A temperature control intelligent optimization method and system for adhesive tape production

Abstract

The present invention relates to a temperature control intelligent optimization method and system for adhesive tape production, comprising the following steps: S1: collecting production data in the drying process through a multimodal sensor; S2: constructing a state prediction model based on GNN and LSTM, inputting the collected production data into the prediction model, and dynamically predicting the state between each temperature control area; S3: based on the prediction of the state between each temperature control area, modeling each temperature control area as an intelligent agent, and realizing the dynamic allocation of heat resources between each heating interval by training the collaborative control strategy between intelligent agents; S4: incorporating a fuzzy control optimization scheme for environmental compensation: for short-term changes in ambient temperature and humidity, using fuzzy control to dynamically adjust the temperature zone gradient distribution between drying sections to optimize energy efficiency; S5: the temperature control module adjusts the temperature and wind speed of the temperature control area in real time according to the optimization scheme obtained in steps S3 and S4. The present invention comprehensively improves the intelligent level of the temperature control system in the drying process of adhesive tape production.

Description

Technical Field

[0001] The present invention relates to the field of intelligent temperature control, and in particular to a temperature control intelligent optimization method and system for adhesive tape production. Background Art

[0002] During the production of adhesive tapes, the coated substrate needs to undergo a strict drying process to ensure that the solvent in the coating can fully evaporate, thereby forming a uniform and stable coating structure. This drying process plays a vital role in the final product quality of the adhesive tape. However, in actual production, due to the complexity of production conditions and high standards of process requirements, the temperature control process in the drying section faces many technical challenges.

[0003] First, the drying section of tape production is usually divided into multiple heating zones, and the temperature and hot air flow rate of each zone need to be precisely controlled to ensure uniform heat transfer and efficient solvent volatilization. Reasonable setting of temperature gradient is an important factor affecting coating quality: too high temperature may cause the coating to dry too quickly and cause cracks or bubbles, while too low temperature may lead to insufficient solvent volatilization and excessive residual solvent, which in turn affects the adhesion and quality of the tape. However, due to thermal interference between different zones in actual production (such as the temperature difference caused by heat transfer from high temperature zone to low temperature zone), the coordinated control between the various zones in the drying section becomes particularly complicated.

[0004] Secondly, the temperature control process in the drying section is not only affected by internal parameters (such as temperature control setpoints, coating thickness, solvent characteristics, etc.), but is also deeply constrained by the dynamic changes of external environmental factors. Changes in workshop ambient temperature and humidity directly affect the volatilization efficiency and the thermal response characteristics of the substrate. At the same time, fluctuations in production rhythm (such as changes in coating speed) further increase the complexity of the drying system. For example, in high-speed coating mode, the demand for heat and volatilization rate in a short period of time increases significantly. Traditional temperature control methods based on fixed parameters are difficult to quickly adapt to these dynamic conditions, resulting in delayed effects or insufficient adjustments.

[0005] In addition, traditional drying section temperature control technology usually relies on simple feedback control strategies (such as PID control) to track the set value through a single adjustment of temperature or flow rate. However, this method cannot fully take into account the coupling between multiple variables, nor can it predict and feedforward the future state of the drying area. When the coating thickness increases or the production line speed increases, this passive control method cannot optimize the heat distribution in real time, and it is easy to cause problems such as excessive local temperature difference or overheating or overcooling. Not only will it cause local quality defects on the coating surface (such as uneven coating, bubbles, insufficient drying, etc.), it may also lead to significant energy waste (such as unreasonable use of hot air energy). Summary of the invention

[0006] In order to solve the above problems, the purpose of the present invention is to provide a temperature control intelligent optimization method and system for tape production, which comprehensively improves the intelligence level of the temperature control system during the drying process of tape production.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A temperature control intelligent optimization method for adhesive tape production comprises the following steps:

