A method for optimizing temperature control in a fully automated laminating machine with multi-parameter coordination during the lamination process.

By standardizing and classifying the multi-source heterogeneous process parameters of the laminating machine using decision trees, generating process semantic labels, establishing a mapping table, and performing safety verification, the problem of safe dynamic reconstruction of the temperature control objective function of the laminating machine under varying operating conditions was solved. This achieved zero-risk smooth transition and millisecond-level rapid response, improving laminating quality and response speed.

CN123114147APending Publication Date: 2026-11-17GUANGZHOU GREAT DRAGON PRINTING & GRAPHIC CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611547886.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-08
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing temperature control methods for laminating machines struggle to achieve safe dynamic reconstruction of the temperature control objective function under varying operating conditions, leading to instability in the temperature control strategy and fluctuations in lamination quality. Furthermore, existing online learning solutions are insufficient to meet the real-time response and low-latency requirements of industrial settings.

Method used

By acquiring multi-source heterogeneous process parameters, standardizing them, and inputting them into a decision tree classifier, process semantic labels are generated, and a semantic-target structure mapping table is established to realize the dynamic calling and replacement of the temperature control objective function. A background pre-simulation unit is introduced for safety verification to ensure a zero-risk smooth transition and millisecond-level rapid response during complex operating condition switching.

Benefits of technology

It achieves a zero-risk smooth transition and millisecond-level rapid response of temperature control strategy under varying operating conditions, solves the safety and real-time issues of temperature control strategy during complex operating condition switching, and improves the stability and response speed of coating quality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114147A_ABST
    Figure CN123114147A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of temperature control optimization in intelligent manufacturing equipment, and relates to a method for optimizing the temperature control of a fully automated coating machine in the coating process through multi-parameter collaboration. The method includes: acquiring and standardizing multi-source heterogeneous process parameters; constructing a decision tree classifier to output process semantic labels; establishing a semantic-objective mapping table and replacing the objective function; inputting the replaced objective function into a background pre-test for verification, and after passing the test, inputting it into the main control loop to generate real-time temperature control commands, while the old function is moved to a standby buffer; driving the heating mechanism based on the real-time temperature control commands and periodically acquiring new parameters to trigger reconstruction. This method solves the problem of safe dynamic reconstruction of the temperature control objective function under varying operating conditions, achieving low-risk smooth transition and rapid response.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of temperature control optimization technology for intelligent manufacturing equipment, and in particular to a method for optimizing the temperature control of a fully automatic laminating machine with multi-parameter coordination in the laminating process. Background Technology

[0002] The temperature control system of fully automatic laminating machines typically employs PID control, fuzzy control, or multi-objective optimization methods based on fixed weighting coefficients. Temperature regulation is achieved by setting a single control objective or assigning fixed weights to multiple objectives. In recent years, some solutions have introduced online adaptive mechanisms, dynamically adjusting the weights of multiple objectives using historical process data, or employing data-driven models such as neural networks and reinforcement learning to improve adaptability to complex operating conditions.

[0003] However, the above solutions still have shortcomings in practical applications. Temperature control strategies based on fixed weights or single objectives struggle to simultaneously optimize multiple performance indicators such as temperature response speed, energy efficiency, and coating quality stability when faced with dynamic changes in multi-source heterogeneous process parameters such as film thickness, lamination speed, ambient temperature and humidity, substrate type, and adhesive viscosity. This can easily lead to temperature control strategy instability or coating quality fluctuations. While solutions based on online learning or deep models possess adaptive capabilities in some scenarios, model inference relies on high computing power, making it difficult to meet the requirements of real-time response and low latency in industrial settings. Furthermore, the lack of a pre-verification mechanism for the safety of control commands during objective function switching poses a risk of temperature control overshoot or equipment safety issues due to sudden strategy changes. Summary of the Invention

[0004] This application provides a temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process, aiming to solve the problem that existing laminating machine temperature control methods are difficult to achieve safe dynamic reconstruction of the temperature control objective function under varying operating conditions.

[0005] This application provides a method for optimizing temperature control in a fully automated laminating machine with multi-parameter coordination during the laminating process, specifically including: Obtain the multi-source heterogeneous process parameters during the coating process, and standardize the multi-source heterogeneous process parameters to generate a standardized process parameter set.

[0006] A decision tree classifier is constructed based on industry expert knowledge and experimental data under typical working conditions. The standardized process parameter set is input into the decision tree classifier for rule-guided training, and process semantic labels are output.

[0007] A semantic-objective structure mapping table is established based on the mapping relationship between the process semantic tags and the pre-configured objective function templates. Each objective function template contains control parameters adapted to specific operating conditions.

[0008] Based on the process semantic label generated at the current moment, the semantic-target structure mapping table is queried, the corresponding target function template is called and its control parameters are loaded to complete the replacement of the temperature control target function.

[0009] The replaced temperature control objective function is input into the background simulation unit for verification. Within one control cycle, the generated temperature control command is verified to meet the safety constraints, and a simulation pass signal or failure signal is generated.

[0010] If the pre-simulation pass signal is received, the temperature control objective function is put into the main control loop to replace the old objective function to generate real-time temperature control instructions, and the old objective function is transferred to the standby buffer to support rollback operation.

[0011] Based on the real-time temperature control command, the heating actuator is driven to adjust the temperature, and new multi-source heterogeneous process parameters are acquired again in the next acquisition cycle to trigger a new round of process semantic label update and dynamic reconstruction of temperature control objective function.

[0012] The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process provided in this application has the following beneficial effects: By standardizing multi-source heterogeneous process parameters and inputting them into a decision tree classifier to output process semantic labels, a mapping relationship between semantic labels and objective function templates is established to achieve dynamic calling and replacement of temperature control objective functions. A background pre-simulation unit is introduced to perform safety verification on the replaced objective function. Only after the pre-simulation passes is the objective function put into the main control loop and the old objective function is transferred to the standby buffer to support rollback. This solves the technical problem that existing temperature control methods for laminating machines are difficult to achieve safe and dynamic reconstruction of temperature control objective functions under changing operating conditions. It realizes zero-risk smooth transition and millisecond-level fast response of temperature control strategies during complex operating condition switching processes. Attached Figure Description

[0013] Figure 1 This is the main flowchart of a fully automatic temperature control optimization method for a laminating machine that coordinates multiple parameters during the laminating process.

[0014] Figure 2 This is a flowchart illustrating the generation of a decision tree classifier model in a fully automated film coating machine temperature control optimization method that utilizes multiple parameters in the film coating process.

[0015] Figure 3 This is a calculation logic diagram of the constraint relaxation coefficient in a temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0018] like Figure 1 As shown, this application provides a temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process, specifically including: S1: Obtain multi-source heterogeneous process parameters during the coating process. The multi-source heterogeneous process parameters include film thickness, pressing speed, ambient temperature and humidity, substrate type and adhesive viscosity. Standardize the multi-source heterogeneous process parameters to generate a standardized process parameter set.

[0019] S2: Construct a decision tree classifier based on industry expert knowledge and typical working condition experimental data. Input the standardized process parameter set into the decision tree classifier for rule-guided training to output process semantic labels that represent common thermal response characteristics.

[0020] S3: Establish a semantic-objective structure mapping table based on the explicit mapping relationship between the process semantic tags and the pre-verified objective function templates, wherein each pre-verified objective function template contains a priority order of objective items, constraint relaxation coefficients, and physical boundary values ​​customized for a specific heat conduction dominant mechanism.

[0021] S4: Based on the process semantic label generated at the current moment, query the semantic-target structure mapping table to call the corresponding pre-verified target function template and load its pre-stored structural parameters and physical boundary values, thereby completing the complete replacement and parameter reloading of the temperature control target function to be executed.

[0022] S5: Input the temperature control objective function to be executed after parameter reloading into the background pre-executing unit of the double-buffered objective function execution mechanism, and verify whether the generated temperature control command meets the actuator safety limit and thermal inertia constraint within one control cycle to generate a pre-executing pass signal or a pre-executing failure signal.

[0023] S6: If the pre-run pass signal is received, the temperature control objective function to be executed is put into the main control loop to replace the old objective function to generate real-time temperature control instructions, and the old objective function is transferred to the standby buffer to support millisecond-level rollback operation.

[0024] S7: Drive the heating actuator to adjust the temperature based on the real-time temperature control command, and reacquire new multi-source heterogeneous process parameters in the next acquisition cycle to trigger a new round of process semantic label update and dynamic reconstruction of temperature control objective function.

[0025] Step S1: Obtain multi-source heterogeneous process parameters during the coating process. These parameters include film thickness, lamination speed, ambient temperature and humidity, substrate type, and adhesive viscosity. Standardize these parameters to generate a standardized process parameter set. Specifically, this includes: S1.1: Acquire the original analog signals or digital pulse signals output by the film thickness sensor, pressing speed encoder, ambient temperature and humidity transmitter, substrate type identification module and adhesive viscosity online detector deployed at the film coating production site, and perform analog-to-digital conversion and time-series alignment processing on the original analog signals or digital pulse signals to generate a multi-source heterogeneous original process parameter sequence containing timestamps.

[0026] Sub-step S1.1 aims to complete the raw signal acquisition, digital conversion, and high-precision timing alignment of multi-source heterogeneous process parameters, providing a time-synchronized data foundation for subsequent standardized processing.

[0027] The film thickness laser sensor, pressing roller rotary encoder, ambient temperature and humidity transmitter, substrate RFID identification module and adhesive viscosity online detector deployed on the film coating production line output 4-20mA analog current signal or TTL level digital pulse signal respectively.

[0028] The analog signals of each channel are sampled by a high-precision analog-to-digital converter (ADC), with a sampling frequency of 1kHz and a resolution of no less than 16 bits, converting continuous analog quantities into discrete digital quantities; the count value of the digital pulse signal is directly read.

[0029] A global hardware clock source is introduced to generate a unified time base, and a timestamp with microsecond precision is added to the data frames of each acquisition channel to form a raw data stream with time stamps.

[0030] To address the inherent differences in response delay among different sensors, a channel delay compensation model is established. Based on the physical installation location of each sensor and the length of the signal transmission path, a fixed time offset is calculated and applied to correct the timestamp of each data frame.

[0031] A linear interpolation algorithm is used to resample asynchronously arriving data points, mapping all channel data onto a unified time grid to eliminate data misalignment caused by sampling phase differences.

[0032] Through the above analog-to-digital conversion and timing alignment processing, the multi-source heterogeneous raw signals from the previous step are transformed into a multi-source heterogeneous raw process parameter sequence containing a unified timestamp, achieving strict synchronization of multi-dimensional data in the time dimension and ensuring the accuracy of subsequent feature extraction.

[0033] For example, the film thickness sensor outputs a 12.5mA current, which is converted into a value of 32768 by a 16-bit ADC, with a timestamp of t0+5μs; the pressure encoder generates 100 pulses at time t0. The system performs reverse compensation and alignment of the timestamp based on a preset 2ms thickness sensor delay and 0.5ms encoder delay. The final generated sequence is {(t0, thickness: 0.05mm, speed: 10m / min, temperature: 25℃, humidity: 40%, substrate: PET, viscosity: 1500cP)}, with the data timing error controlled within ±10μs, significantly improving the data consistency of multi-parameter collaborative modeling.

[0034] S1.2: Based on a predefined industrial communication protocol, the multi-source heterogeneous original process parameter sequence is parsed, and the values ​​of film thickness, pressing speed, ambient temperature, ambient humidity, substrate type code, and adhesive viscosity are extracted. Outlier removal and missing data imputation are performed on the extracted values ​​of film thickness, pressing speed, ambient temperature, ambient humidity, substrate type code, and adhesive viscosity to generate a set of effective process parameters after cleaning.

