average, and the frequency of significant disturbances per unit time, and finally an extended feature vector containing multiple dimensions is generated to reflect the overall dynamic state of the environmental parameters under the current working conditions.

[0034] After the feature construction is completed, the extended feature vector is input into the support vector regression (SVR) model for training. The SVR model adopted by the present invention introduces a combined kernel function structure to enhance the expression ability of the model when dealing with complex nonlinear relationships. The combined kernel function realizes the fitting of the nonlinear coupling relationship between environmental variables and coating thickness by jointly considering the local similarity and overall trend difference between parameters. This combined kernel function not only has the high-precision advantage of the traditional radial basis kernel function in local modeling, but also combines the ability of the polynomial kernel function to grasp the global change trend, thereby constructing a prediction model that is more in line with the characteristics of the dipping process.

[0035] In order to improve the adaptability and generalization ability of the model under actual complex working conditions, the present invention also introduces a sample processing mechanism based on similarity weighting in the SVR training stage. Specifically, the system does not assign equal weights to all historical samples during the training process, but assigns weights by calculating the relative distance between each historical sample and the current real-time working condition in the multidimensional feature space. The smaller the distance, the closer the environmental conditions of the sample are to the current environment, and the higher its weight during the training process. This weighting mechanism can significantly enhance the model's predictive ability for typical or common working conditions, while avoiding fitting offsets caused by extreme samples.

[0036] After training, the SVR model will output a set of fixed prediction coefficient matrices, which can be directly embedded in the real-time control system and used to quickly infer the predicted value of coating thickness based on the currently collected environmental characteristics during the actual production process. This structure enables the model to have both real-time performance and prediction accuracy under complex disturbance conditions, which is very suitable for dip coating control scenarios. After the SVR model training is completed, the SVR model parameters are solidified into a set of weight coefficients and bias values, which are saved in the form of a matrix or tensor, which is the "coating thickness prediction coefficient matrix", which can also be called the SVR model after training.

[0037] Through the above three technical paths - dynamic classification and extended feature extraction of environmental parameters, SVR modeling based on combined kernel functions, and similarity weighted sample training mechanism, the present invention achieves high adaptability and high-precision prediction capabilities for multi-dimensional disturbance environments in the thickness control of the dipping process, effectively improving the consistency of the overall coating and product quality.

[0038] Furthermore, the combined kernel function used in the support vector regression model based on the combined kernel function is This is accomplished by: The input variables and Respectively represent any two environmental parameter vectors after preprocessing and construction of extended features, the environmental parameter vectors include the statistical features of ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece within a fixed time window, the statistical features include the average value, the first-order derivative, the disturbance frequency and the normalized value; The first kernel function For the weighted radial basis function, the following formula 1 is used: ; in, Represents feature dimension; Representation vector In the The value of the feature dimension; Indicates The disturbance weight corresponding to each parameter is calculated based on the sensitivity of the parameter to coating thickness fluctuations in historical data; is the kernel function adjustment coefficient; The second kernel function It is a non-centralized third-order polynomial kernel function, which is implemented using the following formula 2: ; in, It is a constant set empirically or an adjustable coefficient determined by cross-validation; It represents the mean vector of the working conditions obtained from the historical environmental samples, and is used to introduce the correction effect of the sample distribution center on the global trend; Combination Kernel Function Use the following formula 3: ; in, , is the weight coefficient and satisfies .

[0039] In this embodiment, in order to achieve high-precision prediction of the thickness of the dip coating in a multi-dimensional disturbance environment, a combined kernel function structure combining local similarity and global trend relationship is proposed. This structure is embedded in the support vector regression (SVR) model as the core mechanism for establishing a nonlinear mapping between input features and target thickness values. The combined kernel function is denoted as ,in and is the environmental feature vector of any two samples.

[0040] Each vector , They are all input vectors after feature extraction and standardization, and are derived from the extended processing of the original sensor data. Specifically, each environmental parameter (including ambient temperature, relative humidity, viscosity of the plastic dipping liquid, and surface temperature of the workpiece) is extracted within the set time window. The statistical features include: The average value is used to characterize the stable level of the parameter within the window; The first-order derivative is used to reflect the rate of change of the parameter; The disturbance frequency is the number of times the fluctuation amplitude of the parameter exceeds the threshold in a unit time; Normalized value, that is, linearly mapping the current parameter to the [0,1] interval based on the maximum and minimum values ​​of historical samples.

[0041] These dimensions together make up the input feature vector , whose dimensions are .

[0042] The first kernel function is a weighted radial basis kernel function (RBF), which is used to capture the similarity of input features in the local space. Its expression is: ; In this formula: , The samples are , In the The value of the feature dimension; Indicates The perturbation weight of a feature reflects the historical sensitivity of the parameter to the change of coating thickness. This value is calculated by linear regression residual analysis or mutual information entropy calculation method, and the recommended value can be set to 0.1.

[0043] is the adjustment coefficient of the kernel function, which is used to control the compression effect of high-dimensional feature differences on the kernel value. The recommended initial setting is , which can be further optimized through cross-validation.

[0044] The second kernel function is a non-centralized third-order polynomial kernel function, which is used to fit the global trend relationship between the input features and the thickness value. Its expression is: ; In this formula: for and The inner product of is used to measure the global directional similarity; is the mean vector of historical sample features, that is, the average value of each feature dimension is calculated on the training set, indicating the center of typical working conditions; For input sample The Euclidean distance from the mean of the operating condition is used to capture the thickness variation trend caused by deviation from the center of the operating condition; is an adjustable hyperparameter, where Controls the influence of the characteristic inner product term, and the recommended value is 0.1; Controls the influence of the distance term, and the recommended value is 0.5; It is an offset constant used to avoid the kernel value being too low in the case of small samples. It is recommended to set it to 1.

[0045] Finally, the combined kernel function It is a linear combination of two kernel functions: ; in: , is a non-negative weight coefficient, indicating the relative contribution of the two kernel functions in the final prediction; Both satisfy the constraints , which can be determined by reverse tuning the model's error on the validation set. The recommended initial setting is , .

[0046] The entire combined kernel function structure can not only accurately respond to the local fluctuations of the input samples under the disturbance parameters, but also identify the overall deviation trend from the historical working conditions. It is a kernel function design that is highly adapted to the needs of coating thickness prediction in multi-disturbance and strong coupling environments.

[0047] This structure selects parameters through cross-validation during the training process, constructs weights through multi-dimensional perturbation parameter historical statistics, and can be integrated into the standard SVR training process, with good engineering adaptability and implementation feasibility. A conventional machine learning framework (such as scikit-learn or libsvm) can be used to implement a custom kernel function interface based on the above kernel function to complete the combined kernel modeling described in this embodiment.

[0048] Furthermore, the specific implementation of the sample weighting mechanism based on the similarity of environmental features includes: In the support vector regression model training phase, for each historical sample in the historical sample set With the current real-time working condition sample Perform similarity evaluation, where and To pre-standardize and construct a high-dimensional environmental parameter vector after extended features; calculate and The similarity distance between The following formula 4 is used: ; in, is the inverse matrix of the feature covariance matrix calculated based on the historical sample set; It is a tuning parameter used to introduce sensitivity weighting to the dynamic disturbance factor; Indicates The change in the perturbation frequency of the dimension feature in the most recent cycle; Indicates the total number of environmental feature dimensions; For historical samples The value of the feature dimension; For the current sample The value of the feature dimension; The sample weighting factor is calculated according to the following formula 6: : ; in, is the temperature coefficient, which is used to control the degree of compression of the weight distribution due to distance; is the total number of historical samples; Weighting Factor It is used to correct the penalty term coefficient of each sample in the SVR loss function to achieve sample training contribution adjustment based on local similarity, so that the model can focus more on historical samples that are highly matched with the current working conditions in a disturbance-sensitive environment, effectively improving the accuracy and stability of coating thickness prediction in a dynamic environment.