[0009] S1: Collect production data during the drying process through multimodal sensors, including environmental parameters, coating thickness, temperature fluctuations, and solvent evaporation rate;

[0010] S2: Build a state prediction model based on GNN and LSTM, input the collected production data into the prediction model, and dynamically predict the state of each temperature control area;

[0011] S3: Based on the prediction of the state between each temperature control area, each temperature control area is modeled as an intelligent agent, and the dynamic allocation of heat resources in each heating area is achieved by training the collaborative control strategy between intelligent agents;

[0012] S4: Fuzzy control optimization scheme with environmental compensation: In view of short-term changes in ambient temperature and humidity, fuzzy control is used to dynamically adjust the temperature gradient distribution between drying sections to optimize energy efficiency;

[0013] S5: The temperature control module adjusts the temperature and wind speed of the temperature control area in real time according to the optimization scheme obtained in steps S3 and S4.

[0014] Furthermore, multimodal sensors are used to collect production data during the drying process, including environmental parameters, coating thickness, temperature fluctuations, and solvent evaporation rate, as follows:

[0015] After coating and before drying, a pair of laser thickness sensors are installed to measure the physical thickness of the coating, obtain multiple measurements per second, and calculate the difference between the actual coating thickness and the target thickness through the data acquisition module;

[0016] Infrared temperature sensors are installed in each temperature zone of the drying section, arranged on the top and both sides of the drying box to sense the surface temperature of the coating layer in real time. Thermocouples are installed inside the drying box and on the substrate to collect temperature data of the bottom layer of the substrate to reflect the changes in heat conduction between the substrate and the coating. Multiple thermocouples are placed at the entrance and exit of the drying section to monitor the changes in the initial heating and final temperature of the coating layer.

[0017] Install an infrared spectrometer in the exhaust pipe of the drying box to collect real-time data on the solvent concentration in the exhaust gas; install a humidity sensor at the air inlet or sampling port of the drying box to monitor changes in external humidity;

[0018] Use distributed data acquisition nodes to collect coating thickness, bottom layer temperature, surface temperature, and solvent evaporation rate data, and synchronize the asynchronously collected multimodal data with timestamps, and remove noise through Kalman filtering;

[0019] Dynamic Bayesian Network (DBN) was used to infer the potential relationship among coating thickness, temperature fluctuation and solvent concentration.

[0020] Furthermore, the dynamic Bayesian network DBN is used to infer the potential relationship between coating thickness, temperature fluctuation and solvent concentration as follows:

[0021] Assume T t represents the substrate coating thickness at time t; W t represents the local temperature distribution at time t; C t represents the local solvent evaporation concentration at time t; S t Indicates the degree of dryness of the coating at time t; According to the causal characteristics of the production process, a network diagram of the dynamic Bayesian network is constructed:

[0022] Define the conditional probability relationship (conditional probability distribution, CPD) of the Bayesian network. For time t, the overall joint probability is expressed as:

[0023] ;

[0024] The above joint probability is extended to multiple time slices to describe the joint distribution under time dynamic evolution:

[0025] ;

[0026] In real-time production, the variable states at the current time t are inferred using a forward-backward algorithm:

[0027] ;

[0028] Where T represents T time steps; They represent the substrate coating thickness sequence, local temperature sequence, local solvent evaporation concentration sequence and drying degree sequence from time 1 to t respectively;

[0029] According to the observed data of {Tt, Wt, Ct}, the drying completion probability is calculated:

[0030] ;

[0031] Among them, C target is the target value of local solvent evaporation concentration;

[0032] The time series model is used to describe the dynamic evolution of variables over time and reveal the dependencies between multiple time slices.