[0035] The system parses the multi-source heterogeneous raw process parameter sequence containing timestamps and locates the starting address and byte offset of each sensor data in the message based on the predefined ModbusTCP or OPCUA industrial communication protocol frame structure.

[0036] Extract the film thickness, pressing speed, ambient temperature, ambient humidity, substrate type code, and adhesive viscosity. Convert the extracted hexadecimal or binary raw data into decimal floating-point numbers or integer values.

[0037] An outlier detection model based on the 3σ principle is constructed. The mean and standard deviation of each process parameter within the sliding window are calculated to identify and remove outliers that exceed the range of mean plus or minus three times the standard deviation. For data gaps resulting from outlier removal, a linear interpolation algorithm is used to fill in the missing data. The process parameter values ​​at the time of the missing data are estimated using the slope of adjacent valid data points.

[0038] By removing outliers and performing linear interpolation, the original sequences containing noise and missing data are transformed into a continuous, complete set of effective process parameters that conform to physical laws after cleaning, thereby significantly improving data quality.

[0039] S1.3: Read the upper limit and lower limit values ​​of the physical range of each process parameter stored in the controller's local database, construct a normalized mapping matrix using the upper limit and lower limit values ​​of the physical range of each process parameter, and perform minimum-maximum normalization calculation on the film thickness, pressing speed, ambient temperature, ambient humidity and adhesive viscosity values ​​in the effective process parameter set after cleaning based on the normalized mapping matrix to generate a dimensionless continuous process feature vector.

[0040] The upper and lower limits of the physical ranges of each process parameter are read from the controller's local database, and a normalized mapping matrix is ​​constructed. This matrix defines a linear transformation relationship from the original physical quantity space to the dimensionless feature space, ensuring that parameters with different dimensions are comparable in subsequent decision tree classification.

[0041] The values ​​of membrane thickness, pressing speed, ambient temperature, ambient humidity, and adhesive viscosity are extracted from the effective process parameters set after cleaning and used as the input vector for the min-max normalization calculation. For continuous variables, corresponding range boundaries are set according to their physical meaning, with the membrane thickness ranging from 0.01 mm to 0.5 mm and the pressing speed ranging from 1 m / min to 50 m / min.

[0042] The minimum-maximum normalization algorithm is used to perform dimensionless processing on continuous process characteristics. For any process parameter x, its normalized value x' is calculated using the following formula: in, These are the original process parameter values. This parameter is the lower limit of the physical range set in the controller database. This parameter is the upper limit value of the physical range set in the controller database. It is a dimensionless value after normalization, and its range is strictly limited to the interval [0,1].

[0043] The above normalization operations were performed on the membrane thickness, pressing speed, ambient temperature, ambient humidity, and adhesive viscosity to generate corresponding dimensionless components. The calculation results of each component were arranged in a preset order and combined to form a continuous process feature vector. This vector eliminates the influence of dimensional differences on the selection of the decision tree splitting threshold, improving the classifier's efficiency in processing multi-source heterogeneous data.

[0044] By using min-max normalization, the effective process parameters after cleaning in the previous step are transformed into dimensionless continuous process feature vectors, realizing the standardized expression of multi-source heterogeneous data in a unified feature space, and providing a consistent data foundation for rule matching in subsequent decision tree classifiers.

[0045] For example, suppose the thickness of the membrane material after cleaning is 0.12 mm at the current moment, and the pressing speed is 20 m / min. The lower limit of the membrane thickness range in the controller database is 0.01 mm, and the upper limit is 0.5 mm; the lower limit of the pressing speed range is 1 m / min, and the upper limit is 50 m / min. Substituting into the normalization formula, the normalized value of the membrane thickness is (0.12-0.01) / (0.5-0.01)=0.2245, and the normalized value of the pressing speed is (20-1) / (50-1)=0.3878. Similarly, the normalized values ​​of ambient temperature, humidity, and viscosity are calculated, and the final generated continuous process feature vector is [0.2245, 0.3878, ...]. This vector is directly used for subsequent feature splicing and classification processing, which significantly improves the accuracy and response speed of state recognition.

[0046] S1.4: Call the substrate type thermal encoding mapping table preset in the control firmware, perform discrete value mapping transformation on the substrate type encoding in the set of effective process parameters after cleaning, and concatenate the transformed substrate type thermal encoding vector to the end of the dimensionless continuous process feature vector to generate a hybrid dimension process parameter vector that integrates continuous and discrete features.

[0047] The system invokes a substrate type one-hot encoding mapping table pre-set in the non-volatile storage area of ​​the control firmware. This mapping table defines the static correspondence between substrate physical properties and binary vectors.

[0048] Read the substrate type encoding field from the set of valid process parameters after cleaning. This field is a discrete integer identifier that represents different polymer material categories.

[0049] Based on the substrate type encoding retrieval mapping table, the corresponding multi-dimensional binary vector is obtained, where the vector dimension equals the preset total number of substrate types. Only the corresponding type index bit is assigned a value of 1, and the remaining bits are assigned a value of 0. This one-hot encoded vector consists of multiple binary bits, with each bit representing the activation state of a substrate type.

[0050] Obtain the dimensionless continuous process feature vector generated in the previous step. This vector contains normalized continuous values ​​such as film thickness and pressing speed.

[0051] Perform a vector concatenation operation to append the generated one-hot encoded vector to the end of the continuous process feature vector, maintaining a fixed arrangement order of continuous features first and discrete features last.

[0052] A hybrid-dimensional process parameter vector that integrates continuous and discrete features is generated, which eliminates the dimensional conflicts caused by discrete variables directly participating in numerical calculations.

[0053] By employing one-hot encoding mapping and vector concatenation, the results of the previous step are transformed into mixed-dimensional data with unified mathematical properties. This achieves the compatible fusion of discrete substrate attributes and continuous process parameters, providing standardized input for subsequent decision tree classification. For example, the preset substrate types include PET, BOPP, and PE, with n=3. When BOPP is identified as the substrate, it is encoded as 2, resulting in a one-hot vector [0,1,0]. If the continuous feature vector is [0.45,0.62,0.30,0.55,0.40], concatenation generates a mixed vector [0.45,0.62,0.30,0.55,0.40,0,1,0]. This mixed vector preserves the differences in thermal conductivity of the substrates while avoiding interference from integer encoding in distance calculations, significantly improving the accuracy of subsequent semantic classification.

[0054] S1.5: Perform fixed-length sliding window smoothing filtering on the hybrid-dimensional process parameter vector that integrates continuous and discrete features to eliminate high-frequency measurement noise interference and output the final standardized process parameter set, which serves as the direct input data source for the subsequent decision tree classifier.

[0055] Step S2: Construct a decision tree classifier based on industry expert knowledge and typical operating condition experimental data. Input the standardized process parameter set into the decision tree classifier for rule-guided training to output process semantic labels representing common thermal response characteristics. Specifically, this includes: S2.1: Based on industry expert knowledge and experimental data of typical working conditions, a rule base for a decision tree classifier is constructed. The threshold boundaries of the standardized process parameter set, such as film thickness, pressing speed, ambient temperature and humidity, substrate type and adhesive viscosity, are defined and processed to generate an initial rule base containing node splitting conditions and leaf node mapping relationships.

[0056] It should be noted that rule-guided training differs from traditional decision tree algorithms that rely entirely on data-driven node splitting. This method pre-constructs an initial rule base based on industry expert knowledge and experimental data under typical operating conditions, containing the mapping relationship between node splitting conditions and leaf nodes. Then, using this initial rule base as a constraint framework, feature space partitioning is performed on the standardized process parameter set. By calculating the information gain ratio, the optimal splitting attribute is selected within the candidate splitting attribute range defined by the rule base. This integrates expert experience into the decision tree construction process, enabling the generated decision tree classifier model to possess both data-driven adaptive capabilities and conform to the physical laws of process thermal response.

[0057] S2.1 Based on industry expert knowledge and experimental data of typical working conditions, a rule base for a decision tree classifier is constructed. The threshold boundaries of the standardized process parameter set, such as film thickness, pressing speed, ambient temperature and humidity, substrate type and adhesive viscosity, are defined and processed to generate an initial rule base containing node splitting conditions and leaf node mapping relationships.

[0058] The thermal response characteristic criteria stored in the coating process expert knowledge base are read, and the physical critical values ​​of film thickness, pressing speed, ambient temperature, ambient humidity and adhesive viscosity are extracted as the benchmark for setting the initial splitting threshold.

[0059] For continuous process parameters, the interquartile range method is used to calculate the distribution range of each parameter under typical operating conditions, determine the normal fluctuation range and abnormal deviation boundary, and construct the preliminary numerical segmentation points of the parameter dimension.

[0060] The ability of each parameter dimension to distinguish the categories of thermal response states is evaluated using the information entropy calculation formula, as follows: in, For dataset Information entropy This represents the total number of thermal response state categories. For the first The proportion of class samples in the dataset It represents a logarithmic operation with base 2.

[0061] Based on the principle of maximizing information gain, the top three key feature dimensions that contribute the most to the classification of hot states are selected and prioritized as the root node and secondary child node splitting attributes of the decision tree.

[0062] By combining the adhesive layer curing kinetic model, a coupling constraint condition between the adhesive layer viscosity change rate and temperature gradient is set, and the parameter combination that satisfies the coupling constraint is marked as a high confidence rule branch.

[0063] Discretized substrate type encoding is used as an independent branch condition. Independent sub-rule tree structures are established for different substrates with different thermal capacities to avoid feature space confusion.

[0064] By integrating the above-mentioned splitting threshold, preferred feature dimensions, and coupling constraints, an initial rule base containing the node splitting judgment logic and the mapping relationship between leaf node semantic labels is generated.

[0065] By using rule-guided training, the initial rule base is instantiated into the topology of a decision tree classifier, enabling rapid mapping from multi-dimensional process parameters to discrete semantic labels, which significantly improves the interpretability and real-time response capability of the classifier.

[0066] For example, in a PET substrate lamination scenario, the film thickness threshold is set to 0.05 mm, and the lamination speed threshold is set to 15 m / min. When the film thickness is detected to be less than 0.05 mm and the lamination speed is greater than 15 m / min, the root node of the decision tree is determined to be the "thin material, high speed" branch. Further considering the ambient temperature of 25℃ and the adhesive viscosity of 500 mPa·s, based on the information gain ratio calculation results, the adhesive viscosity is selected as the second-order splitting attribute. If the viscosity is lower than 600 mPa·s, the path points to the leaf node "thin material, high speed, low viscosity state," corresponding to the objective function template ID T01 (T01 represents the first pre-configured objective function template under this condition).

[0067] S2.2: The initial rule base is used to perform feature space partitioning on the standardized process parameter set. The optimal splitting attribute is determined by calculating the information gain ratio of each process parameter dimension, so as to generate a decision tree classifier model with a hierarchical logical structure.

[0068] like Figure 2 As shown, Figure 2 This is the flowchart for generating the decision tree classifier model in step S2.2, specifically: Receive the standardized process parameter set generated by S1.5. This dataset contains dimensionless continuous process feature vectors and discrete substrate thermal encoding vectors, which serve as the input sample space for constructing the decision tree model.

[0069] Iterate through each dimension of the standardized process parameter set, and for the sample subset under the current node, calculate the information gain ratio of each candidate splitting attribute under a specific threshold split. The information gain ratio is calculated based on the following formula: in, The information gain ratio of attribute A. For the information gain of attribute A, This is the splitting information for attribute A, used to penalize attributes with a large number of values ​​to avoid overfitting.

[0070] By comparing the information gain ratio values ​​of all process parameter dimensions, the process parameter with the highest information gain ratio is selected as the optimal splitting attribute for the current decision tree node. This optimal splitting attribute represents the key influencing factor with the highest distinguishing effect on thermal response characteristics under the current operating conditions.