[0049] In the method for predicting the thickness of the dip coating of the present invention, the training of the support vector regression (SVR) model depends on the correspondence between a large number of historical environmental samples and the current actual working conditions. In order to improve the adaptability of the model in a non-steady-state environment, this embodiment introduces a training mechanism based on similarity weighting, that is, the influence of the training samples is dynamically allocated according to the similarity between the historical samples and the current real-time working conditions in the high-dimensional environmental feature space. The more similar the historical samples are, the higher the weights are, and the less similar the samples are, the lower the weights are, so that the model can focus more on the data that best matches the current state during the training process.

[0050] Specifically, for each historical sample With the current real-time sample , are all represented as environmental parameter vectors after standardization and feature expansion. The environmental parameter vector contains multiple physical quantities collected by sensors, including but not limited to ambient temperature, relative humidity, viscosity of the dipping liquid, and surface temperature of the workpiece. The following features are extracted for each parameter within a specific time window: sliding average, first-order derivative (i.e. parameter change rate), disturbance frequency (the number of times the set threshold is exceeded per unit time), and the normalized current value.

[0051] The present invention is used to measure historical samples With the current sample The distance function of the similarity between them is a composite distance function , defined by the following formula: ; The first term is a statistical distance measure based on the Mahalanobis distance. Represents the item-by-item difference between historical samples and current samples in all feature dimensions. Covariance matrix is the environmental feature covariance matrix calculated based on all historical samples, which reflects the correlation between the features. In order to eliminate the interference of inconsistent dimensions or redundant dimensions between features on distance calculation, the covariance matrix needs to be inverted to obtain Compared with the Euclidean distance, the Mahalanobis distance can better reflect the effective distance structure in the high-dimensional feature space and is the preferred measurement method suitable for the complex disturbance coupling environment of this embodiment.

[0052] The second term is the perturbation-weighted Manhattan distance term, designed to respond to environmental perturbations. Indicates The change in the disturbance frequency of a feature in the most recent period (for example, the past 30 minutes or the past 3 sliding windows) is calculated as: ; in is the perturbation frequency of the feature in the current period, is the disturbance frequency of the previous cycle. The disturbance is defined as the number of times the parameter exceeds the historical fluctuation threshold. Through this item, this embodiment can reflect whether the current environment is in a fast-changing state, and then dynamically correct the similarity evaluation. Adjustment coefficient It is used to balance the contribution of the disturbance correction term to the overall distance. The recommended initial setting is 0.1, which can be adjusted within the range of 0.05~0.5 through cross-validation.

[0053] In obtaining the sample similarity distance Then, the weighting factor of the historical sample is calculated by the following normalization formula : ; in, Represents the total number of historical samples; is the temperature coefficient, which controls the degree of nonlinear compression of the similarity distance in the weight. This will cause similar samples to be given extremely high weights, while samples that are slightly farther away are quickly suppressed. The recommended value range is 1 to 5, and the recommended initial value is 2.

[0054] Weighting Factor Finally, it is used in the loss function of the SVR training process, for example, in the ε-insensitive loss function, the error term of each sample is multiplied by the corresponding , or by introducing the sample weight matrix to participate in the construction of the Lagrangian objective function, the contribution of each sample in model training can be adjusted.

[0055] Through the above-mentioned sample weighting mechanism, the SVR model will focus more on the historical samples that are closest to the current working conditions during training, reducing the interference of abnormal samples or data far from the center of the working conditions, greatly improving the accuracy and stability of the coating thickness prediction model in actual disturbance complex environments. At the same time, this method has good feasibility and is suitable for integrated optimization based on scikit-learn, XGBoost or custom SVR frameworks.

[0056] Step S103: using the current real-time environmental parameters as input, combined with the coating thickness prediction coefficient matrix, dynamically calculating the target coating thickness range and its change trend of the workpiece under the current conditions, and generating a thickness deviation compensation signal including the deviation direction and degree.

[0057] In step S103, the system uses the coating thickness prediction coefficient matrix constructed in step S102 and combines it with the real-time environmental parameters collected by the sensor at the current moment to perform dynamic calculations to obtain the coating thickness range and its change trend that may be formed on the current workpiece under a given environment. The core of this step is to highly integrate the model prediction results with the real-time working conditions, thereby generating a thickness deviation compensation signal for regulating process parameters.

[0058] First, the system inputs the currently collected parameters such as ambient temperature, relative humidity, viscosity of the dipping liquid, and surface temperature of the workpiece into the trained nonlinear coupling prediction model. The coating thickness prediction coefficient matrix stored in the model will perform weighted operations on these parameters and output the theoretical coating thickness prediction value under the current environment. Considering that fluctuations in environmental conditions may cause the coating thickness to show a trend deviation in a short period of time, the system will introduce a short-term sliding window (such as environmental parameter change data within the past 3 to 5 minutes) to calculate the trend derivative, and further obtain the directional information of the coating thickness change over time, that is, to judge whether the thickness is tending to thicken, thin, or in a stable state.

[0059] After completing the above prediction, the system compares the currently predicted theoretical thickness value with the preset target thickness range, calculates the current deviation, and determines the direction of the deviation (positive or negative) and the degree of deviation (numerical level). For example, when the theoretically predicted thickness is significantly greater than the upper limit of the target range, the system will generate a negative compensation signal, prompting the subsequent process to adopt a thinning strategy; if the thickness is close to the lower limit, a positive compensation signal is generated to guide the process to increase the coating amount.

[0060] In order to improve the control accuracy and response speed, the compensation signal should not only contain the numerical value of the thickness deviation, but also the weight information coupled with the environmental fluctuation trend. For example, when the ambient humidity rises rapidly, the same thickness deviation may correspond to different response amplitudes. In addition, the system can also introduce a threshold mechanism to grade the prediction deviation to avoid over-compensation for small disturbances, which will cause the control system to be unstable.

[0061] Finally, the system encodes the deviation direction, degree and trend characteristics calculated above into a thickness deviation compensation signal and transmits it to the next control step in a unified format. This signal serves as the core feedback basis to drive the subsequent steps to adaptively adjust the process parameters, so that the entire control chain forms a closed-loop logic from prediction, analysis to response, ensuring that the coating thickness is always within the controllable range under dynamic conditions.

[0062] In summary, step S103 generates a thickness compensation signal with clear directionality and quantitative information by fusing the real-time environmental parameters with the historical modeling results, which provides a scientific basis for subsequent process parameter adjustments and is a key link in achieving dynamic thickness control and real-time compensation.

[0063] Furthermore, the current real-time environmental parameters are used as input, combined with the coating thickness prediction coefficient matrix, to dynamically calculate the target coating thickness range and its change trend of the workpiece under the current conditions, and generate a thickness deviation compensation signal including the deviation direction and degree, including: The ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece collected and standardized by the sensor at the current moment are used to form a real-time environmental parameter vector, and the parameter vector is passed as input to the coating thickness prediction model obtained by the previous training, and the theoretical coating thickness prediction value corresponding to the current working condition is obtained through model reasoning; The difference between the theoretical prediction value and the target coating thickness interval preset in the process is calculated to obtain the deviation value and deviation sign information, and a trend vector constructed by a continuous environmental parameter change sequence in the previous period is introduced to identify the trend direction of the coating thickness change by comparing the relative increase and decrease relationship between the current prediction value and the prediction values ​​of multiple past periods; The deviation value, deviation sign information and trend direction data are used as joint input to generate a thickness deviation compensation signal, which includes an identifier indicating the direction of adjustment of the workpiece coating thickness, a grading value indicating the required correction amplitude level, and a structured control field for subsequent linkage regulation of process parameters.