[0033] Furthermore, a state prediction model is constructed based on GNN and LSTM, as follows:

[0034] The prediction model inputs: multimodal data collected in real time, including coating thickness, temperature fluctuation, solvent concentration, and environmental parameters in each temperature control zone;

[0035] Output: predict the state of each temperature control area in the next several time steps, including solvent concentration distribution, coating drying state, and energy consumption index;

[0036] The drying system is divided into N regions, and the state of each region at any time t is expressed as:

[0037] ;

[0038] in, is the coating thickness of the nth region; is the temperature fluctuation of the nth region; is the solvent concentration in the nth region; It is a global environment parameter.

[0039] The state of the entire system at time t is expressed as:

[0040] ;

[0041] Define a graph G = (V, E): Node V corresponds to N dry areas, each node represents the state of an area; edge E represents the spatial association between areas;

[0042] GNN is used to extract spatial interaction features between regions, and node features are updated through neighborhood aggregation. The formula is as follows:

[0043] ;

[0044] in, is the feature representation of node n in layer l, and the initial input is ; is the set of neighbor nodes of node n; is the edge weight between nodes n and j; is the learnable weight matrix of the lth layer;

[0045] σ is the activation function;

[0046] After L layers of graph convolution, the final spatial feature representation of each node is obtained:

[0047]

[0048] in Contains the node n’s own state information and the interaction information of its neighbors;

[0049] Combining the features of all nodes, we get the spatial features of the entire system:

[0050] ;

[0051] Dynamic modeling is performed through LSTM. For each node n, its time series input is , the output of LSTM is several time steps in the future Status prediction:

[0052] ;

[0053] in, is the predicted state of region n in the future time steps t+1 to t+τ;

[0054] During the training process, the loss function is defined to optimize the model parameters:

[0055] ;

[0056] in, , are the actual and predicted solvent concentrations, respectively; , are the actual and predicted temperature distributions, respectively; is the weight parameter.

[0057] Furthermore, step S3 is specifically as follows:

[0058] Each temperature control area corresponds to an intelligent agent, which adjusts control parameters by sensing the state of its area and the state of neighboring areas to make resource allocation decisions;

[0059] A centralized training framework is used to simulate the dynamic behavior of the drying process using a simulator in the offline stage to train the global strategy model; agents learn collaborative optimization strategies by sharing global feedback rewards;

[0060] In online operation, each agent only perceives the current local state and performs actions independently.

[0061] Further, the agent is as follows:

[0062] The state description of the i-th agent includes:

[0063] ;

[0064] in, is the coating thickness of area i; is the temperature distribution of region i; is the solvent concentration in region i; is the environmental parameter;

[0065] Actions of agent i Including control parameter adjustment:

[0066] ;

[0067] in, The hot air velocity changes; The target temperature adjustment amount for the current area; is the drying oven outlet temperature adjustment;

[0068] The immediate reward for each agent Taking drying effect and energy efficiency into consideration, the specific definition is:

[0069] ;

[0070] Among them, C target is the target value of solvent concentration; T target is the temperature target value; For energy consumption; , and is the weight coefficient;

[0071] The policy-based reinforcement learning algorithm PPO is used to optimize the agent's strategy, enabling the agent to dynamically adjust control parameters.

[0072] Furthermore, the policy-based reinforcement learning algorithm PPO is used to optimize the agent's strategy, so that the agent can dynamically adjust the control parameters, as follows:

[0073] Initialize the policy network π θ (A∣S) and Value Network The weight parameters are:

[0074] Run the current policy πθ in the environment, sample T time steps of data, and record the state S t 、Action A t , Reward R t , new state S t+1 , calculate the advantage function Y t :

[0075] ;

[0076] in, Status The value network output below; is the learning parameter;

[0077] Using the policy objective function L PPO (θ) Update the policy network weights θ:

[0078] ;

[0079] ;

[0080] Among them, α π is the learning rate; is the ratio of the new and old strategies; To limit the scope of policy updates, is a preset constant; represents the expected function;

[0081] Use the value network loss function L value Update value network weights :

[0082] ;

[0083] ;

[0084] in, is the learning rate of the value network;

[0085] Repeat the above steps until the strategy converges.