[0071] Based on the predefined expert experience thresholds in the initial rule base, the specific splitting point value of the optimal splitting attribute is determined. If the optimal splitting attribute is a continuous variable such as membrane thickness or pressing speed, a bisection method is used to divide the sample space into two subspaces: one less than or equal to the splitting point and the other greater than the splitting point. If it is a discrete variable such as substrate type, a multi-branch path is generated based on the unique thermal encoding bit.

[0072] The feature space partitioning process described above is executed recursively. For each generated subspace, the information gain ratio is repeatedly calculated and the optimal splitting attribute is selected until a preset stopping condition is met. The stopping condition includes the node sample purity reaching a set threshold, the tree depth reaching the maximum limit, or the number of node samples falling below the minimum number of splits.

[0073] By hierarchically partitioning the feature space and selecting the optimal attribute, the complex nonlinear process parameter mapping relationship is transformed into a decision tree classifier model with a clear logical structure, thus achieving efficient semantic classification of the coating process status.

[0074] For example, using membrane thickness, pressing speed, and ambient temperature as input features, and the target variable as the semantic label of the process state, the information gain ratio of membrane thickness is calculated to be 0.45, pressing speed to 0.32, and ambient temperature to 0.18 at the root node. Membrane thickness is selected as the optimal splitting attribute, and the split point is set to 0.5 (normalized value). The left child node contains samples with a thickness ≤ 0.5, and the right child node contains samples with a thickness > 0.5. In the left child node, the information gain ratio of the remaining features is calculated again, and pressing speed is selected as the splitting attribute, with the split point set to 0.6. The final generated decision tree has a depth of 3, and the leaf nodes correspond to semantic labels such as "thin material high speed", "thin material low speed", "thick material high speed", and "thick material low speed". This model achieves a classification accuracy of 98% on the test set, with a single inference time of less than 1 millisecond, significantly improving the real-time performance and accuracy of state recognition.

[0075] S2.3: Based on the decision tree classifier model, the standardized process parameter set input in real time is processed by path deduction. The process parameter values ​​are matched level by level according to the node splitting conditions to generate a classification path sequence pointing to a specific leaf node.

[0076] The system receives a standardized set of process parameters after S1.5 sliding window smoothing filtering. This set of parameters includes dimensionless continuous process feature vectors and discrete features of substrate type with unique thermal encoding, which serve as the real-time input data source for the decision tree classifier.

[0077] The pre-built decision tree classifier model is invoked, and hierarchical traversal is performed starting from the root node. The membrane thickness value in the current standardized process parameter set is extracted and compared with the preset first layer splitting threshold.

[0078] If the membrane thickness is less than or equal to the first layer splitting threshold, the process is determined to enter the thin material branch path, and the left child node is activated; otherwise, if the membrane thickness is greater than the first layer splitting threshold, the process is determined to enter the thick material branch path, and the right child node is activated.

[0079] At the activated child node, continue to extract the pressing speed value and the ambient temperature and humidity value, and perform logical matching and judgment with the speed threshold range and humidity critical value in the second layer of splitting conditions stored in the node.

[0080] By using a step-by-step recursive matching method, the process parameter values ​​are passed down the decision tree branches until a leaf node that cannot be split further is reached, and the sequence of splitting conditions of all internal nodes traversed from the root node to that leaf node is recorded.

[0081] The recorded internal node splitting condition sequences are combined to generate a classification path sequence pointing to a specific leaf node. This sequence uniquely represents the landing point of the current mulching condition in the multidimensional feature space.

[0082] Through the above path deduction process, the standardized process parameter set of the previous step is transformed into a classification path sequence with clear physical meaning, realizing the accurate mapping of process state from continuous numerical space to discrete logical path, and providing a definite topological structure basis for subsequent semantic tag extraction.

[0083] For example, the decision tree root node splitting threshold is set to a membrane thickness of 0.15 mm. When the input membrane thickness is 0.12 mm, the pressing speed is 30 m / min, and the ambient humidity is 40%, the system first enters the left child node (thin material branch) because 0.12 mm ≤ 0.15 mm. At the left child node, because the pressing speed 30 m / min > 20 m / min and the ambient humidity 40% < 60%, it matches the leaf node corresponding to "thin material high-speed dry state". The generated classification path sequence is: root node (thickness ≤ 0.15) → left child node (speed > 20 AND humidity < 60) → leaf node "ID_03" (ID_03 is the unique identifier of this leaf node, used to index the corresponding semantic label). This path sequence directly points to the predefined "thin material high-speed dry state" semantic cluster, ensuring the accuracy and real-time performance of subsequent objective function calls.

[0084] S2.4: Perform semantic label extraction processing based on the pre-set mapping relationship of the leaf nodes at the end of the classification path sequence, convert the thermal response characteristic descriptor corresponding to the end point of the path into a process semantic label, and output a process semantic label representing the current coating condition.

[0085] Obtain the classification path sequence pointing to a specific leaf node generated by step S2.3. This sequence contains all the splitting conditions and corresponding process parameter threshold ranges from the root node to the leaf node.

[0086] The pre-defined thermal response characteristic descriptor data structure in the terminal leaf node of the classification path sequence is parsed, and physical semantic fields such as temperature hysteresis coefficient, heat conduction time constant and adhesive layer curing activation energy index are extracted.

[0087] The extracted physical semantic fields are discretized and mapped according to predefined semantic coding rules, where the temperature hysteresis coefficient is mapped to the "fast response" or "slow response" label, and the heat conduction time constant is mapped to the "shallow heating" or "deep penetration" label.

[0088] By combining the substrate type code and the adhesive viscosity grade, the initially mapped label combination is logically verified. If a high viscosity and fast response label are detected to coexist, the label is forcibly corrected to "steady state priority" according to the expert rule base to avoid the risk of bubbles.

[0089] The verified tags are combined and spliced ​​to generate process semantic tags with unique identifiers, such as "thin material - high speed - dry state - fast response" or "thick material - low speed - wet state - deep penetration".

[0090] By using semantic encoding and logical verification, the classification path results from the previous step are transformed into process semantic tags with clear thermodynamic meanings, thus achieving a precise semantic bridge between coating conditions and temperature control strategies.

[0091] For example, when the classification path determines that the film thickness is less than 0.05 mm and the lamination speed is greater than 10 m / min, the leaf node descriptor displays a heat conduction time constant of 0.8 s. The system maps this time constant to a "shallow heating" tag, and combines it with a "dry state" tag indicating an ambient temperature below 20°C to generate a semantic tag of "high-speed dry shallow heating of thin materials". This tag is directly indexed to the objective function template T1 in S3, ensuring that the temperature control system uses high proportional differential control to quickly respond to surface temperature changes, significantly improving the temperature tracking accuracy during high-speed lamination of thin materials.

[0092] Step S3: Establish a semantic-objective structure mapping table based on the explicit mapping relationship between the process semantic tags and the pre-validated objective function templates. Each pre-validated objective function template contains a priority order of objective items, constraint relaxation coefficients, and physical boundary values ​​customized for a specific heat conduction dominant mechanism. Specifically, this includes: S3.1: Based on the set of process semantic tags, perform heat conduction dominant mechanism analysis on each type of process combination with common thermal response characteristics to generate heat conduction dominant mechanism identifiers that characterize heat transfer characteristics under specific operating conditions.

[0093] It should be noted that the control parameters include at least the priority order of objective items, constraint relaxation coefficients, and physical boundary values. The priority order of objective items defines the weighting of temperature response speed, energy efficiency, and coating quality stability in the optimization calculation; the constraint relaxation coefficient adjusts the degree of transformation from hard constraints to soft constraints in the objective function to balance control accuracy and system robustness; and the physical boundary values ​​limit the actuator's safety limits and thermal inertia constraints, including the maximum allowable temperature rise slope and the upper limit of power output.

[0094] Receive process semantic tags and parse the key operating condition characteristic parameters contained in the tags, such as film thickness, pressing speed, and substrate type.

[0095] The system calls upon a pre-set knowledge base of heat conduction mechanisms to retrieve thermophysical property parameters that match the characteristics of the current operating conditions, including the thermal conductivity of the substrate, the specific heat capacity of the adhesive layer, and the interfacial contact thermal resistance coefficient.

[0096] A transient thermal equilibrium equation is constructed based on Fourier's law of heat conduction, with the membrane thickness as a characteristic length variable and the pressing speed as a convective heat transfer boundary condition variable substituted into the equation.

[0097] The Biot number is calculated to assess the relative magnitude of internal thermal conduction resistance and surface convective heat transfer resistance, and to determine whether the dominant heat conduction mode is a lumped parameter model or a distributed parameter model.

[0098] If the Biot number is less than 0.1, it is determined to be a lumped parameter dominant mechanism, and the identifier "Lumped_Thermal_Dominant" is generated; if the Biot number is greater than 0.1, it is determined to be a distributed parameter dominant mechanism, and the Fourier number is further calculated to distinguish the transient response stage, and the identifier "Transient_Conduction_Fast" or "Steady_State_Slow" is generated according to the range of Fourier values.

[0099] The identifier is modified by combining the influence of substrate type on thermal inertia. For example, a deformation heat generation term is introduced for high-elastic substrates to generate a composite identifier "Deformation_Coupled_Thermal".

[0100] Through the above-mentioned thermodynamic parameter calculation and mechanism discrimination logic, the abstract semantic tags are transformed into concrete identifiers of the dominant heat conduction mechanism, thereby quantifying the physical basis for the selection of the temperature control objective function structure.

[0101] S3.2: Based on the heat conduction dominant mechanism identifier, design a targeted target item priority order configuration scheme using industry expert knowledge and typical working condition experimental data to generate a target item priority order sequence that includes weighted sorting of temperature response speed, energy consumption efficiency and coating quality stability.

[0102] Receive the thermal conduction dominant mechanism identifier generated by S3.1, which characterizes the physical properties of heat transfer under the current coating conditions.

[0103] Based on the predefined process-control mapping rules in the industry expert knowledge base, the control requirement priority matrix corresponding to the heat conduction dominant mechanism identifier is parsed.

[0104] For mechanisms dominated by rapid heating, the temperature response speed is set as the highest priority, energy efficiency as the second priority, and film quality stability as the basic constraint.

[0105] For mechanisms dominated by maintaining thermal equilibrium, the coating quality stability is set as the highest priority, the temperature response speed as the second priority, and the energy efficiency as the lowest priority.

[0106] A multi-objective weighted evaluation function is constructed, which linearly combines the normalized performance indicators of each objective item with their corresponding priority weights. The comprehensive objective function value is determined by the weighted sum of temperature response speed, energy efficiency, and coating quality stability, with the sum of the weight coefficients of each item being 1.

[0107] Based on the preset priority level, the weight coefficient values ​​are dynamically allocated. The weight coefficient value range for high priority is set to 0.6 to 0.8, for medium priority it is 0.2 to 0.4, and for low priority it is 0.0 to 0.2.

[0108] Generate a priority order configuration scheme for target items containing a sequence of specific weight values. This sequence directly determines the search direction of subsequent optimization algorithms.

[0109] By using the above-mentioned dynamic weight allocation method, the abstract heat conduction dominant mechanism is transformed into a quantifiable objective function weight sequence, thereby achieving precise adaptation of the temperature control strategy to the thermal characteristics of specific operating conditions.

[0110] For example, when the heat conduction dominant mechanism identifier indicates "thin material high-speed dry state," the system determines that heat dissipation is rapid and requires a quick attainment of the curing temperature. In this case, temperature response speed is given the highest priority, with a weighting coefficient w1 set to 0.7; energy efficiency is the secondary priority, with a weighting coefficient w2 set to 0.2; and film quality stability is the basic constraint, with a weighting coefficient w3 set to 0.1. Since the temperature response speed item has the highest weight, the optimization algorithm prioritizes minimizing the temperature response speed cost function, thereby driving the controller to output a larger heating power to shorten the heating time. If the identifier indicates "thick material low-speed wet state," the film quality stability weight w3 is increased to 0.7, and the temperature response speed weight w1 is reduced to 0.1, ensuring the adhesive layer is fully cured without bubbles, significantly improving the film yield.