[0064] In the plastic coating thickness control method of the present invention, in order to realize real-time compensation control, it is necessary to dynamically generate a thickness compensation signal according to the current environmental state to drive the subsequent process parameter adjustment. To this end, the system first collects key process parameters such as ambient temperature, relative humidity, viscosity of the plastic dipping liquid and surface temperature of the workpiece in real time through multiple sensors distributed at different positions of the plastic dipping production line. After all the collected data are standardized, they are combined into a high-dimensional environmental parameter vector according to the preset feature construction rules. The vector contains the current value, sliding mean, rate of change and other contents of each physical parameter. This vector represents the state of the working condition at the current moment in a multidimensional environment.

[0065] The environmental parameter vector constructed above is input into the coating thickness prediction model that has been trained in the previous step. The model is based on the support vector regression algorithm or other machine learning structures. It uses the obtained thickness prediction coefficient matrix for inference and outputs the theoretical coating thickness value corresponding to the workpiece under the current working condition. The predicted value is the natural coating thickness response caused by the current environment without any intervention or regulation.

[0066] In order to determine whether the predicted thickness meets the process requirements, the system presets a target coating thickness range, which is usually set by process expert experience, quality standards or historical statistical data to reflect the required upper and lower thickness tolerances. The current predicted value is compared with the target range. If it exceeds the upper limit or is lower than the lower limit, the difference between it and the target boundary is calculated to generate a deviation value, and its positive and negative directions are recorded to characterize whether it is too thick or too thin.

[0067] After the difference calculation is completed, the system further introduces a trend judgment mechanism. This mechanism constructs a trend vector based on the current and past prediction results. Through the time series analysis method, the continuous prediction values ​​in the past several time windows are sorted and the change direction is analyzed to determine whether the current thickness deviation is continuing to expand, stabilizing, or starting to regress. Trend judgment not only considers the direction of change of the value size, but also combines the frequency and amplitude of the predicted value fluctuation to form an information vector with more contextual semantics.

[0068] Finally, the system inputs the three types of information, namely, deviation value, deviation direction sign and trend vector, into the deviation compensation signal generation module. This module constructs a set of structured control fields, including a direction indicator indicating whether thickness adjustment is currently required (such as thickening or thinning), a level field indicating the adjustment range (such as first-level fine-tuning, second-level medium-tuning or third-level emphasis), and a standardized output interface field for subsequent linkage controller calls, such as numbers encoded as different strategy templates, correction ratio parameters connected to the PID controller, etc. The compensation signal will be transmitted to the process control module in real time to dynamically drive the linkage adjustment of key parameters such as workpiece preheating temperature, dipping residence time or lifting speed.

[0069] The above steps constitute the whole process from environmental collection, prediction calculation, deviation assessment, trend identification to control output. It has a clear data flow chain and module division logic, and can be directly integrated into the actual dip coating control system to ensure that the coating quality remains stable under multiple disturbance conditions.

[0070] Furthermore, the difference between the theoretical prediction value and the target coating thickness interval preset by the process is calculated to obtain the deviation value and deviation sign information, and a trend vector constructed by a continuous environmental parameter change sequence in the previous period is introduced. By comparing the relative increase and decrease relationship between the current prediction value and the prediction values ​​of multiple past periods, the trend direction of the coating thickness change is identified, including: Dynamically expand the target coating thickness range, call the environmental adaptation coefficient stored in the process knowledge base according to the current workpiece type, material batch and actual temperature state of the dipping solution, and adjust the original thickness setting range to the effective control tolerance range under the current working conditions; The environmental parameter input vectors and the corresponding thickness prediction values ​​in multiple consecutive time segments are sorted in time series, and a multi-period prediction trajectory is constructed based on a sliding window structure to calculate the monotonicity discrimination results, amplitude statistics, and direction consistency scores in each period to generate intermediate trend data for trend identification; The current predicted value is compared with the boundary of the extended target interval. Combined with the directional consistency score extracted from the trend data and the recent fluctuation frequency level, a comprehensive judgment is made as to whether the thickness change direction is continuous deviation, fluctuation stabilization or regression to the target. Based on this, a trend attribute identifier is attached to the deviation symbol for subsequent compensation signal structure generation.

[0071] In the present invention, in order to further improve the dynamic response capability of coating thickness control, it is necessary not only to rely on the difference between the static thickness prediction value and the target value for control, but also to consider the change trend of the current prediction value, so as to achieve a more forward-looking and stable control strategy. Therefore, before generating a thickness deviation compensation signal, the system needs to determine the deviation between the current prediction value and the target coating thickness interval, and identify whether the deviation is in an expansion, convergence or random fluctuation state.

[0072] First, for the target coating thickness range preset by the process, the present invention introduces a dynamic expansion mechanism to avoid misjudgment caused by the overly rigid static target definition in different batches, different materials or different dipping liquid states. In implementation, the system will obtain the type identification of the current workpiece, the material batch code, and the real-time collected dipping liquid temperature value, and use this as an index to call the preset environmental adaptation coefficient in the process knowledge base. These coefficients are adjustment factors obtained through training and modeling of a large amount of historical production data, which can reflect the degree of influence of different working conditions on the coating thickness control boundary. The original upper and lower limits of the thickness will be fine-tuned under the action of these factors, thereby forming a dynamic control tolerance range suitable for the current working conditions.

[0073] Subsequently, the system serializes the thickness prediction value at the current moment with the prediction values ​​in multiple consecutive time segments of the previous period. Each time segment corresponds to a set of environmental parameter inputs and the coating thickness prediction value output by the model. The system arranges these prediction results in chronological order and applies a sliding window structure for analysis. The sliding window can be set to a fixed number of time points (for example, 10 sampling periods), and the window slides one bit each time to cover the local change trends in different time periods. In each window, the system calculates the monotonicity judgment result of the current period (that is, whether the predicted value continues to rise, fall or oscillate), the maximum and minimum value difference as a variation index, and the consistency score of the direction of change of the predicted value, which is used to measure the strength of the trend stability in the time period. These statistics together constitute the intermediate trend data required for trend identification.

[0074] Finally, the system compares the boundary of the current predicted value with the aforementioned dynamically expanded target coating thickness interval to determine the direction and distance in which it exceeds the target interval. On this basis, the evolution trend of the predicted value is determined in combination with the directional consistency score and fluctuation frequency level in the intermediate trend data. If the deviation continues to expand and the directional consistency score is high, it is judged as "continuous deviation"; if the predicted value fluctuates back and forth between the upper and lower limits of the target and the consistency is low, it is judged as "fluctuation stabilization"; if the predicted value gradually returns to the target interval from the deviation section and is accompanied by a positive change in the directional consistency score, it is judged as "return to target". The judgment result will be attached to the deviation direction symbol in the form of a trend attribute identifier, which is used to achieve differentiated control in the subsequent compensation signal structure generation.