[0086] Furthermore, S4 is specifically:

[0087] The system monitors the ambient temperature Tenv and the ambient humidity Henv, and combines key data of the drying section, including substrate surface temperature and coating speed;

[0088] The collected environmental data is fuzzified and input into the fuzzy control rule base; fuzzy reasoning is performed through the rule base to determine the gradient adjustment value ΔT that is suitable for the current environment set and hot air adjustment value ΔF set ;

[0089] The specific parameter adjustment value of the fuzzy output is calculated by the center of gravity method, and the adjustment value is fed back to the drying section temperature control module to complete the compensation adjustment in real time.

[0090] A temperature control intelligent optimization system for adhesive tape production includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps of the temperature control intelligent optimization method for adhesive tape production as described above.

[0091] The present invention has the following beneficial effects:

[0092] 1. The present invention can describe the complex dynamic relationship between coating thickness, temperature fluctuation and solvent concentration in the drying process through modeling and inference of dynamic Bayesian networks, and can infer the drying state in real time, and optimize production parameters accordingly, improve drying efficiency, save energy consumption, and ensure the consistency of product quality;

[0093] 2. The present invention uses GNN to model the internal correlation of the drying area, captures the interaction characteristics between the drying sections, and combines LSTM to learn the temporal dynamic change law of the production process. It can predict the temperature fluctuation and solvent volatilization rate of the temperature control area in real time. Compared with the traditional single-area, static prediction model, it can accurately capture the local state and dynamic response, and adapt to complex industrial dynamic scenarios;

[0094] 3. The present invention models each temperature control area as an intelligent entity, and realizes the dynamic distribution of heat among the drying areas through reinforcement learning training collaborative strategy, which solves the problems of energy reuse and insufficient overall efficiency optimization caused by traditional independent control of each area, and makes the global performance of the system better; and in response to sudden fluctuations in ambient temperature and humidity, a dynamic fuzzy control strategy is designed to automatically adjust the temperature zone gradient according to the fluctuation amount to realize adaptive control, improve the system's resistance to external interference and stability, and have good adaptability to changes in the production environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 is a flow chart of the method of the present invention;

[0096] Figure 2 Schematic diagram of cause-effect relationship within a time slice in one embodiment of the present invention. DETAILED DESCRIPTION

[0097] The following is combined with Figure 1-2 The present invention is further described in detail with specific embodiments:

[0098] A temperature control intelligent optimization method for adhesive tape production comprises the following steps:

[0099] S1: Collect production data during the drying process through multimodal sensors, including environmental parameters, coating thickness, temperature fluctuations, and solvent evaporation rate;

[0100] S2: Build a state prediction model based on GNN and LSTM, input the collected production data into the prediction model, and dynamically predict the state of each temperature control area;

[0101] S3: Based on the prediction of the state between each temperature control area, each temperature control area is modeled as an intelligent agent, and the dynamic allocation of heat resources in each heating area is achieved by training the collaborative control strategy between intelligent agents;

[0102] S4: Fuzzy control optimization scheme with environmental compensation: In view of short-term changes in ambient temperature and humidity, fuzzy control is used to dynamically adjust the temperature gradient distribution between drying sections to optimize energy efficiency;

[0103] S5: The temperature control module adjusts the temperature and wind speed of the temperature control area in real time according to the optimization scheme obtained in steps S3 and S4.

[0104] In this embodiment, the production data during the drying process is collected by a multimodal sensor, including environmental parameters, coating thickness, temperature fluctuations, and solvent evaporation rate, as follows:

[0105] After coating and before drying, a pair of laser thickness sensors (arranged symmetrically up and down) are installed to measure the physical thickness of the coating, obtain multiple measurements per second, and calculate the difference between the actual coating thickness and the target thickness through the data acquisition module;

[0106] Infrared temperature sensors are installed in each temperature zone of the drying section, arranged on the top and both sides of the drying box to sense the surface temperature of the coating layer in real time. Thermocouples are installed inside the drying box and on the substrate to collect temperature data of the bottom layer of the substrate to reflect the changes in heat conduction between the substrate and the coating. Multiple thermocouples are placed at the entrance and exit of the drying section to monitor the changes in the initial heating and final temperature of the coating layer.