[0111] S3.3: Based on the priority sequence of the target items, combined with the material deformation feedback requirements and the adhesive layer curing gradient control requirements, calculate the constraint relaxation coefficient values ​​that are adapted to the current heat conduction dominant mechanism, so as to generate a set of constraint relaxation coefficients for adjusting the constraint boundary of the objective function.

[0112] like Figure 3 As shown, Figure 3 This is the logic diagram for calculating the constraint relaxation coefficient in step S3.3, specifically: The priority sequence of target items generated by S3.2 is received and combined with the material deformation feedback requirements and adhesive layer curing gradient control requirements under the current process state as the input benchmark for calculating the constraint relaxation coefficient.

[0113] Extract the substrate elastic modulus data and adhesive layer rheological property curves pre-stored locally on the controller, and determine the deformation tolerance threshold and curing reaction rate constant based on the heat conduction dominant mechanism corresponding to the current semantic tag.

[0114] A constraint relaxation coefficient calculation model based on deviation tolerance is constructed. This model aims to quantify the degree of transformation from hard constraints to soft constraints in the objective function, in order to balance control accuracy and system robustness. The relaxation coefficient of the temperature response constraint is obtained by adding 1 to the ratio of the deviation between the current measured temperature and the target set temperature relative to the maximum allowable temperature deviation threshold.

[0115] To address the material deformation feedback requirement, a deformation-sensitive factor is introduced. When high-speed operation of a highly elastic substrate or thin material is detected, the relaxation coefficient of the deformation constraint term is increased, allowing the temperature to fluctuate within a small range to avoid tensile deformation of the substrate caused by local overheating.

[0116] To address the requirements for controlling the curing gradient of the adhesive layer, the curing reaction rate at the current temperature is calculated based on the Arrhenius equation. If the reaction rate is lower than the critical value, the relaxation coefficient of the energy consumption constraint is reduced, and the heating power is forcibly increased to accelerate the curing process.

[0117] The calculated relaxation coefficients of each dimension are integrated into a vector form to form a set of constraint relaxation coefficients that include three dimensions: temperature response, deformation tolerance, and curing efficiency.

[0118] By using the above processing method, the target priority of the previous step is transformed into specific mathematical constraint boundary adjustment parameters, realizing dynamic adaptation of the objective function constraint rigidity under different heat conduction mechanisms, and ensuring the optimal trade-off of the temperature control system when multiple objectives conflict.

[0119] For example, under the semantic tag "thick material, low-speed, wet state," the system reads that the substrate is a PET film with a thickness of 0.15 mm and an ambient humidity of 65%. At this point, the target priority sequence is "interface thermal balance > curing gradient > energy consumption." When calculating the temperature relaxation coefficient, with the current temperature at 120℃, the target at 130℃, and a maximum allowable deviation of 5℃, the result is λ = 1 + |120 ​​- 130| / 5 = 3.0, indicating that the temperature constraint is significantly relaxed to prioritize heat penetration. Simultaneously, due to the high humidity, the adhesive layer curing rate decreases, so the system sets the curing constraint relaxation coefficient to 0.8, tightening the constraint to enhance heating. The final generated set of constraint relaxation coefficients is {λtemp: 3.0, λdeform: 1.2, λcure: 0.8}, which is loaded into the objective function template. This significantly improves the coating adhesion of thick materials in a humid environment, avoiding substrate wrinkling caused by excessive pursuit of rapid temperature response.

[0120] S3.4: Based on the set of constraint relaxation coefficients and the maximum allowable temperature rise slope limit, determine the physical boundary value parameters of each objective function template to generate a physical boundary value array that limits the actuator safety amplitude and thermal inertia constraint.

[0121] Obtain the set of constraint relaxation coefficients generated in step S3.3, and extract physical limit parameters such as the maximum allowable temperature rise slope threshold, power output upper limit, and temperature overshoot tolerance by combining them with the hardware specifications of the heating actuator of the coating equipment.

[0122] Based on the identifier of the dominant heat conduction mechanism, the pre-stored database of substrate thermal property parameters is called to read the specific heat capacity, thermal conductivity and density data of the substrate under the current working conditions, and a dynamic boundary calculation model reflecting the thermal inertia of the material is constructed.

[0123] Using the maximum permissible temperature rise slope constraint, a hard constraint boundary value for the temperature change rate is calculated based on the actuator power limit, heat conversion efficiency, heated area mass, and substrate specific heat capacity. The calculated maximum permissible temperature rise slope is then linearly weighted with the constraint relaxation coefficient to generate a temperature gradient safety boundary value adapted to the current semantic tag.

[0124] Based on the adhesive layer curing kinetic model, the minimum maintenance time and maximum tolerance temperature of the adhesive layer interface temperature are set. Combined with the aforementioned temperature gradient safety boundary value, the upper and lower limit threshold array of the actuator duty cycle is derived.

[0125] The derived physical boundary value array is discretized and mapped to an integer instruction code recognizable by the controller register, forming a physical boundary value array containing the heating slope limit, power duty cycle range, and steady-state error tolerance band.

[0126] Through the above chain-like derivation process, the constraint relaxation coefficients of the previous step are transformed into an array of physical boundary values ​​that limit the actuator's safety amplitude and thermal inertia constraints, thereby achieving an inherently safe adaptation of the temperature control objective function under dynamic operating conditions.

[0127] For example, for the semantic tag "thick material, low-speed wet state", the specific heat capacity of the substrate (c) is read as 1.5 J / (g·℃), the heated mass (m) is 200 g, the actuator power limit is 1000 W, and the heat conversion efficiency is 0.8. The maximum allowable temperature rise slope is calculated to be 2.67℃ / s. Combining the constraint relaxation coefficient of 0.9 output from S3.3, the corrected temperature gradient safety boundary value is 2.4℃ / s. Based on this, the actuator duty cycle is set to an upper limit of 85% and a lower limit of 10%, and the maximum withstand temperature is locked at 180℃. This array of physical boundary values ​​ensures that during the thick material coating process, the heating rate meets the curing requirements of the adhesive layer while avoiding substrate deformation or adhesive layer blistering due to excessively rapid heating, significantly improving the stability of the coating quality.

[0128] S3.5: Using the process semantic tag as the index key, associate it with the corresponding target item priority order sequence, constraint relaxation coefficient set and physical boundary value array to construct a complete mapping record, so as to generate a semantic-target structure mapping table that supports millisecond-level query calls.

[0129] Obtain the process semantic tags, target item priority sequence, constraint relaxation coefficient set, and physical boundary value array generated by steps S3.1 to S3.4, as the basic data elements for constructing the mapping table.

[0130] A hash digest calculation is performed on the process semantic tags, and the SHA-256 algorithm is used to generate a fixed-length unique index key to ensure the uniformity of the address distribution of different working condition tags in the storage space and avoid query delays caused by hash collisions.

[0131] The priority order of the target items is serialized into a one-dimensional integer array, the set of constraint relaxation coefficients is encapsulated into a floating-point matrix, the physical boundary value array is converted into a double-precision threshold list, and all are packaged into a standard binary data block structure.

[0132] Establish a key-value pair storage structure, using the generated unique index key as the key and the binary data block containing serialized parameter data as the value. Construct a memory-resident semantic-target structure mapping table and pre-allocate contiguous memory space to optimize cache hit rate.

[0133] Perform integrity checks on the constructed mapping table to verify the correctness of the pointer association between each index key and the corresponding data block, ensuring the atomicity and consistency of data reading in high-frequency query scenarios at the millisecond level.

[0134] By using hash index mapping and binary serialization encapsulation, the scattered structured parameters from the previous step are transformed into a memory mapping table that supports fast retrieval with O(1) time complexity, thus achieving the expected technical effect of instant loading and zero-latency switching of temperature control objective function parameters.

[0135] For example, for the semantic tag "thin material high-speed dry state", the first 8 bits of its SHA-256 hash value are calculated as the index key 0xA1B2C3D4. The priority sequence [1,3,2] (representing response speed first, quality stability second, and energy consumption last) is serialized into a byte array. The constraint relaxation coefficient is set to 0.95, and the maximum temperature rise slope in the physical boundary values ​​is set to 5℃ / s. The system stores the binary block containing the above parameters, pointed to by the index key 0xA1B2C3D4, in the controller's SRAM. When the real-time operating condition matches this tag, the controller directly reads the data block through the index key, taking less than 1 microsecond, significantly improving the dynamic reconfiguration efficiency.

[0136] Step S4: Based on the process semantic label generated at the current moment, query the semantic-target structure mapping table to call the corresponding pre-verified target function template and load its pre-stored structural parameters and physical boundary values, thereby completing the complete replacement and parameter reloading of the temperature control target function to be executed. Specifically, this includes: S4.1: Obtain the process semantic label generated at the current moment as the index key, and perform a fast retrieval of the semantic-target structure mapping table based on the hash lookup algorithm to locate the address pointer of the pre-verified target function template that uniquely matches the process semantic label.

[0137] S4.2: Read the pre-verified objective function template storage area pointed to by the address pointer, extract the objective item priority order data sequence, constraint relaxation coefficient matrix and physical boundary value threshold set contained therein, so as to generate the structural parameter set to be loaded and the physical boundary value set to be loaded.

[0138] Receive the pre-verification target function template address pointer output from step S4.1. This pointer points to the parameter storage block corresponding to a specific semantic tag in the controller's non-volatile memory.

[0139] Based on the address pointer, the starting position of the storage block is located, and a fixed-length target item priority order data sequence is read. This sequence defines the hierarchical weight relationship between temperature response speed, energy efficiency, and coating quality stability in the optimization calculation.

[0140] The priority order data sequence of the objectives is analyzed, the normalized weight coefficients and penalty factors of each objective are extracted, and the initial vector of the structure parameter set to be loaded is constructed to ensure the logical consistency of multi-objective collaborative optimization.

[0141] Continue reading the constraint relaxation coefficient matrix from the storage block offset address. This matrix contains soft constraint boundary values ​​that are dynamically adjusted for the current dominant heat conduction mechanism, used to balance control accuracy and system robustness.

[0142] The constraint relaxation coefficient matrix is ​​validated for dimension and converted for format, mapped to a floating-point array, and then incorporated into the set of structural parameters to be loaded, in order to adapt to the numerical calculation accuracy requirements of the background pre-simulation unit.

[0143] Read the set of physical boundary value thresholds at the end of the storage block, including hard safety limit parameters such as the maximum allowable temperature rise rate, actuator power limit, and minimum steady-state duration.

[0144] The set of physical boundary value thresholds is encapsulated into an immutable constant structure to generate the set of physical boundary values ​​to be loaded, thus preventing security risks caused by parameter overflow at runtime.

[0145] The set of structural parameters to be loaded and the set of physical boundary values ​​to be loaded are merged to form a complete temperature control objective function configuration data package, which is then prepared to be written into the background register of the double-buffered mechanism.

[0146] Through the chained reading and parsing process described above, the discrete semantic tags are transformed into executable mathematical model parameters, enabling the precise instantiation of the temperature control strategy from abstract semantics to specific control laws.