[0075] This step not only enhances the dynamics and contextual relevance of thickness deviation judgment, but also provides rich trend feature input for subsequent process parameter regulation. It has significant practical value and wide adaptability under production conditions where coating thickness is significantly affected by complex disturbances.

[0076] Step S104: adjusting the dipping control parameters including the workpiece preheating temperature, dipping holding time and lifting speed according to the thickness deviation compensation signal to form a process parameter optimization plan.

[0077] In step S104, the system adjusts multiple key dipping process parameters in conjunction with the thickness deviation compensation signal generated in step S103, thereby forming an optimized control scheme that dynamically matches the current environmental conditions. This step is the core link in the transformation of the compensation strategy into actual control instructions, requiring comprehensive coordination of the mutual influence between different parameters to ensure that the final coating thickness approaches the target value.

[0078] In the specific implementation, the thickness deviation compensation signal contains the difference between the coating thickness and the target value, the deviation direction (thickening or thinning), the change trend (increasing or decreasing), and the corresponding environmental disturbance factor weight. The system first analyzes the compensation signal to extract the control direction and control amplitude that need to be adjusted. For example, when the system recognizes that the current coating thickness is too thick and the predicted trend continues to thicken, the system should actively reduce the formation strength of the coating and achieve the control target by shortening the dipping time, reducing the preheating temperature of the workpiece, or speeding up the lifting speed.

[0079] Specifically, the workpiece preheating temperature is an important parameter that affects the amount of plastic liquid adsorbed. A higher preheating temperature will cause the plastic liquid to adhere more to the workpiece surface, forming a thicker coating. The system can lower the target preheating temperature by adjusting the power output of the infrared heater or hot air system. In order to achieve fast response and uniform heating, it is preferred to use a PID thermostat with closed-loop feedback to ensure stable temperature adjustment.

[0080] The dipping and holding time directly determines how long the workpiece absorbs the material in the dipping liquid. If the thickness deviation is positive, the system can extend the immersion time to increase the coating thickness; if it is a negative deviation, the dwell time can be shortened. This parameter can be used to adjust the motor control logic through PLC or industrial control software to accurately control the time the robot arm or fixture remains under the liquid surface within the error range (for example, ±0.2 seconds).

[0081] The lifting speed affects the time it takes for the liquid to flow naturally from the surface of the workpiece, and plays an important role in the uniformity of the coating and the thickness boundary. A slower lifting speed is conducive to a smooth transition and a thicker coating, while a faster speed reduces the residual thickness of the liquid. The system can use a servo motor to control the lifting mechanism and set a multi-stage speed curve according to the thickness compensation signal. For example, a slower speed is set at the beginning of the lifting stage to stabilize the bottom thickness, and then gradually accelerate to reduce top accumulation.

[0082] The above three process parameters are not adjusted in isolation, but constitute a linkage optimization mechanism. The system comprehensively balances the adjustment range of different parameters by establishing preset weights or online optimization algorithms to ensure that the control strategy achieves the best balance between response accuracy and physical feasibility. For example, under high temperature and high humidity conditions, in order to prevent local condensation from affecting the adsorption performance of the workpiece, the system prefers to adjust the lifting speed rather than the preheating temperature.

[0083] Ultimately, all adjustment parameters will be packaged into a structured data set, namely the "process parameter optimization plan". This plan is transmitted to the control module of the dipping equipment in a standardized interface format to drive each physical component to perform corresponding actions, thereby forming a control strategy that is highly consistent with the current compensation requirements.

[0084] In summary, step S104 not only realizes the conversion of deviation analysis results into equipment action instructions, but also ensures the accuracy, stability and real-time response capability of the adjustment process through a multi-parameter coordinated control mechanism, thus laying a foundation for the precise control of subsequent dipping equipment.

[0085] Furthermore, the plastic dipping control parameters including the workpiece preheating temperature, the plastic dipping holding time and the lifting speed are adjusted according to the thickness deviation compensation signal to form a process parameter optimization scheme, including: The deviation direction identification, correction amplitude level and trend attribute fields contained in the thickness deviation compensation signal are mapped to control strategy indexes respectively, and grouped and processed according to the process parameter type, and divided into temperature parameters, time parameters and speed parameters, and the priority order is set according to the historical response sensitivity of each parameter to thickness adjustment; The preset process rule library is called according to the strategy index. The process rule library is constructed based on a large amount of historical production sample data, and includes the empirical correspondence between various control parameters and coating thickness under different working conditions, adjustment amplitude boundaries, and parameter linkage mapping tables. The candidate parameter combination that meets the response conditions is selected from the process rule library in combination with the current actual environmental state, and the rule consistency check and correction are performed; A dynamic evaluation mechanism is introduced into the candidate parameter combinations, and the thickness control effect of each parameter combination in similar historical working conditions is fitted and backtested. Combined with the confidence level of the current prediction model and the working condition disturbance stability index, the optimal control path is selected to generate the final process parameter optimization plan.

[0086] In the present invention, the real-time control of the thickness of the dip coating depends on the refined process adjustment for the deviation condition. In order to realize the automatic and intelligent control strategy generation, the system first receives the thickness deviation compensation signal generated from the previous step. This signal not only contains the positive and negative directions and numerical values ​​of the deviation value, but also further encapsulates the dynamic properties related to the development trend of the deviation, such as whether the deviation continues to expand, tends to stabilize, or gradually returns to the target interval. The system will extract three core fields from the signal, namely the deviation direction identification, the correction amplitude level, and the trend attribute. These three contents are used to map to the target direction, adjustment intensity, and dynamic response model of the subsequent control strategy.

[0087] In order to improve the pertinence and execution efficiency of parameter control, the system classifies the process parameters to be adjusted according to the physical control logic, and divides them into temperature parameters (such as workpiece preheating temperature), time parameters (such as dipping insulation time) and speed parameters (such as lifting speed). This grouping method makes it easy to set the adjustment logic and interaction boundaries of each type of parameter separately. Furthermore, the system sets response priorities for different types of parameters based on the response sensitivity of the coating thickness to various control parameters recorded in the historical production data. If the historical data shows that the thickness is particularly sensitive to changes in lifting speed, the system will give priority to adjusting the lifting speed parameter when the current deviation is negative and the trend continues to decline.

[0088] The system then calls the preset process rule library. The rule library is built on the basis of a large amount of historical production sample data, covering process response strategies under different materials, working conditions, and batch conditions. Each record in the rule library includes the empirical correspondence between the control parameter combination and the thickness response, the upper and lower limits of the parameter adjustment, and the linkage mapping relationship between the parameters. For example, if the lifting speed increases, it may be necessary to shorten the insulation time or increase the preheating temperature of the workpiece accordingly. The linkage rule will ensure the overall consistency of the control strategy. The system selects qualified candidate parameter combinations from the rule library based on the aforementioned control strategy index and the current real-time environmental parameters, such as the temperature of the dipping liquid, the ambient humidity, or the viscosity change state. If some combinations have logical conflicts because the current state does not meet the boundary conditions, the system will automatically perform a consistency check and fine-tune the parameters according to the priority strategy.

[0089] In order to select the optimal control path from multiple candidate solutions, the system introduces a dynamic evaluation mechanism. This mechanism relies on historical similar working condition backtracking data to re-analyze the thickness adjustment effect caused by each set of parameter combinations in the past dipping process. The system not only pays attention to whether the final thickness enters the target range, but also evaluates the time required for adjustment, fluctuation amplitude and convergence trend, and combines the confidence level of the current prediction model as a stability reference. In addition, the system also introduces current environmental disturbance stability indicators, such as ambient temperature change rate, viscosity disturbance frequency and workpiece temperature uniformity distribution, as additional criteria for the adaptability of candidate solutions.