[0107] Install an infrared spectrometer in the exhaust pipe of the drying box to collect real-time data on the solvent concentration in the exhaust gas; install a humidity sensor at the air inlet or sampling port of the drying box to monitor changes in external humidity;

[0108] Use distributed data acquisition nodes to collect coating thickness, bottom layer temperature, surface temperature, and solvent evaporation rate data, and synchronize the asynchronously collected multimodal data with timestamps, and remove noise through Kalman filtering;

[0109] Dynamic Bayesian Network (DBN) was used to infer the potential relationship among coating thickness, temperature fluctuation and solvent concentration.

[0110] In this embodiment, the potential relationship between coating thickness, temperature fluctuation and solvent concentration is inferred using a dynamic Bayesian network DBN, as follows:

[0111] Assume Tt represents the substrate coating thickness at time t; W t represents the local temperature distribution at time t; C t represents the local solvent evaporation concentration at time t; S t Indicates the degree of dryness of the coating at time t; According to the causal characteristics of the production process, a network diagram of the dynamic Bayesian network is constructed:

[0112] Causality within a time slice:

[0113] T t →W t : Coating thickness affects local temperature distribution. Areas with thicker coating require higher heat and have greater temperature fluctuations.

[0114] W t →C t :Local temperature increase will lead to faster solvent evaporation rate.

[0115] C t →S t : The solvent concentration drops below the target concentration and the coating layer reaches a dry state;

[0116] Dynamic relationship between time slices (t-1 to t):

[0117] T t →T t+1 : The current coating thickness affects the thickness at the next time (thinning is caused by the drying process);

[0118] W t →W t+1 :The current local temperature fluctuation affects the temperature continuity behavior at the next time;

[0119] C t →C t+1 :The current solvent evaporation concentration decreases over time, affected by environmental conditions and drying speed;

[0120] Define the conditional probability relationship (conditional probability distribution, CPD) of the Bayesian network. For time t, the overall joint probability is expressed as:

[0121] ;

[0122] The above joint probability is extended to multiple time slices to describe the joint distribution under time dynamic evolution:

[0123] ;

[0124] Where T represents T time steps; They represent the substrate coating thickness sequence, local temperature sequence, local solvent evaporation concentration sequence and drying degree sequence from time 1 to t respectively;

[0125] In real-time production, the variable states at the current time t are inferred using a forward-backward algorithm:

[0126] ;

[0127] According to {T t ,W t ,C t} observation data, calculate the drying completion probability:

[0128] ;

[0129] Among them, C target is the target value of local solvent evaporation concentration;

[0130] The time series model is used to describe the dynamic evolution of variables over time and reveal the dependencies between multiple time slices.

[0131] In this embodiment, a state prediction model is constructed based on GNN and LSTM, as follows:

[0132] The prediction model inputs: multimodal data collected in real time, including coating thickness, temperature fluctuation, solvent concentration, and environmental parameters (humidity and air inlet temperature) in each temperature control zone;

[0133] Output: predict the state of each temperature control area in the next several time steps, including solvent concentration distribution, coating drying state, and energy consumption index;

[0134] The drying system is divided into N regions, and the state of each region at any time t is expressed as:

[0135] ;

[0136] in, is the coating thickness of the nth region; is the temperature fluctuation of the nth region; is the solvent concentration in the nth region; It is a global environment parameter.