[0147] For example, when the semantic tag is "thick material, low-speed wet state", the address pointer points to the 0x08A4 memory area (0x08A4 is a hexadecimal address in the controller memory). The priority sequence [0.3, 0.2, 0.5] is read, indicating that quality stability has the highest weight. The diagonal element of the constraint relaxation coefficient matrix is ​​read as 1.2, relaxing temperature fluctuation constraints to accommodate the thermal inertia of thick materials. The maximum temperature rise slope in the physical boundary values ​​is read as 2.5℃ / s. The final generated configuration data package contains the above parameters, ensuring that the uniformity of adhesive layer curing is prioritized during thick material coating, while limiting heating power to prevent overheating, significantly improving coating bonding strength and process stability.

[0148] S4.3: Write the set of structural parameters to be loaded and the set of physical boundary values ​​to be loaded into the background pre-executing unit register in the double-buffered objective function execution mechanism, overwriting the original old structural parameters and old physical boundary values, so as to instantiate and generate the complete mathematical model architecture of the temperature control objective function to be executed.

[0149] The set of structural parameters and physical boundary values ​​extracted in step S4.2 are obtained as the input data source for the background pre-simulation unit register write operation.

[0150] The priority sequence of target items in the set of parameters to be loaded is analyzed and mapped to the control law weight index address of the background register in the double buffer mechanism, thus establishing the primary and secondary relationship of the optimization dimensions of the new objective function.

[0151] The constraint relaxation coefficient set is read, converted into fixed-point format according to the floating-point precision standard, and written into the constraint boundary register group of the background pre-simulation unit to define the dynamic balance range between temperature response speed and energy efficiency.

[0152] The maximum allowable temperature rise slope threshold and the power output upper limit threshold in the physical boundary value array are loaded into the reference level register of the actuator safety limiting comparator to construct the hardware-level instruction truncation protection logic.

[0153] The instantiation engine is invoked to reconstruct the mathematical expression architecture of the temperature control objective function in the background memory space based on the written weight index, constraint boundary and safety limit parameters. This generates an instance object of the objective function containing a complete coefficient matrix and calculates the instantaneous evaluation value of the reconstructed objective function to ensure the integrity of the mathematical model architecture.

[0154] The generated target function instance object is bound to a memory address, and its pointer is set to the core entry point of the background pre-simulation unit, thus completing the transformation from static parameters to an executable mathematical model architecture.

[0155] By using the above register writing and instantiation reconstruction process, the parameter set of the previous step is transformed into a temperature control objective function model with complete computational logic to be executed, thereby achieving millisecond-level non-intrusive reloading and instantaneous preparation of the objective function structure.

[0156] S4.4: Perform memory integrity verification and logical consistency verification on the complete mathematical model architecture of the temperature control objective function to be executed, confirm that there is no conflict in the priority order data sequence of the objective items and that the set of physical boundary value thresholds is within a safe range, so as to generate the parameter reload completion flag.

[0157] Read the instantiated mathematical model architecture data of the temperature control objective function to be executed from the background pre-simulation unit register, and extract the priority sequence of objective items, the set of constraint relaxation coefficients, and the array of physical boundary values ​​as verification objects.

[0158] Perform logical mutual exclusion detection on the priority order data sequence of target items, traverse the weight index of each target in the sequence for temperature response speed, energy efficiency and coating quality stability, confirm that there are no cases where the same optimization dimension is assigned multiple conflicting priorities, and generate priority logic compliance flags.

[0159] Based on the set of physical boundary value thresholds, a safety domain judgment matrix is ​​constructed, and parameters such as the maximum allowable temperature rise slope, the upper limit of power output, and the minimum steady-state maintenance time are compared and calculated with the hardware execution limit value of the controller.

[0160] If any physical boundary value exceeds the hardware safety limit or falls below the minimum thermal inertia constraint threshold, the template is marked as illegal and the loading process is blocked to prevent the risk of thermal shock caused by parameter out-of-bounds.

[0161] The set of constraint relaxation coefficients is checked for validity in terms of numerical range. It is confirmed that all relaxation coefficients are within the preset open interval (0,1) and satisfy the normalized sum constraint, thus ensuring the convexity and convergence of the objective function during optimization.

[0162] Based on the comprehensive priority logic compliance flags, the legality determination results of physical boundary values, and the validity check results of constraint relaxation coefficients, perform logical AND operations.

[0163] When all verification items pass, the parameter reload completion flag is set, triggering subsequent master control loop switching instructions; if any verification fails, an exception lock signal is generated and the old objective function running state is retained.

[0164] By using memory integrity verification and logical consistency verification mechanisms, the static parameters of the temperature control objective function to be executed are transformed into a secure and certified executable control law configuration, thereby achieving zero-fault switching and inherent system safety during the dynamic reconstruction process of the objective function.

[0165] For example, for the objective function template corresponding to the semantic tag "thick material, low-speed wet state", the system extracts its physical boundary value array, where the maximum allowable temperature rise slope is set to 2.5℃ / s. The controller reads the hardware limit value as 5.0℃ / s, and the comparison shows that 2.5 ≤ 5.0 is valid. The constraint relaxation coefficient set is {0.8, 0.1, 0.1}, all within the interval (0,1) and summing to 1.0. The priority sequence shows that the adhesive layer curing gradient control takes precedence over energy efficiency, with no index conflicts. All checks pass, a high-level parameter reload completion flag is generated, allowing entry into step S4.5 to activate the new control law.

[0166] S4.5: The response parameter overload completion flag activates the complete mathematical model architecture of the temperature control objective function to be executed that has been instantiated in the background pre-play unit into an effective control law of the front-end main control loop. At the same time, the old objective function in the original front-end main control loop is archived to the standby buffer to complete the whole replacement operation of the temperature control objective function.

[0167] The response parameter reload completion flag triggers the state switching logic of the double buffer mechanism. The complete mathematical model architecture data block of the temperature control objective function to be executed, which has already been instantiated in the background pre-executing unit, is read.

[0168] Copy the data block to the real-time calculation register group of the front-end master loop, overwriting the original old objective function pointer and parameter context. Perform a context snapshot operation on the old objective function in the original front-end master loop to extract its current integral term state, differential term history value, and constraint boundary cache.

[0169] The extracted snapshot data is encapsulated into a standby buffer object and assigned a unique version identifier and timestamp. The state machine flags of the master control loop are updated, and the control law source is pointed to the newly activated temperature control objective function to be executed. The lock on the background pre-render unit is released, and the old objective function is marked as rollbackable.

[0170] Through the above atomic switching process, the parameter reload result of the previous step is transformed into an effective control law for the front-end main control loop, realizing the seamless replacement of the temperature control strategy and the millisecond-level fault rollback capability.

[0171] For example, when the process semantic label changes from "thin material high-speed dry state" to "thick material low-speed wet state", the system detects that the parameter reload completion flag generated by S4.4 is high. The controller immediately locks the new objective function template at the background pre-playback unit address 0x4000-0x4FFF, which contains piecewise constraint coefficients for the thermal inertia of thick materials. The system directly copies this memory block data to the core calculation area 0x2000-0x2FFF of the front-end main control loop via the DMA (Direct Memory Access) channel, taking less than 5 microseconds.

[0172] Step S5: The temperature control objective function to be executed, after parameter reloading, is input into the background pre-executing unit of the double-buffered objective function execution mechanism. Within one control cycle, the generated temperature control command is verified to ensure it meets the actuator safety limits and thermal inertia constraints, thereby generating a pre-executing pass signal or a pre-executing failure signal. Specifically, this includes: S5.1: Obtain the temperature control objective function to be executed after parameter reloading and the process status feedback data at the current moment. Based on the double-buffered objective function execution mechanism, the background pre-simulation unit constructs a virtual control closed-loop environment and loads the temperature control objective function to be executed into the virtual control closed-loop environment to initialize the pre-simulation calculation context.

[0173] It should be noted that safety constraints include at least actuator safety limit constraints and thermal inertia constraints. Actuator safety limit constraints mean that the rate of change of amplitude in the simulated temperature control command sequence must not exceed the maximum allowable temperature rise slope threshold, and the power output must not exceed the upper limit threshold of power output; thermal inertia constraints mean that the temperature field evolution trend caused by the simulated temperature control command sequence must not exceed the thermal inertia safety boundary determined by the heat capacity of the substrate and the curing kinetics of the adhesive layer.

[0174] Read the instantiated mathematical model architecture of the temperature control objective function to be executed from the background pre-executing unit register, and extract the complete parameter set including the priority sequence of objective items, the constraint relaxation coefficient matrix, and the physical boundary value threshold.

[0175] Simultaneously collect real-time temperature distribution data, online detection value of adhesive viscosity, and substrate type code on the surface of the coating pressing roller at the current moment to construct an initial process feedback vector reflecting the current thermodynamic state.

[0176] Based on the isolation characteristics of the double buffer mechanism, a virtual control closed-loop environment is instantiated in the independent memory space of the background pre-simulation unit. This environment maps the transfer function and thermal inertia delay characteristics of the real heating actuator.

[0177] The objective function for temperature control to be executed is loaded into the computational core of the virtual control closed-loop environment, and a data path connection is established between the output of the objective function and the input of the virtual actuator.

[0178] Initialize the pre-simulation calculation context, set the pre-simulation time window length to a single control cycle, and inject the current process feedback vector as the initial state variable of the virtual environment.

[0179] Configure the solver parameters for the heat conduction differential equation in the virtual environment to match the dominant heat conduction mechanism identifier corresponding to the current semantic tag, thus ensuring the accuracy of the pre-simulated physical model.

[0180] Through the above processing method, the parameter overload result of the previous step is transformed into a virtual pre-simulation environment with physical consistency, realizing the initial deployment of the target function to be executed in a safe and isolated space, and providing a reliable computing basis for subsequent instruction derivation.

[0181] For example, for the semantic tag "thick material, low-speed wet state," the background pre-simulation unit loads the corresponding piecewise objective function template. It reads the current pressing roller surface temperature as 180℃, adhesive viscosity as 3500 mPa·s, and substrate as PET. A virtual environment is constructed in an independent memory space, mapping a solid-state relay response delay of 50ms and a ceramic heating rod thermal time constant of 12s. The objective function priority sequence (interface thermal balance priority) and constraint relaxation coefficient of 0.85 are loaded into the calculation core. The pre-simulation time window is set to 200ms, and the initial temperature field vector is injected. The solver is configured to use an implicit difference scheme to match the thermal conductivity characteristics of thick materials. After context initialization, the virtual environment is in a ready state, ensuring that the subsequently generated simulated temperature control commands can accurately reflect the thermal response trend under this condition, significantly improving the confidence of the pre-simulation verification.

[0182] S5.2: Utilize the initialized pre-calculation context to perform forward inference calculations within a complete control cycle, generate a simulated temperature control command sequence based on the mathematical structure of the temperature control objective function to be executed, and use the simulated temperature control command sequence as the basic input data for subsequent constraint verification.

[0183] Read the mathematical framework of the temperature control objective function to be executed that has been instantiated in the background pre-simulation unit register, and extract the process status feedback data in the current control cycle as the initial conditions for deduction.

[0184] A virtual time step sequence is constructed to map the optimization variable space of the temperature control objective function to be executed to the discretized time axis, and the sampling interval is set to be consistent with the control cycle.

[0185] Based on the temperature response rate term in the objective function, and combined with the thermal conduction differential equation of the coating pressing roller, the theoretical temperature rise trajectory under the current heating power assumption is calculated.

[0186] The predicted temperature value at the k-th time step is calculated using the following formula: in, For the first Predicting temperature step by step, To simulate heating power, To control the cycle duration, For specific heat capacity, For equivalent thermal mass, For heat dissipation coefficient, The ambient temperature.

[0187] Based on the energy efficiency term constraint in the objective function, the simulated heating power sequence is iteratively optimized to ensure that the power output is minimized while satisfying the temperature rise trajectory.

[0188] A curing kinetic model for the adhesive layer is introduced, and the predicted temperature sequence is substituted into the Arrhenius equation to calculate the evolution trend of the adhesive layer viscosity over time, thereby evaluating the stability index of the coating quality.