[0090] Based on the above information, the system constructs a multi-dimensional control scoring system, and finally selects a set of parameter combinations with the best backtesting effect, high model confidence and the most adaptable in the current disturbance environment, and outputs it as the final process parameter optimization solution. The solution adopts the form of structured configuration, including target parameter items, adjustment magnitude, execution timing, linkage logic and fallback strategy, etc., and is transmitted to the control equipment execution layer through a standard interface to achieve closed-loop dynamic control of the dipping process.

[0091] Through the above technical path, the present invention can achieve precise control of coating thickness while effectively improving the adaptability of process parameter response, convergence efficiency of adjustment and fault tolerance to disturbance conditions, and is suitable for the dip-molding manufacturing process in multiple batches and variable environments. The entire process does not rely on manual judgment and can be fully embedded in the industrial automation control system for deployment.

[0092] Furthermore, a dynamic evaluation mechanism is introduced into the candidate parameter combinations to perform fitting backtesting on the thickness control effect of each parameter combination in historical similar working conditions, and the optimal control path is selected to generate the final process parameter optimization solution in combination with the confidence level of the current prediction model and the working condition disturbance stability index, including: Construct a historical environment similarity index corresponding to the candidate parameter combination, and based on the environmental characteristic range of each set of parameters in previous dipping batches, screen out sample records implemented under similar working conditions, and extract the corresponding coating thickness deviation correction results as the backtest data set; A multi-period backtest evaluation is performed on the backtest data set of each candidate parameter combination, and the stability of the thickness adjustment effect, the adjustment convergence rate, and the final compression rate of the thickness deviation under different disturbance levels are statistically analyzed to construct a multi-dimensional control performance vector. The confidence output of the current prediction model is used as a control risk compensation factor to perform weighted aggregation on the performance vector. The weighted aggregation result is integrated with the real-time working condition disturbance sensitivity matching value. The sensitivity matching value is jointly generated by the ambient temperature fluctuation rate, viscosity mutation frequency and workpiece temperature gradient in the current working condition. It is used to evaluate the adaptability of the parameter combination to the current disturbance situation, and finally select the parameter combination with the best control performance and the lowest risk as the final output process parameter optimization solution.

[0093] In the thickness control method proposed in the present invention, in order to select the optimal control path from multiple feasible process parameter combinations, the system introduces a dynamic evaluation mechanism in the process parameter optimization stage. This mechanism does not only rely on static rule matching or empirical judgment, but is based on a data-driven method. Based on historical similar working condition data, the control capability of each group of parameter combinations is retrospectively analyzed and multi-dimensionally evaluated, combined with the current working condition disturbance characteristics and model prediction confidence, so as to generate the most executable control plan.

[0094] In order to establish the basis of the evaluation mechanism, the system first constructs the corresponding historical environmental similarity index for each set of candidate parameter combinations. The index is matched based on the environmental characteristic intervals that the parameter combination has been used for in the previous dipping production process, including but not limited to the ambient temperature distribution, dipping liquid viscosity level, workpiece initial temperature when entering the mold, and relative humidity fluctuation range. After the system performs multi-dimensional standardization on the environmental data in the historical samples, it calculates the distance or similarity score with the current working conditions, selects historical samples with similar working condition structures, and extracts the coating thickness adjustment results generated by the actual use of the same or similar parameter combinations in these samples as the backtest data set for the combination.

[0095] Based on the above backtest data, the system conducts a multi-period retrospective evaluation of each set of candidate parameter combinations. Under historical environmental conditions with different disturbance levels, the system separately counts the thickness control performance of the parameter combination under various working conditions, including its fluctuation amplitude during the adjustment process, the thickness deviation convergence rate, and whether the deviation can be compressed to within the target range in the end. These results constitute a multi-dimensional control performance vector, which is used to quantify the control ability of the parameter combination. In addition, considering that the output results of the current prediction model may be affected by noise or extrapolated from working condition characteristics, the system introduces the current confidence of the prediction model for thickness prediction into the evaluation system as a risk compensation factor, and weights the above control performance vector, thereby balancing the risk of uncertainty in the control effect.

[0096] In order to further enhance the adaptability between the selected parameter combination and the current actual disturbance environment, the system also introduces a working condition disturbance sensitivity matching mechanism. Based on the current real-time environmental data, this mechanism calculates key disturbance indicators, such as the ambient temperature fluctuation rate per unit time, the viscosity change mutation frequency, and the workpiece temperature gradient. The system compares these disturbance characteristics with the performance of the corresponding parameter combination in the historical data to evaluate whether the parameter combination still has good robustness and execution efficiency under the current disturbance background.

[0097] Finally, the system fuses the weighted control performance vector with the disturbance sensitivity matching value to form a comprehensive evaluation index, and selects a set of parameters from all candidate parameter combinations that have excellent performance in historically similar environments, are highly adaptable under current working conditions, and have the lowest predicted risk as the final output process parameter optimization solution. This solution can be used for parameter configuration of automatic control equipment, or uploaded to the industrial execution system interface as a manual intervention suggestion to achieve dynamic fine-tuning and rapid response to process parameters.

[0098] Through the above construction method, the present invention realizes closed-loop evaluation of process optimization at the parameter control level, and takes into account the comprehensive balance of prediction ability, execution risk and disturbance adaptability, which is particularly suitable for dipping application scenarios with multiple disturbances, easy fluctuations, and high thickness precision requirements. This method does not need to rely on external expert judgment, and can be completely automatically executed by the edge controller or central control platform to achieve intelligent, adaptive, and highly robust process control.

[0099] Step S105: Using the process parameter optimization scheme to control the execution process of the dipping equipment so that the coating thickness dynamically approaches the target value, and using the execution result as feedback input to continuously optimize the nonlinear coupling prediction model.

[0100] In step S105, the system will directly control the operation of the dipping equipment according to the process parameter optimization scheme generated in the previous step to perform precise coating thickness adjustment operations. At the same time, this step also includes real-time collection and comparison of the actual thickness of the finished coating product, and the results are input back into the nonlinear coupling prediction model as feedback information to achieve dynamic self-learning and continuous optimization of the model.

[0101] In actual implementation, the system first sends the various control instructions in the optimization plan to the corresponding equipment control unit. The control of the preheating temperature of the workpiece is usually achieved through an infrared heater or a hot air circulation furnace. The system sets the target temperature according to the optimization plan, and adjusts the heating power and time in real time through the temperature control module to ensure that the workpiece reaches the set thermal state before entering the dipping tank. The control of the dipping and insulation time is achieved by setting the movement residence time of the manipulator or conveying device. The system accurately controls the length of time the workpiece stays in the dipping liquid according to the optimization plan, which can generally be accurate to within 0.1 seconds. The adjustment of the lifting speed depends on the stepper motor or servo mechanism. By setting the acceleration, maximum speed and speed change curve, the workpiece can be separated from the dipping liquid at the optimal rhythm to ensure the uniformity and integrity of the coating thickness.

[0102] During the entire equipment operation process, the system continuously collects the actual operating parameters of each key link through the embedded sensor module, such as the actual workpiece temperature, dwell time, lifting speed execution trajectory, etc., and transmits them back to the central control system through the communication bus (such as Modbus, etc.). At the same time, after the workpiece is dipped, the system measures the actual thickness of the coating in real time through a non-contact laser thickness gauge or ultrasonic thickness gauge. The measured actual thickness data is compared with the model prediction value and the target thickness to determine the effectiveness of the current process execution.