[0137] The state of the entire system at time t is expressed as:

[0138] ;

[0139] Define a graph G = (V, E): Node V corresponds to N dry areas, each node represents the state of an area; edge E represents the spatial association between areas;

[0140] GNN is used to extract spatial interaction features between regions, and node features are updated through neighborhood aggregation. The formula is as follows:

[0141] ;

[0142] in, is the feature representation of node n in layer l, and the initial input is ; is the set of neighbor nodes of node n; is the edge weight between nodes n and j; is the learnable weight matrix of the lth layer;

[0143] σ is the activation function;

[0144] After L layers of graph convolution, the final spatial feature representation of each node is obtained:

[0145]

[0146] in Contains the node n’s own state information and the interaction information of its neighbors;

[0147] Combining the features of all nodes, we get the spatial features of the entire system:

[0148] ;

[0149] Dynamic modeling is performed through LSTM. For each node n, its time series input is , the output of LSTM is several time steps in the future Status prediction:

[0150] ;

[0151] in, is the predicted state of region n in the future time steps t+1 to t+τ;

[0152] During the training process, the loss function is defined to optimize the model parameters:

[0153] ;

[0154] in, , are the actual and predicted solvent concentrations, respectively; , are the actual and predicted temperature distributions, respectively; is the weight parameter.

[0155] In this embodiment, step S3 is specifically as follows:

[0156] Each temperature control area corresponds to an intelligent agent, which adjusts control parameters by sensing the state of its area and the state of neighboring areas to make resource allocation decisions;

[0157] A centralized training framework is used to simulate the dynamic behavior of the drying process using a simulator (based on historical production data or physical simulation) in the offline stage to train the global strategy model. Agents learn collaborative optimization strategies by sharing global feedback rewards.

[0158] In online operation, each agent only perceives the current local state (the state of its own region and its neighborhood) and performs actions independently.

[0159] In this embodiment, the intelligent agent is as follows:

[0160] The state description of the i-th agent includes:

[0161] ;

[0162] in, is the coating thickness of area i; is the temperature distribution of region i; is the solvent concentration in region i; is the environmental parameter;

[0163] Actions of agent i Including control parameter adjustment:

[0164] ;

[0165] in, The hot air velocity changes; The target temperature adjustment amount for the current area; is the drying oven outlet temperature adjustment;

[0166] The immediate reward for each agent Taking drying effect and energy efficiency into consideration, the specific definition is:

[0167] ;

[0168] Among them, C target is the target value of solvent concentration; T target is the temperature target value; For energy consumption; , and is the weight coefficient;

[0169] The policy-based reinforcement learning algorithm PPO is used to optimize the agent's strategy, enabling the agent to dynamically adjust control parameters.

[0170] In this embodiment, a policy-based reinforcement learning algorithm PPO is used to optimize the agent's strategy so that the agent can dynamically adjust control parameters, as follows:

[0171] Initialize the policy network π θ (A∣S) and Value Network The weight parameters are:

[0172] Run the current policy πθ in the environment, sample T time steps of data, and record the state S t 、Action A t , Reward R t , new state S t+1 , calculate the advantage function Y t :

[0173] ;

[0174] in, Status The value network output below; is the learning parameter;

[0175] Using the policy objective function L PPO (θ) Update the policy network weights θ:

[0176] ;

[0177] ;

[0178] Among them, α π is the learning rate; is the ratio of the new and old strategies; To limit the scope of policy updates, is a preset constant; represents the expected function;

[0179] Use the value network loss function L value Update value network weights :

[0180] ;

[0181] ;

[0182] in, is the learning rate of the value network;

[0183] Repeat the above steps until the strategy converges.

[0184] In this embodiment, S4 is specifically:

[0185] The system monitors the ambient temperature Tenv and the ambient humidity Henv, and combines key data of the drying section, including substrate surface temperature and coating speed;

[0186] The collected environmental data is fuzzified and input into the fuzzy control rule base; fuzzy reasoning is performed through the rule base to determine the gradient adjustment value ΔT that is suitable for the current environment set and hot air adjustment value ΔF set ;

[0187] The specific parameter adjustment value of the fuzzy output (including the new temperature zone gradient setting value and hot air flow rate) is calculated through the center of gravity method, and the adjustment value is fed back to the drying section temperature control module to complete the compensation adjustment in real time.