[0189] The generated simulated temperature control command sequence, which includes heating power duty cycle, heating slope, and expected temperature field, is encapsulated into a standard data structure and stored in the output buffer of the pre-simulation unit.

[0190] By using forward extrapolation calculation, the static parameters of the temperature control objective function to be executed are transformed into dynamic simulated temperature control command sequence data, thereby achieving the expected technical effect of pre-verification of potential risks and safety assessment of the new control strategy.

[0191] For example, for the semantic tag "thick material, low-speed wet state", the control period Δt is set to 200ms, the equivalent thermal mass m is 5.2kg, the specific heat capacity C is 850J / (kg·℃), and the heat dissipation coefficient K is 0.03W / ℃. The initial temperature T(0) is 120℃, and the ambient temperature T env The target temperature is 25℃. The objective function requires reaching 160℃ within the 10th cycle with minimal power fluctuation. Through deduction, the simulated power P(1) in the first step is calculated to be 1200W. Substituting this into the formula, the predicted temperature T(1) is calculated as: T(1) = 120 + 1200 × 0.2 / (850 × 5.2) - 0.03 × (120 - 25) × 0.2 = 119.5℃. After 10 consecutive iterations, a simulated command sequence containing 10 power values ​​is generated. The final predicted temperature stabilizes at 159.8℃, and the viscosity change rate of the adhesive layer meets the curing gradient requirements. A complete simulated temperature control command sequence is output for subsequent verification.

[0192] S5.3: Perform actuator safety limiting constraint verification processing on the generated simulated temperature control command sequence. Compare and analyze the amplitude change rate in the simulated temperature control command sequence with the maximum allowable temperature rise slope threshold and the power output upper limit threshold to generate the limiting compliance judgment result.

[0193] Obtain the simulated temperature control command sequence generated by the background simulation unit. This sequence contains the predicted heating power duty cycle and temperature rise slope data within the current control cycle.

[0194] The timestamps and corresponding power amplitudes in the simulated temperature control command sequence are analyzed to construct a discretized power-time mapping array, which serves as the basic dataset for amplitude limiting verification.

[0195] Read the physical boundary value array loaded in the semantic-target structure mapping table, extract the maximum allowable temperature rise slope threshold and the power output upper limit threshold, and set them as static constraint benchmarks.

[0196] Perform a difference operation on the power-time mapping array to calculate the power change within adjacent time steps, and derive the instantaneous temperature rise slope value by combining the sampling period.

[0197] The calculated instantaneous temperature rise slope value is compared point by point with the maximum allowable temperature rise slope threshold, and all abnormal time points exceeding the threshold are marked.

[0198] The power amplitude in the simulated temperature control command sequence is compared with the upper limit threshold of power output to identify power peaks that exceed the hardware safety limit.

[0199] By combining the results of abnormal time point marking and power peak identification, a logical OR operation is performed. If any violation exists, the amplitude limiting is determined to be non-compliant; otherwise, the amplitude limiting is determined to be compliant.

[0200] Generate a limit compliance determination result containing a compliance status identifier and a violation details index, which serves as the input condition for subsequent thermal inertia constraint matching degree calculation.

[0201] Through the above chain derivation process, the simulated temperature control command of the previous step is transformed into a quantified limit compliance judgment index, which realizes the accurate verification of the physical safety boundary of the actuator and effectively prevents the risk of hardware overload caused by sudden changes in the objective function.

[0202] For example, the control period Δt is set to 200ms, the maximum allowable temperature rise slope threshold is 5℃ / s, and the power output upper limit threshold is 90%. In the simulated temperature control command sequence, the power jumps from 40% to 60% at the 5th time point. The calculated instantaneous temperature rise slope is (60-40) / 0.2=100% / s, which far exceeds the threshold of 5℃ / s (equivalent power change rate). The system immediately marks this point as abnormal, generates a limit non-compliance judgment result, triggers the abnormal locking mechanism, prevents the objective function from being put into the main control loop, and ensures the safe operation of the equipment.

[0203] S5.4: Based on the amplitude limit compliance judgment result, perform thermal inertia constraint matching degree calculation on the simulated temperature control command sequence, and combine the substrate heat capacity parameter and the adhesive layer curing kinetic model to evaluate whether the temperature field evolution trend caused by the simulated temperature control command sequence exceeds the thermal inertia safety boundary, so as to generate thermal inertia stability assessment index.

[0204] Based on the compliance determination results, the power change rate data in the simulated temperature control command sequence and the substrate thermophysical parameters of the current coating process section are extracted to construct a thermal inertia constraint verification model. The specific heat capacity, density, and adhesive layer curing activation energy parameters of the substrate stored locally in the controller are read, and combined with the current contact area of ​​the pressing roller, the equivalent heat capacity coefficient and heat conduction time constant are calculated. The simulated temperature control command sequence is mapped to a heating power input vector with discrete time steps, and substituted into the one-dimensional unsteady-state heat conduction equation to deduce the temperature field evolution. The finite difference method is used to numerically solve the heat conduction equation, calculating the predicted temperature distribution curve of the adhesive layer interface in the next control cycle. According to the Arrhenius equation of adhesive layer curing kinetics, the rate of change of adhesive layer curing degree is calculated using the predicted temperature distribution curve to assess whether there is a risk of uneven curing or bubble defects due to excessively rapid heating. A thermal inertia safety boundary threshold is set, including the maximum allowable temperature gradient and the minimum curing stabilization time. The predicted temperature distribution curve and the rate of change of curing degree are compared with the safety boundary threshold. If the predicted temperature gradient exceeds the maximum allowable value or the rate of change in curing degree exceeds the stable range, a thermal inertia mismatch is determined, and a high thermal inertia risk marker is generated; otherwise, a thermal inertia safety marker is generated. Through the above-mentioned coupled deduction of heat conduction and curing kinetics, the simulation command is transformed into a quantitative thermal inertia stability assessment index, realizing the safety verification of the new objective function at the physical thermal response level.

[0205] S5.5: Execute logic and operations based on the comprehensive amplitude limit compliance judgment result and thermal inertia stability evaluation index. If both indicators meet the preset safety conditions, generate a pre-run pass signal and mark the temperature control objective function to be executed as available. Otherwise, generate a pre-run failure signal and trigger the abnormal locking mechanism.

[0206] The system receives the amplitude limiting compliance determination results and thermal inertia stability assessment indicators as input data for logical judgment.

[0207] Boolean value parsing is performed on the amplitude limit compliance judgment result to confirm that the amplitude change rate in the simulated temperature control command sequence does not exceed the maximum allowable temperature rise slope threshold and the power output does not exceed the upper limit threshold.

[0208] Numerical comparisons were performed on the thermal inertial stability assessment indicators to verify that the temperature field evolution trend calculated based on the substrate heat capacity parameters and the adhesive curing kinetic model is within the thermal inertial safety boundary.

[0209] Construct a logic AND gate circuit to synchronously input the truth state of the amplitude limiting compliance judgment result and the compliance state of the thermal inertia stability evaluation index.

[0210] When both input states are true, the logical AND operation outputs a high-level signal, generates a pre-run pass signal, and sets the status flag of the temperature control objective function to be executed to the available state.

[0211] When any input state is false, the logical AND operation outputs a low-level signal, generating a pre-play failure signal and immediately triggering an exception locking mechanism to freeze the parameter update operation of the current background pre-play unit.

[0212] By using logical AND operations, the constraint verification results from the previous step are transformed into explicit control authority signals, achieving the expected technical effect of safe interlocking when switching the temperature control objective function.

[0213] For example, a maximum allowable temperature rise rate threshold is set to 5℃ / s, and a power output upper limit threshold is set to 80%. If the maximum temperature rise rate of the simulated command sequence is 4.2℃ / s and the peak power is 75%, the limiting compliance judgment result is true. If the thermal inertia stability assessment index calculated based on the substrate heat capacity shows that the temperature overshoot risk is lower than the safety threshold of 0.5℃, the thermal inertia stability assessment index is true. The logic AND operation outputs a pre-simulation pass signal, allowing the activation of new objective functions. If the temperature rise rate of the simulated command sequence reaches 6.1℃ / s, the limiting compliance judgment result is false, the logic AND operation outputs a pre-simulation failure signal, triggering anomaly locking, preventing unsafe objective functions from being put into the main control loop, and significantly improving the system's operational safety.

[0214] Step S6: If the pre-run pass signal is received, the temperature control objective function to be executed is put into the main control loop to replace the old objective function to generate real-time temperature control instructions, and the old objective function is transferred to the standby buffer to support millisecond-level rollback operations. Specifically, this includes: S6.1: Based on the pre-drill, the main control loop switching enable flag is generated by signal triggering. The status confirmation information output by the background pre-drill unit is judged by logic level to generate a highly effective real-time temperature control command switching authorization signal as the input condition for subsequent data path reconstruction.

[0215] It should be noted that the standby buffer is a memory storage area independent of the background pre-simulation unit and the main control loop front-end calculation unit. It is used to store the complete mathematical structure parameters, historical constraint coefficients, and current control cycle state snapshot of the replaced old objective function, and to establish version tags and rollback index entries. When the objective function newly added to the main control loop encounters an anomaly, the system quickly reads the backup package from the standby buffer through the rollback index entries to restore to the previous stable operating condition.

[0216] The system receives a pre-performance pass signal from the background pre-performance unit. This signal is a high-level logic state, indicating that the temperature control objective function to be executed has passed the safety limit and thermal inertia constraint verification.

[0217] The pre-simulation signal is subjected to jitter filtering, and a digital low-pass filter is used to eliminate high-frequency noise interference at the signal edges to ensure the stability of the state determination.

[0218] The filtered signal is input to the switching enable logic module of the main control circuit to perform a logic level determination operation to confirm that the signal duration exceeds the preset minimum effective pulse width threshold.

[0219] Based on the judgment result, a real-time temperature control command switching authorization signal is generated. This signal is a global synchronization trigger pulse, which is used to coordinate the data path switching action of the dual buffer register group.

[0220] The switching authorization signal is broadcast to the data bus controller through a hardware interrupt mechanism, locking the data write permission for the current control cycle to prevent data races or instruction conflicts during the switching process.

[0221] By using the above processing method, the pre-test verification results of the previous step are transformed into a switching authorization signal with timing synchronization characteristics, thereby realizing the prerequisite for smooth switching of the main control loop and ensuring the atomicity and security of system state transition.

[0222] For example, the pre-test signal duration is 5ms. After processing by a low-pass filter with a cutoff frequency of 100Hz, the signal fluctuation amplitude is reduced to within 0.1V. The switching enable logic module sets the minimum effective pulse width threshold to 2ms. After determining that the signal is valid, it generates a global synchronization trigger pulse with a width of 100μs. This pulse triggers a hardware interrupt, locks the data bus write permission, and ensures that the atomic replacement of the objective function instance is completed before the start of the next control cycle. This avoids heating power oscillations caused by instruction jumps and significantly improves the response stability of the temperature control system.

[0223] S6.2: Use the real-time temperature control instruction switching authorization signal to control the data path switching action of the double buffer register group, activate the main channel mapping of the temperature control objective function to be executed that has completed parameter reloading, so as to establish the temperature control objective function to be executed as the only valid calculation source of the current main control loop and output the newly activated temperature control objective function instance.

[0224] The system receives the real-time temperature control command switching authorization signal and analyzes the signal level to confirm that the background simulation unit has completed the security verification. Based on the authorization signal, it triggers the write protection release instruction of the double-buffered register group, granting write access to the storage area of ​​the temperature control target function to be executed.

[0225] The system reads the instantiated target item priority sequence, constraint relaxation coefficient matrix, and physical boundary value array from the register address of the background pre-simulation unit. The read set of structural parameters is then transmitted to the front-end calculation unit register of the main control loop via the internal high-speed data bus.