[0103] In order to further improve the adaptability and long-term accuracy of the model, the system takes the above comparison results and the actual environmental parameters as new samples and adds them to the model update queue. According to the set update strategy, such as fixed-period retraining or error-exceeding-triggered training, the system can use the new data to perform incremental training or full retraining on the original nonlinear coupling prediction model. After the training is completed, the system will automatically replace the parameter matrix in the original model to achieve dynamic optimization of the model. This feedback learning mechanism can significantly improve the model's adaptability to factors such as equipment aging, material fluctuations, and long-term environmental changes, so that the system has the characteristics of continuously evolving and self-correcting intelligent control.

[0104] During the execution of the entire step, all control logic and data processing processes can be integrated and run in the industrial control system (such as PLC or edge computing gateway), and cooperate with the SCADA system for visual monitoring and abnormal alarm. Through the above method, not only a closed-loop system from prediction, control to feedback optimization is realized, but also the thickness of the dip coating is always close to the target value under dynamic conditions, meeting the industrial production needs of high consistency and high stability.

[0105] Furthermore, the process parameter optimization scheme is used to control the execution process of the dipping equipment so that the coating thickness dynamically approaches the target value, and the execution result is used as feedback input to continuously optimize the nonlinear coupling prediction model, including: Before controlling the equipment, the system sets an execution buffer for each control parameter involved in the optimization plan, and sets a delayed response strategy based on the current working condition stability level, so that the equipment can execute in stages or gradually after receiving the command, avoiding coating overshoot or abnormal mechanical response due to sudden changes in parameters; During the dipping operation, the initial coating thickness of the workpiece after it leaves the liquid surface is monitored in real time by sensors, and the difference is compared with the predicted thickness target. At the same time, the environmental disturbance state and the equipment response behavior are packaged to form an execution feedback data segment, which contains process parameter instructions, actual response values ​​and process disturbance information, and is used to mark the source of the offset between the model input and the actual output. In the feedback input processing stage, the above data fragments are screened according to the intensity of the disturbance impact, and representative samples are extracted to trigger the incremental model optimization process. Without resetting the existing prediction structure, some parameter weights of the nonlinear coupling prediction model are fine-tuned or updated, thereby improving the model's adaptability and prediction robustness under specific disturbance conditions.

[0106] After the process parameter optimization plan is generated, the system will not immediately send all parameters directly to the dipping equipment for full adjustment, but will first set a buffer execution interval for each control parameter based on the disturbance level of the current working conditions, system stability, and equipment response characteristics. The setting of the buffer zone can be based on historical experience rules or obtained through real-time simulation. Its main purpose is to divide the execution of instructions into multiple stages to prevent sudden changes in parameters from causing overshoot on the coating thickness or causing lag and violent vibration in the equipment action. For example, for the adjustment of the lifting speed, the system can set an interval that gradually increases from the current speed to the target speed. The speed change amplitude and response time in each interval are limited by preset rules, thereby ensuring the executability of the control instructions and the smoothness of the process.

[0107] During the execution of the optimization plan, the system continuously monitors key indicators in real time through multiple sets of sensors, especially the initial coating thickness after the workpiece exits the liquid surface as the most sensitive control result, which is regarded as the benchmark for direct comparison with the predicted thickness target. Through this comparison, the system can obtain the thickness deviation caused by the actual control behavior, and combine the environmental disturbance data in the process with the equipment action response behavior to package the relevant information into a structured execution feedback data fragment. Each data fragment contains specific process parameter instructions, actual equipment response values ​​(such as execution temperature, residence time or speed curve), coating thickness results, and disturbance information such as ambient temperature fluctuations, viscosity mutations or humidity interference recorded during the cycle. This structured data provides a complete context for the subsequent analysis of the root cause of the deviation between the model prediction and the actual working conditions.

[0108] When the system completes the execution and data collection of a round of optimization solutions, it will enter the feedback processing stage. In this stage, the system will first grade and screen the collected execution feedback data fragments according to the disturbance intensity to eliminate those samples with weak disturbances, stable change trends or negligible thickness deviations, so as to avoid unnecessary adjustments to the model due to noise interference. The determination of disturbance intensity can be based on the combination of parameters such as fluctuation frequency, gradient change rate or temperature difference offset range to form a scoring standard.

[0109] For the high-value samples after screening, the system further extracts representative data for model fine-tuning. These representative samples generally have the following characteristics: the parameter adjustment amplitude is close to the control boundary, the execution response delay exceeds the average level, or the predicted trend deviates significantly from the actual change direction. The selected samples will enter the model incremental optimization process as an update trigger signal. In this process, the original nonlinear coupling prediction model structure will not be reconstructed as a whole, but a subset of parameters in the model that are highly correlated with the current deviation will be identified, such as the weighted connection relationship between certain environmental features and thickness output, and they will be adjusted or replaced to a limited extent. In order to avoid overfitting of the model or excessive structural disturbance, the optimization process introduces learning rate limits, disturbance suppression functions and strategy interpolation mechanisms, so that the model can improve its adaptability and robustness to specific disturbance environments while maintaining its original prediction capabilities.

[0110] This method constructs a closed-loop path of real-time response and continuous learning in the coating thickness control system through five continuous links: "process parameter execution-thickness response monitoring-structured feedback collection-disturbance sample screening-model local correction". Its design takes into account engineering feasibility, data utilization efficiency and long-term model stability. It is especially suitable for dip-molding manufacturing scenarios with high-frequency disturbances and strong parameter nonlinear coupling. It also has the ability to integrate with existing MES systems and industrial edge computing platforms, and can achieve real-time deployment and automatic updates.

[0111] Furthermore, in the feedback input processing stage, the above data segments are screened according to the intensity of the disturbance impact, and representative samples are extracted to trigger the incremental model optimization process. Without resetting the existing prediction structure, some parameter weights of the nonlinear coupling prediction model are fine-tuned or updated, thereby improving the model's adaptability and prediction robustness under specific disturbance conditions, including: Based on the disturbance information contained in the feedback data segment, including the ambient temperature fluctuation frequency, the viscosity mutation amplitude of the dipping liquid, and the thermal non-uniformity index of the workpiece surface, a disturbance impact intensity score is constructed to quantify the potential interference ability of the data segment on the coating thickness prediction deviation, and a disturbance threshold screening rule is set to eliminate invalid feedback data with no significant disturbance or errors attributable to measurement errors; In the data segments that meet the disturbance intensity threshold, identify the feedback samples where the parameter adjustment amplitude is at the edge of the control range, the equipment response delay time exceeds the empirical expectation, or the thickness recovery trend is inconsistent with the predicted direction, and give priority to selecting data with stronger correlation with model deviation performance as representative samples for subsequent update processes; During the model updating stage, the entire nonlinear prediction structure is not retrained. Instead, based on the systematic differences between the input features and the actual deviations in the current representative samples, some feature weights or support vector sets in the model are locally corrected. Through parameter interpolation, learning rate constraints and perturbation conservative strategies, the update amplitude and update area are limited to keep the stability and adaptability of the model structure balanced in a disturbance-dominated environment.

[0112] In the present invention, the real-time and accuracy of the dip coating thickness control are highly dependent on the prediction model's ability to sensitively identify the current working condition disturbance. However, traditional prediction models usually maintain a fixed structure after initial training, and are difficult to adapt to the changing disturbance characteristics in the actual environment. Therefore, the present invention proposes to introduce a feedback data-driven incremental optimization strategy during model operation. The first step of this strategy is to score and judge the intensity of the disturbance impact on each execution feedback data segment to identify which feedback information is a representative sample that is truly worth correcting.