[0188] Fuzzy rules:

[0189] IF Tenv=low AND Henv=low → ΔTset=slight decrease, ΔFset=no change;

[0190] IF Tenv= low AND Henv= high → ΔTset= small increase, ΔFset= small increase;

[0191] IF Tenv=medium AND Henv=high → ΔTset=substantially increased, ΔFset=substantially increased;

[0192] IF Tenv=high AND Henv=low → ΔTset=large decrease, ΔFset=small decrease;

[0193] IF Tenv=high AND Henv=high → ΔTset=no change, ΔFset=no change.

[0194] A temperature control intelligent optimization system for adhesive tape production includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps of the temperature control intelligent optimization method for adhesive tape production as described above.

[0195] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0196] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0197] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0198] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0199] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A temperature control intelligent optimization method for tape production, characterized in that: The following steps are involved: S1: Collect production data during the drying process through multimodal sensors, including environmental parameters, coating thickness, temperature fluctuations, and solvent evaporation rate; S2: Build a state prediction model based on GNN and LSTM, input the collected production data into the prediction model, and dynamically predict the state of each temperature control area; S3: Based on the prediction of the state between each temperature control area, each temperature control area is modeled as an intelligent agent, and the dynamic allocation of heat resources in each heating area is achieved by training the collaborative control strategy between intelligent agents; S4: Fuzzy control optimization scheme with environmental compensation: In view of short-term changes in ambient temperature and humidity, fuzzy control is used to dynamically adjust the temperature gradient distribution between drying sections to optimize energy efficiency; S5: The temperature control module adjusts the temperature and wind speed of the temperature control area in real time according to the optimization scheme obtained in steps S3 and S4; The state prediction model based on GNN and LSTM is constructed as follows: The prediction model inputs: multimodal data collected in real time, including coating thickness, temperature fluctuation, solvent concentration, and environmental parameters in each temperature control zone; Output: predict the state of each temperature control area in the next several time steps, including solvent concentration distribution, coating drying state, and energy consumption index; The drying system is divided into N regions, and the state of each region at any time t is expressed as: ； in, is the coating thickness of the nth region; is the temperature fluctuation of the nth region; is the solvent concentration in the nth region; It is the global environment parameter; The state of the entire system at time t is expressed as: ； Define a graph G = (V, E): Node V corresponds to N dry areas, each node represents the state of an area; edge E represents the spatial association between areas; GNN is used to extract spatial interaction features between regions, and node features are updated through neighborhood aggregation. The formula is as follows: ； in, is the feature representation of node n in layer l, and the initial input is ; is the set of neighbor nodes of node n; is the edge weight between nodes n and j; is the learnable weight matrix of the lth layer; σ is the activation function; After L layers of graph convolution, the final spatial feature representation of each node is obtained: in Contains the node n’s own state information and the interaction information of its neighbors; Combining the features of all nodes, we get the spatial features of the entire system: ; Dynamic modeling is performed through LSTM. For each node n, its time series input is , the output of LSTM is several time steps in the future Status prediction: ； in, is the predicted state of region n in the future time steps t+1 to t+τ; During the training process, the loss function is defined to optimize the model parameters: ; in, , are the actual and predicted solvent concentrations, respectively; , are the actual and predicted temperature distributions, respectively; a 1 、a 2 is the weight parameter.