[0226] Perform an atomic swap operation in the registers to overwrite the old objective function parameters in the original front-end computation unit with the newly loaded structure parameters, ensuring that there is no intermediate state interference during the data update process.

[0227] Lock the computation source pointer of the foreground computation unit, pointing it to the memory address of the newly loaded target function instance, thus blocking the data access path of the old target function. Activate the computation enable bit of the new target function instance, starting the real-time solution logic of the main control loop.

[0228] By using atomic swaps and pointer redirection, the verification result from the previous step is transformed into a unique and effective control law instance, enabling seamless switching and zero-jitter transition of the temperature control strategy.

[0229] For example, when the system receives a high-level switching authorization signal, write protection is released in the address region 0x4000-0x40FF. A parameter block containing three target item weight coefficients of 0.7, 0.2, and 0.1, and a temperature ramp limit of 5°C / s is read from the background address 0x5000. The parameter block is copied to the foreground address 0x4000 within 2 microseconds via the DMA controller. A pointer switching instruction is executed, changing the calculation source pointer from the old instance 0x4000 to the new instance 0x4000. The new instance immediately responds to the current temperature deviation of 2°C and calculates a PWM (Pulse Width Modulation) instruction with a 60% duty cycle.

[0230] S6.3: Based on the newly activated temperature control objective function instance, replace the old objective function to perform real-time temperature control command generation calculation. Solve the output variables of the newly activated temperature control objective function instance in the current control cycle to generate a new generation of real-time temperature control command sequence containing heating power duty cycle and temperature rise slope limit information.

[0231] Obtain the instance of the newly activated temperature control objective function, as well as the real-time temperature deviation value and historical control quantity data within the current control cycle, as the initial input variables for instruction calculation.

[0232] Read the priority sequence of target items pre-stored in the objective function template to determine the calculation weight coefficients of temperature response speed, energy consumption efficiency and coating quality stability under the current operating conditions.

[0233] A multidimensional control law equation containing proportional, integral, derivative, and feedforward compensation terms is constructed, and the real-time temperature deviation value is substituted into the equation for linear combination calculation.

[0234] in, To control the output, For temperature deviation, , , These are the proportional, integral, and differential gain coefficients, which are dynamically adjusted according to the priority sequence.

[0235] The calculated preliminary control values ​​are subjected to boundary limiting processing based on the constraint relaxation coefficient set to eliminate abnormal values ​​that exceed the safety range of the physical actuator.

[0236] By combining the maximum allowable temperature rise slope limit, the rate of change of the control quantity after the limit is smoothed and filtered to generate the heating power duty cycle parameter that meets the thermal inertia requirements.

[0237] Simultaneously calculate the expected temperature trajectory for the next moment, derive and generate the corresponding temperature rise slope limit information, and form a complete new generation of real-time temperature control command sequence.

[0238] Through the above-mentioned multi-dimensional weight dynamic allocation and dual constraint verification processing method, the abstract objective function mathematical model is transformed into specific executable heating power duty cycle and heating slope limit data, so as to realize the accurate adaptation and stable output of temperature control commands to dynamic process states.

[0239] For example, for the semantic tag "thick material, low-speed wet state", the system loads the corresponding objective function template, setting Kp=1.2, Ki=0.05, and Kd=0.8. The current temperature deviation e(t) is -5℃, the cumulative value of the integral term is -10℃·s, and the differential term is 0.5℃ / s. Substituting into the formula, the initial control quantity u(t) is calculated as 1.2×(-5)+0.05×(-10)+0.8×0.5=-6.1. After constraint relaxation coefficient correction and a maximum heating slope limit of 2℃ / s, a real-time temperature control command with a heating power duty cycle of 65% and a heating slope limit of 1.8℃ / s is finally generated, effectively avoiding temperature overshoot caused by heat absorption by the adhesive layer and significantly improving the thermal balance stability of the coating interface.

[0240] S6.4: In response to the data path switching completion event, perform the old objective function state transition operation, encapsulate the replaced old objective function and its associated historical constraint parameters with a context snapshot, and generate a standby buffer old objective function backup package with complete state recovery capability.

[0241] In response to the data path switching completion event, the system captures the current running state pointer of the old objective function instance in the main control loop. It reads the accumulated integral term, historical deviation of the derivative term, and real-time dynamic adjustment of the constraint relaxation coefficient of the old objective function at the end of the previous control cycle. The system serializes and packages these running state variables with the static structural parameters of the old objective function to construct a state snapshot data structure containing complete context information. It performs cyclic redundancy check (CRC) calculation on the state snapshot data structure to generate a checksum field for integrity verification. The state snapshot with the checksum is written to the non-volatile storage area of ​​the standby buffer, and a unique version identifier is assigned to support backtracking indexing. Through serialization and encapsulation, the replaced old objective function and its associated historical constraint parameters are transformed into a standby buffer old objective function backup package with complete state recovery capabilities, achieving the expected technical effect of millisecond-level condition rollback.

[0242] For example, when the system switches from "thin material high-speed dry state" to "thick material low-speed wet state", the accumulated integral value of the old objective function at time t is captured as 12.5, the differential history deviation is -0.3, and the constraint relaxation coefficient is 0.85. These values ​​are packaged with the priority sequence [1,0,0] of the old objective function, and the CRC32 (cyclic redundancy check) checksum is calculated to be 0xA1B2C3D4. This data packet is written to the standby buffer address 0x00FF and marked with the version number V1.0. If the new objective function pre-simulation fails, the system restores the old objective function and its state within 2ms according to the V1.0 index to ensure continuous temperature control without disturbance.

[0243] S6.5: Perform millisecond-level rollback index registration processing based on the old objective function backup package in the standby buffer, bind pointers and mark versions for the storage address of the generated old objective function backup package in the standby buffer, so as to establish rollback index entries that support rapid recovery to the previous stable operating condition under abnormal conditions.

[0244] Receive the standby buffer old objective function backup packet generated by S6.4. This data packet contains the complete mathematical structure parameters of the replaced objective function, historical constraint coefficients, and a state snapshot of the current control cycle.

[0245] Parse the metadata in the backup package header, extract the unique identifier ID and generation timestamp of the old objective function, and use them as the primary key for the rollback index to ensure that the control law version corresponding to the most recent stable operating condition can be accurately located when an anomaly is triggered.

[0246] Calculate the starting physical address of the currently available storage block in the standby buffer, and based on the allocation algorithm of the memory management unit, write the old target function backup package into the specified non-volatile storage area, and record the byte length and checksum value it occupies to prevent bit flips or corruption of data during storage.

[0247] Construct a rollback index entry data structure, encapsulating the extracted unique identifier ID, storage start address pointer, data length, and version number into a fixed-length index record. The version number adopts a monotonically increasing counter mechanism to ensure that the timing logic of the old and new versions is strictly ordered.

[0248] The completed rollback index entries are registered to the circular index queue inside the controller, and the head and tail pointers of the queue are updated. If the queue is full, the oldest index entry is overwritten to maintain the utilization of limited storage space and ensure the priority of the latest stable state for recovery.

[0249] Integrity verification is performed on the registered index entries. This is done by reading the data at the storage address and recalculating the checksum. The result is then compared with the checksum stored in the index record to confirm that the data was written correctly and the index points to the correct location. Finally, a flag indicating successful index registration is generated.

[0250] Through the pointer binding and version marking processes described above, the backup data from the previous step is transformed into rollback index entries with fast addressing capabilities, achieving the expected technical effect of restoring to the previous stable temperature control strategy in milliseconds when the main control loop malfunctions.

[0251] For example, the standby buffer capacity is set to 10KB, and each backup package is 2KB in size. When the old target function backup package is generated, the system allocates a physical address of 0x2000A000 and calculates a CRC32 checksum of 0xA1B2C3D4. The ID "Func_V2_Thick" is extracted, with a timestamp of T=1050ms and a version number of V=12. An index entry {ID:"Func_V2_Thick",Addr:0x2000A000,Len:2048,CRC:0xA1B2C3D4,Ver:12} is constructed. This entry is written to the current position of the circular queue, and the tail pointer is updated. Data at 0x2000A000 is read to verify the CRC match, and registration is marked as successful. If a thermal inertia over-limit anomaly occurs subsequently, the system directly reads the address pointed to by this index to load the V12 version function, with the recovery time controlled within 2ms, significantly improving the system's fault self-healing capability.

[0252] Step S7: Based on the real-time temperature control command, the heating actuator is driven to adjust the temperature, and new multi-source heterogeneous process parameters are acquired again in the next acquisition cycle to trigger a new round of process semantic label updates and dynamic reconstruction of the temperature control objective function. Specifically, this includes: S7.1: Based on the real-time temperature control command output by the main control circuit, the duty cycle parameter and phase angle parameter in the command are processed by pulse width modulation analysis to generate a high-frequency pulse width modulation drive signal with specific timing characteristics, thereby providing a standardized digital control input for the power execution unit.

[0253] S7.2: The on / off logic control of the solid-state relay switch array is performed using a high-frequency pulse width modulation drive signal to adjust the average current intensity flowing into the ceramic heating rod assembly, thereby generating a controllable heat flow output that linearly corresponds to the real-time temperature control command.

[0254] Receive the high-frequency pulse width modulation drive signal generated by the analysis in step S7.1. This signal contains duty cycle parameters and phase control information characterizing the target heating power.

[0255] The high-frequency pulse width modulation drive signal is input to the logic control interface of the solid-state relay switch array, and the low-voltage signal on the control side is electrically isolated from the high-voltage circuit on the load side through opto-isolation circuit.

[0256] The optocoupler inside the solid-state relay responds to the rising and falling edges of the drive signal, triggering the bidirectional thyristor or triac to turn on and off, performing on / off switching operations with millisecond-level precision.

[0257] The switching action is completed at a specific zero-crossing point or phase angle of the alternating current cycle to suppress electromagnetic interference caused by sudden current changes and ensure the integrity of the current waveform flowing into the ceramic heating rod assembly.

[0258] Based on the duty cycle value in the real-time temperature control command, calculate the ratio of the solid-state relay conduction time to the total cycle within a complete PWM cycle, and determine the effective power delivery ratio.

[0259] As a purely resistive load, the instantaneous heating power of the ceramic heating rod assembly is proportional to the square of the effective value of the current flowing through it, and follows Joule's law for electrothermal conversion. By adjusting the duty cycle to change the equivalent average current intensity, the heat flux density distribution on the surface of the ceramic heating rod can be linearly controlled.

[0260] Eliminates temperature fluctuations caused by contact bounce in traditional mechanical relays, enabling stepless and smooth adjustment of heating power.

[0261] By using the high-frequency on / off logic control processing of solid-state relay switch arrays, the duty cycle command in the digital domain is converted into a controllable heat flow output in the analog domain, thereby achieving accurate linear tracking and rapid dynamic response of the temperature field at the coating interface.

[0262] For example, the PWM control frequency is set to 10kHz, and the period T is 0.1ms. When the real-time temperature control command requires 50% heating power, the duty cycle is set to 0.5. The solid-state relay is turned on for 0.05ms and off for 0.05ms in each cycle. If the effective value of the power supply voltage is 220V and the heating rod resistance is 10Ω, the peak current during full conduction is approximately 31.1A. Calculations show that the equivalent average current intensity is significantly reduced, and the generated average heat power precisely corresponds to 50% of the target value. In the "thin material, high speed" condition, the system needs to heat up rapidly, the duty cycle is instantly increased to 0.9, the heat flow output increases rapidly, and the temperature response time is shortened to the second level. In the "thick material, low speed" heat preservation stage, the duty cycle is finely adjusted to 0.3 to maintain a constant thermal balance, and the temperature fluctuation range is controlled within ±0.5℃, significantly improving the stability of the coating quality.