[0113] Each feedback data fragment contains a set of multi-dimensional disturbance features and corresponding execution effect data, mainly including parameters such as the frequency of ambient temperature fluctuations, the sudden change amplitude of the viscosity of the plastic dipping liquid, and the non-uniformity index of the heat distribution on the workpiece surface. This information is collected and recorded in real time during the execution phase through multiple sets of sensors on the device side. Based on the preset disturbance evaluation model, the system maps the above disturbance dimensions into a unified disturbance impact score to quantify the degree of interference that the data fragment may cause to the coating thickness prediction. Combined with the set disturbance intensity threshold, the system automatically eliminates invalid feedback fragments with weak disturbances, stable trends, or errors fluctuating within the normal range, ensuring that the model update is based only on data that has a substantial impact on the accuracy of the prediction.

[0114] In the valid data segments, the system further mines representative samples that are highly correlated with the model prediction deviation. The following three types of samples are selected: First, the parameter adjustment amplitude has approached or reached the minimum or maximum value allowed by the system, indicating that the control process has reached the boundary state; second, there is a significant delay in the actual response of the equipment, indicating that the control execution lag may affect the timeliness of the model prediction; third, the recovery trend of the workpiece coating thickness is obviously inconsistent with the direction predicted by the model, reflecting that the model has structural errors under this type of disturbance conditions. The system preferentially selects these samples to drive the subsequent local model optimization process.

[0115] In the model optimization stage, in order to avoid problems such as unstable model structure, large consumption of computing resources and high data requirements caused by retraining, the present invention adopts an incremental fine-tuning mechanism. Specifically, the entire nonlinear coupling prediction model is not rebuilt, but only some feature parameter weights or support vector sets in the model are locally corrected while keeping the original structure unchanged. This correction method is based on the analysis of the systematic differences between the input features of representative samples and the corresponding prediction deviations. By identifying the dimensions of the main error sources, a refined adjustment strategy is implemented. During the fine-tuning process, the system introduces a parameter interpolation mechanism to avoid sudden changes in weights; at the same time, the upper limit of the learning rate is set to control the update speed, and a disturbance suppression mechanism is configured to ensure that the correction amplitude is within the tolerance range of the model. All these strategies serve the stability of the model structure while improving its responsiveness to new working conditions under disturbance-dominated scenarios.

[0116] Through the above three-layer mechanism, namely, perturbation score screening, high deviation sample identification and constrained local update, the present invention constructs a model self-evolution path that dynamically adapts to process disturbance changes without destroying the original modeling logic. This mechanism is extremely suitable for the manufacturing process of dip coating, which has frequent working conditions fluctuations, short control feedback cycle, and high requirements for thickness stability. It can significantly extend the model life cycle, improve prediction accuracy, and reduce the frequency of manual maintenance, realizing a truly intelligent, low-intervention process control system.

[0117] A second embodiment of the present application provides an electronic device, the electronic device comprising: processor; The memory is used to store a program. When the program is read and executed by the processor, it executes a dip coating thickness control method based on real-time compensation of multi-dimensional environmental parameters provided in the first embodiment of the present application.

[0118] The third embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, a method for controlling the thickness of a dip coating based on real-time compensation of multi-dimensional environmental parameters provided in the first embodiment of the present application is executed.

[0119] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A method for controlling the thickness of a plastic coating based on real-time compensation of multi-dimensional environmental parameters, characterized in that: include: The ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece in the dipping environment are collected through multiple sets of sensors, and the collected data are standardized to construct an environmental parameter data set containing time series characteristics; Based on the environmental parameter data set, a nonlinear coupling prediction model between coating thickness and multidimensional environmental factors is constructed, and a coating thickness prediction coefficient matrix for real-time reasoning is generated during the training phase; Using the current real-time environmental parameters as input, combined with the coating thickness prediction coefficient matrix, dynamically calculate the target coating thickness range and its change trend of the workpiece under the current conditions, and generate a thickness deviation compensation signal including the deviation direction and degree; Adjust the dipping control parameters including the workpiece preheating temperature, dipping insulation time and lifting speed according to the thickness deviation compensation signal to form a process parameter optimization plan; The process parameter optimization scheme is used to control the execution process of the dipping equipment so that the coating thickness dynamically approaches the target value, and the execution result is used as feedback input to continuously optimize the nonlinear coupling prediction model.

2. The method for controlling the thickness of the plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 1 is characterized in that: The method collects the ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece in the dipping environment through multiple sets of sensors, standardizes the collected data, and constructs an environmental parameter data set containing time series characteristics, including: An integrated environmental temperature and humidity sensor, an online rotary viscometer and a non-contact infrared temperature sensor are respectively installed in the preheating section, the dipping section and the lifting section of the dipping production line. Each of the sensors is equipped with a unified timestamp synchronization module to associate the collected data with the same time axis to form a raw data stream with time consistency. Data cleaning and trend extraction are performed on each type of collected data through edge processing nodes. The trend extraction includes calculating the first-order derivative and local fluctuation variance of the parameter within a preset sliding time window, and judging whether it exceeds the set threshold. If it exceeds, the time point is marked as a disturbance point and a disturbance level label is assigned. The standardized raw data, first-order derivatives, local fluctuation variance values ​​and disturbance labels are combined to form an extended feature vector of environmental parameters, and the data window is encapsulated in a fixed period to form a time series parameter data set containing dynamic feature dimensions for subsequent thickness prediction model calls.

3. The method for controlling the thickness of the plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 1, characterized in that: Based on the environmental parameter data set, a nonlinear coupling prediction model between coating thickness and multidimensional environmental factors is constructed, and a coating thickness prediction coefficient matrix for real-time reasoning is generated in the training stage, including: Each parameter of ambient temperature, relative humidity, viscosity of the plastic dipping liquid and workpiece surface temperature is classified according to its change rate in historical data to form two feature sets representing fast-changing parameters and slow-changing parameters respectively. The change amplitude, average trend and disturbance frequency of the parameters in each set within the preset time window are extracted to construct an extended feature vector. Based on the extended feature vector, a support vector regression model based on a combined kernel function is trained, wherein the combined kernel function simultaneously considers the local similarity and nonlinear trend relationship between parameters, and is used to establish a nonlinear mapping relationship between coating thickness and multidimensional environmental factors, and after the training is completed, a coating thickness prediction coefficient matrix for real-time prediction is output; During the model training process, a sample weighting mechanism based on the similarity of environmental characteristics is introduced for historical environmental data samples. The weighting mechanism assigns weights according to the distance between the sample and the current working condition in the multidimensional parameter space, so that historical samples that are closer to the current working condition contribute more to the model training, thereby improving the adaptability and robustness of the prediction results under actual dipping conditions.

4. The method for controlling the thickness of the plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 1, characterized in that: The method uses the current real-time environmental parameters as input, combines the coating thickness prediction coefficient matrix, dynamically calculates the target coating thickness range and its change trend of the workpiece under the current conditions, and generates a thickness deviation compensation signal including the deviation direction and degree, including: The ambient temperature, relative humidity, viscosity of the dipping liquid and the surface temperature of the workpiece collected and standardized by the sensor at the current moment are used to form a real-time environmental parameter vector, and the parameter vector is passed as input to the coating thickness prediction model obtained by the previous training, and the theoretical coating thickness prediction value corresponding to the current working condition is obtained through model reasoning; The difference between the theoretical prediction value and the target coating thickness interval preset in the process is calculated to obtain the deviation value and deviation sign information, and a trend vector constructed by a continuous environmental parameter change sequence in the previous period is introduced to identify the trend direction of the coating thickness change by comparing the relative increase and decrease relationship between the current prediction value and the prediction values ​​of multiple past periods; The deviation value, deviation sign information and trend direction data are used as joint input to generate a thickness deviation compensation signal, which includes an identifier indicating the direction of adjustment of the workpiece coating thickness, a grading value indicating the required correction amplitude level, and a structured control field for subsequent linkage regulation of process parameters.

5. The method for controlling the thickness of the plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 4 is characterized in that: The difference between the theoretical prediction value and the target coating thickness interval preset in the process is calculated to obtain the deviation value and deviation sign information, and a trend vector constructed by a continuous environmental parameter change sequence in the previous period is introduced. By comparing the relative increase and decrease relationship between the current prediction value and the prediction values ​​in the past multiple periods, the trend direction of the coating thickness change is identified, including: Dynamically expand the target coating thickness range, call the environmental adaptation coefficient stored in the process knowledge base according to the current workpiece type, material batch and actual temperature state of the dipping solution, and adjust the original thickness setting range to the effective control tolerance range under the current working conditions; The environmental parameter input vectors and the corresponding thickness prediction values ​​in multiple consecutive time segments are sorted in time series, and a multi-period prediction trajectory is constructed based on a sliding window structure to calculate the monotonicity discrimination results, amplitude statistics, and direction consistency scores in each period to generate intermediate trend data for trend identification; The current predicted value is compared with the boundary of the extended target interval. Combined with the directional consistency score extracted from the trend data and the recent fluctuation frequency level, a comprehensive judgment is made as to whether the thickness change direction is continuous deviation, fluctuation stabilization or regression to the target. Based on this, a trend attribute identifier is attached to the deviation symbol for subsequent compensation signal structure generation.

6. The method for controlling the thickness of a plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 1, characterized in that: The step of adjusting the plastic dipping control parameters including the workpiece preheating temperature, the plastic dipping holding time and the lifting speed according to the thickness deviation compensation signal to form a process parameter optimization scheme includes: The deviation direction identification, correction amplitude level and trend attribute fields contained in the thickness deviation compensation signal are mapped to control strategy indexes respectively, and grouped and processed according to the process parameter type, and divided into temperature parameters, time parameters and speed parameters, and the priority order is set according to the historical response sensitivity of each parameter to thickness adjustment; The preset process rule library is called according to the strategy index. The process rule library is constructed based on a large amount of historical production sample data, and includes the empirical correspondence between various control parameters and coating thickness under different working conditions, adjustment amplitude boundaries, and parameter linkage mapping tables. The candidate parameter combination that meets the response conditions is selected from the process rule library in combination with the current actual environmental state, and the rule consistency check and correction are performed; A dynamic evaluation mechanism is introduced into the candidate parameter combinations, and the thickness control effect of each parameter combination in similar historical working conditions is fitted and backtested. Combined with the confidence level of the current prediction model and the working condition disturbance stability index, the optimal control path is selected to generate the final process parameter optimization plan.

7. The method for controlling the thickness of a plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 6, characterized in that: The dynamic evaluation mechanism is introduced into the candidate parameter combination, and the thickness control effect of each parameter combination in similar historical working conditions is fitted and backtested. In combination with the confidence level of the current prediction model and the working condition disturbance stability index, the optimal control path is selected to generate the final process parameter optimization plan, including: Construct a historical environment similarity index corresponding to the candidate parameter combination, and based on the environmental characteristic range of each set of parameters in previous dipping batches, screen out sample records implemented under similar working conditions, and extract the corresponding coating thickness deviation correction results as the backtest data set; A multi-period backtest evaluation is performed on the backtest data set of each candidate parameter combination, and the stability of the thickness adjustment effect, the adjustment convergence rate, and the final compression rate of the thickness deviation under different disturbance levels are statistically analyzed to construct a multi-dimensional control performance vector. The confidence output of the current prediction model is used as a control risk compensation factor to perform weighted aggregation on the performance vector. The weighted aggregation result is integrated with the real-time working condition disturbance sensitivity matching value. The sensitivity matching value is jointly generated by the ambient temperature fluctuation rate, viscosity mutation frequency and workpiece temperature gradient in the current working condition. It is used to evaluate the adaptability of the parameter combination to the current disturbance situation, and finally select the parameter combination with the best control performance and the lowest risk as the final output process parameter optimization solution.

8. The method for controlling the thickness of a plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 1, characterized in that: The process parameter optimization scheme is used to control the execution process of the dipping equipment so that the coating thickness dynamically approaches the target value, and the execution result is used as feedback input to continuously optimize the nonlinear coupling prediction model, including: Before controlling the equipment, the system sets an execution buffer for each control parameter involved in the optimization plan, and sets a delayed response strategy based on the current working condition stability level, so that the equipment can execute in stages or gradually after receiving the command, avoiding coating overshoot or abnormal mechanical response due to sudden changes in parameters; During the dipping operation, the initial coating thickness of the workpiece after it leaves the liquid surface is monitored in real time by sensors, and the difference is compared with the predicted thickness target. At the same time, the environmental disturbance state and the equipment response behavior are packaged to form an execution feedback data segment, which contains process parameter instructions, actual response values ​​and process disturbance information, and is used to mark the source of the offset between the model input and the actual output. In the feedback input processing stage, the above data fragments are screened according to the intensity of the disturbance impact, and representative samples are extracted to trigger the incremental model optimization process. Without resetting the existing prediction structure, some parameter weights of the nonlinear coupling prediction model are fine-tuned or updated, thereby improving the model's adaptability and prediction robustness under specific disturbance conditions.

9. The method for controlling the thickness of the plastic coating based on real-time compensation of multi-dimensional environmental parameters according to claim 8, characterized in that: In the feedback input processing stage, the above data segments are screened according to the intensity of disturbance impact, and representative samples are extracted to trigger the incremental model optimization process. Without resetting the existing prediction structure, some parameter weights of the nonlinear coupling prediction model are fine-tuned or updated, thereby improving the model's adaptability and prediction robustness under specific disturbance conditions, including: Based on the disturbance information contained in the feedback data segment, including the ambient temperature fluctuation frequency, the viscosity mutation amplitude of the dipping liquid, and the thermal non-uniformity index of the workpiece surface, a disturbance impact intensity score is constructed to quantify the potential interference ability of the data segment on the coating thickness prediction deviation, and a disturbance threshold screening rule is set to eliminate invalid feedback data with no significant disturbance or errors attributable to measurement errors; In the data segments that meet the disturbance intensity threshold, identify the feedback samples where the parameter adjustment amplitude is at the edge of the control range, the equipment response delay time exceeds the empirical expectation, or the thickness recovery trend is inconsistent with the predicted direction, and give priority to selecting data with stronger correlation with model deviation performance as representative samples for subsequent update processes; During the model updating stage, the entire nonlinear prediction structure is not retrained. Instead, based on the systematic differences between the input features and the actual deviations in the current representative samples, some feature weights or support vector sets in the model are locally corrected. Through parameter interpolation, learning rate constraints and perturbation conservative strategies, the update amplitude and update area are limited to keep the stability and adaptability of the model structure balanced in a disturbance-dominated environment.

Images