[0263] S7.3: Based on the controllable heat flow output, the contact interface of the coating pressing roller surface is subjected to heat conduction coupling treatment to change the instantaneous temperature field distribution in the adhesive melting area, thereby realizing the tracking of the target temperature trajectory matched by the current process semantic tag.

[0264] The system receives the controllable heat flux power signal output from S7.2 and maps it to the boundary heat flux condition of the heating zone on the surface of the laminating pressing roller. Based on Fourier's law of heat conduction, a three-dimensional transient heat conduction model of the adhesive layer melting region is constructed, dividing the pressing roller metal substrate, adhesive layer, and substrate into discretized grid nodes. The thermal conductivity matrix between adjacent grid nodes is calculated, and the temperature gradient distribution of each node is solved by combining the material thermal property parameters determined by the current process state semantic tag. Convective heat transfer boundary conditions are introduced to correct the influence of ambient temperature and humidity on the heat dissipation coefficient of the roller surface, and a dynamic heat balance equation is established. The temperature field distribution vector of the adhesive layer interface at the next moment is solved by a numerical iterative algorithm, and compared with the target temperature trajectory curve matched by the current process semantic tag. The deviation energy function between the actual temperature field and the target trajectory is calculated to generate a temperature compensation correction factor. This correction factor is fed back to the PID control loop of the heating actuator to fine-tune the pulse width modulation duty cycle of subsequent cycles. Through the above heat conduction coupling processing and closed-loop correction mechanism, the heat flux output of the previous step is converted into instantaneous temperature field distribution data that conforms to the target trajectory, achieving a high-precision temperature tracking control effect for dynamic working conditions.

[0265] S7.4: Multi-point synchronous acquisition and processing of instantaneous temperature field distribution is performed through an embedded infrared temperature sensor array to obtain the latest multi-source heterogeneous process parameter set containing spatial temperature gradient information, which serves as the initial input data for the next acquisition cycle.

[0266] S7.5: Based on the update event of the latest multi-source heterogeneous process parameter set, the decision tree classifier re-evaluation mechanism is triggered to generate a new round of process semantic labels, thereby initiating the dynamic reconstruction process of the temperature control objective function, which includes semantic mapping query and objective function template reloading.

[0267] Receive the latest multi-source heterogeneous process parameter set output from step S7.4. This dataset includes the temperature field distribution at multiple points on the surface of the coating pressing roller, real-time values ​​of ambient temperature and humidity, and the current production line operating status indicator.

[0268] Perform timing consistency checks on the latest multi-source heterogeneous process parameter set, eliminate transient anomalies caused by sensor jitter, and ensure the physical rationality and timing continuity of the input data.

[0269] The decision tree classifier kernel is invoked to map the cleaned process parameter vectors to the pre-constructed feature space, and the current working condition attributes are matched level by level according to the node splitting conditions.

[0270] Based on the matching path, locate the leaf node of the decision tree and extract the process semantic tags bound to the leaf node, such as "high viscosity thick material slow state" or "low viscosity thin material high speed state".

[0271] Monitor the change in semantic tags relative to the previous control cycle. If a tag transition is detected, generate an interrupt signal to trigger the reconstruction of the objective function.

[0272] The interrupt signal is sent to the state machine controller of the collaborative modeling module to activate the query interface of the semantic-target structure mapping table.

[0273] The address pointer of the pre-validated temperature control objective function template, which uniquely corresponds to the new semantic label, is quickly retrieved using a hash index.

[0274] Read the target item priority sequence, constraint relaxation coefficients, and physical boundary value array pointed to by the address, and prepare to execute the background pre-load.

[0275] Through the above chain processing, real-time process data is transformed into semantic tags with clear thermal response characteristics, and the dynamic reconstruction process of the temperature control objective function is precisely triggered, so as to realize the millisecond-level adaptive response of the control system to complex operating condition changes.

[0276] For example, when the infrared sensor array detects a sudden change in the temperature gradient at the center of the pressing roller from 5℃ / cm to 12℃ / cm, and the online viscosity meter reading rises from 300mPa·s to 800mPa·s, the system determines that the operating condition transitions from "normal state" to "high viscosity, thick material, slow-speed state". The lightweight decision tree, based on preset rules, outputs a label ID of 0x0A (0x0A is the hexadecimal index key of this semantic label) if the viscosity is >500mPa·s and the thickness is >0.2mm. The system immediately queries the mapping table to obtain the objective function template corresponding to ID0x0A, setting its priority to "interface thermal balance weight 0.7, energy consumption weight 0.3", and adjusting the constraint relaxation coefficient to 1.2 to allow for a larger heating slope.

[0277] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0278] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the element or object preceding “comprising” or “including” encompasses the element or object listed following “comprising” or “including” and its equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0279] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing temperature control in a fully automated laminating machine with multi-parameter coordination during the laminating process, characterized in that, include: Obtain multi-source heterogeneous process parameters during the coating process, and standardize the multi-source heterogeneous process parameters to generate a standardized process parameter set; A decision tree classifier is constructed based on industry expert knowledge and experimental data of typical working conditions. The standardized process parameter set is input into the decision tree classifier for rule-guided training, and process semantic labels are output. A semantic-objective structure mapping table is established based on the mapping relationship between the process semantic tags and the pre-configured objective function templates. Each objective function template contains control parameters adapted to specific operating conditions. Based on the process semantic tags generated at the current moment, the semantic-target structure mapping table is queried, the corresponding target function template is called and its control parameters are loaded to complete the replacement of the temperature control target function; The replaced temperature control objective function is input into the background pre-simulation unit for verification. Within one control cycle, the generated temperature control command is verified to meet the safety constraints, and a pre-simulation pass signal or failure signal is generated. If the pre-run pass signal is received, the temperature control objective function is put into the main control loop to replace the old objective function to generate real-time temperature control instructions, and the old objective function is transferred to the standby buffer to support rollback operation; Based on the real-time temperature control command, the heating actuator is driven to adjust the temperature, and new multi-source heterogeneous process parameters are acquired again in the next acquisition cycle to trigger a new round of process semantic label update and dynamic reconstruction of temperature control objective function.

2. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, A semantic-objective structure mapping table is established based on the mapping relationship between the process semantic tags and the pre-configured objective function templates. Each objective function template contains control parameters adapted to specific operating conditions, specifically including: Based on the set of process semantic tags, the dominant heat conduction mechanism is analyzed for each type of process combination with common thermal response characteristics, so as to generate a dominant heat conduction mechanism identifier that characterizes the heat transfer characteristics under specific operating conditions. Based on the heat conduction dominant mechanism identifier, a targeted target item priority order configuration scheme is designed using industry expert knowledge and typical working condition experimental data to generate a target item priority order sequence that includes weighted sorting of temperature response speed, energy consumption efficiency and coating quality stability. Based on the priority sequence of the target items, combined with the material deformation feedback requirements and the adhesive layer curing gradient control requirements, the constraint relaxation coefficient values ​​adapted to the current heat conduction dominant mechanism are calculated to generate a set of constraint relaxation coefficients for adjusting the constraint boundary of the objective function. Based on the set of constraint relaxation coefficients and the maximum allowable temperature rise slope limit, the physical boundary value parameters of each objective function template are determined to generate a physical boundary value array that limits the actuator's safety amplitude and thermal inertia constraints. Using the process semantic tags as index keys, and associating them with the corresponding target item priority order sequence, constraint relaxation coefficient set, and physical boundary value array, a complete mapping record is constructed to generate a semantic-target structure mapping table that supports millisecond-level query calls.

3. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, The replaced temperature control objective function is input into the background simulation unit for verification. Within one control cycle, the generated temperature control command is verified to meet safety constraints, generating a simulation pass signal or a simulation failure signal, specifically including: The system acquires the temperature control objective function to be executed after parameter reloading and the process status feedback data at the current moment. Based on the double-buffered objective function execution mechanism, the background pre-simulation unit constructs a virtual control closed-loop environment and loads the temperature control objective function to be executed into the virtual control closed-loop environment to initialize the pre-simulation calculation context. The forward inference calculation is performed within a complete control cycle using the initialized pre-calculation context. A simulated temperature control command sequence is generated based on the mathematical structure of the temperature control objective function to be executed, and the simulated temperature control command sequence is used as the basic input data for subsequent constraint verification. The generated simulated temperature control command sequence is subjected to actuator safety limit constraint verification processing. The amplitude change rate in the simulated temperature control command sequence is compared and analyzed with the maximum allowable temperature rise slope threshold and the power output upper limit threshold to generate the limit compliance judgment result. Based on the amplitude limit compliance judgment result, the thermal inertia constraint matching degree of the simulated temperature control command sequence is calculated. Combined with the substrate heat capacity parameter and the adhesive layer curing gradient control requirements, the temperature field evolution trend caused by the simulated temperature control command sequence is evaluated to determine whether it exceeds the thermal inertia safety boundary, so as to generate thermal inertia stability assessment index. The logic and operation are executed by combining the comprehensive amplitude limit compliance judgment result and the thermal inertia stability assessment index. If both indicators meet the preset safety conditions, a pre-run pass signal is generated and the temperature control objective function to be executed is marked as available. Otherwise, a pre-run failure signal is generated and an abnormal locking mechanism is triggered.

4. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, The multi-source heterogeneous process parameters include at least five of the following: film thickness, pressing speed, ambient temperature and humidity, substrate type, and adhesive viscosity.

5. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, The steps of obtaining and standardizing multi-source heterogeneous process parameters specifically include: The multi-source heterogeneous process parameters are subjected to analog-to-digital conversion and timing alignment to generate a multi-source heterogeneous original process parameter sequence; the multi-source heterogeneous original process parameter sequence is parsed, anomaly removal is performed, and missing parameters are imputed to generate a cleaned set of effective process parameters.

6. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 5, characterized in that, The step of obtaining and standardizing multi-source heterogeneous process parameters further includes: A normalized mapping matrix is ​​constructed using the upper and lower limits of the physical range of each process parameter. Min-max normalization is then performed on the continuous process parameters in the set of effective process parameters after cleaning to generate dimensionless continuous process feature vectors.

7. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 6, characterized in that, The step of obtaining and standardizing multi-source heterogeneous process parameters further includes: Perform a unique thermal encoding mapping transformation on the substrate type encoding in the effective process parameter set after cleaning, and concatenate the transformed unique thermal encoding vector to the end of the dimensionless continuous process feature vector to generate a mixed-dimensional process parameter vector.

8. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, The steps of constructing a decision tree classifier, performing rule-guided training, and outputting process semantic labels specifically include: An initial rule base is constructed based on industry expert knowledge and experimental data under typical working conditions. The initial rule base is used to divide the feature space of the standardized process parameter set. The optimal splitting attribute is determined by the information gain ratio, and a decision tree classifier model is generated.

9. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, The steps of constructing a decision tree classifier, performing rule-guided training, and outputting process semantic labels also include: Based on the decision tree classifier model, path deduction is performed on the standardized process parameter set, and a classification path sequence is generated by matching the node splitting conditions step by step; the process semantic labels are extracted according to the mapping relationship of the leaf nodes at the end of the classification path sequence.

10. The temperature control optimization method for a fully automatic laminating machine with multi-parameter coordination in the laminating process according to claim 1, characterized in that, The steps of querying the semantic-target structure mapping table, calling the corresponding target function template and loading its control parameters to complete the replacement of the temperature control target function specifically include: Using the process semantic tag as the index key, the corresponding target function template address pointer is located from the semantic-target structure mapping table through hash lookup; The set of structural parameters and the set of physical boundary values ​​pointed to by the address pointer are read, and the set of structural parameters and the set of physical boundary values ​​are written into the background pre-drilling unit register to instantiate and generate the temperature control target function to be executed.