2. The temperature control intelligent optimization method for adhesive tape production according to claim 1 is characterized in that: The multimodal sensor is used to collect production data during the drying process, including environmental parameters, coating thickness, temperature fluctuations, and solvent evaporation rate, as follows: After coating and before drying, a pair of laser thickness sensors are installed to measure the physical thickness of the coating, obtain multiple measurements per second, and calculate the difference between the actual coating thickness and the target thickness through the data acquisition module; Infrared temperature sensors are installed in each temperature zone of the drying section, arranged on the top and both sides of the drying box to sense the surface temperature of the coating layer in real time. Thermocouples are installed inside the drying box and on the substrate to collect temperature data of the bottom layer of the substrate to reflect the changes in heat conduction between the substrate and the coating. Multiple thermocouples are placed at the entrance and exit of the drying section to monitor the changes in the initial heating and final temperature of the coating layer. Install an infrared spectrometer in the exhaust pipe of the drying box to collect real-time data on the solvent concentration in the exhaust gas; install a humidity sensor at the air inlet or sampling port of the drying box to monitor changes in external humidity; Use distributed data acquisition nodes to collect coating thickness, bottom layer temperature, surface temperature, and solvent evaporation rate data, and synchronize the asynchronously collected multimodal data with timestamps, and remove noise through Kalman filtering; Dynamic Bayesian Network (DBN) was used to infer the potential relationship among coating thickness, temperature fluctuation and solvent concentration.

3. The temperature control intelligent optimization method for adhesive tape production according to claim 2 is characterized in that: The potential relationship between coating thickness, temperature fluctuation and solvent concentration is inferred by using a dynamic Bayesian network DBN, as follows: Assume T t represents the substrate coating thickness at time t; W t represents the local temperature distribution at time t; C t represents the local solvent evaporation concentration at time t; S t Indicates the degree of dryness of the coating at time t; According to the causal characteristics of the production process, a network diagram of the dynamic Bayesian network is constructed: Define the conditional probability relationship of the Bayesian network. For time t, the overall joint probability is expressed as: ; The above joint probability is extended to multiple time slices to describe the joint distribution under time dynamic evolution: ； Where T represents T time steps; They represent the substrate coating thickness sequence, local temperature sequence, local solvent evaporation concentration sequence and drying degree sequence from time 1 to t respectively; In real-time production, the variable states at the current time t are inferred using a forward-backward algorithm: ； According to the observed data of {Tt, Wt, Ct}, the drying completion probability is calculated: ； Among them, C target is the target value of local solvent evaporation concentration; The time series model is used to describe the dynamic evolution of variables over time and reveal the dependencies between multiple time slices.

4. The temperature control intelligent optimization method for adhesive tape production according to claim 1 is characterized in that: The step S3 is specifically as follows: Each temperature control area corresponds to an intelligent agent, which adjusts control parameters by sensing the state of its area and the state of neighboring areas to make resource allocation decisions; A centralized training framework is used to simulate the dynamic behavior of the drying process using a simulator in the offline stage to train the global strategy model; agents learn collaborative optimization strategies by sharing global feedback rewards; In online operation, each agent only perceives the current local state and performs actions independently.

5. The temperature control intelligent optimization method for adhesive tape production according to claim 4 is characterized in that: The intelligent agent is specifically as follows: The state description of the i-th agent includes: ; in, is the coating thickness of area i; is the temperature distribution in region i; is the solvent concentration in region i; is the environmental parameter; Actions of agent i Including control parameter adjustment: ； in, The hot air velocity changes; The target temperature adjustment amount for the current area; is the drying oven outlet temperature adjustment; The immediate reward for each agent Taking drying effect and energy efficiency into consideration, the specific definition is: ; Among them, C target is the target value of solvent concentration; T target is the temperature target value; For energy consumption; , and is the weight coefficient; The policy-based reinforcement learning algorithm PPO is used to optimize the agent's strategy, enabling the agent to dynamically adjust control parameters.

6. The temperature control intelligent optimization method for adhesive tape production according to claim 5, characterized in that: The strategy of the agent is optimized by using the policy-based reinforcement learning algorithm PPO, so that the agent can dynamically adjust the control parameters, as follows: Initialize the policy network π θ (A∣S) and Value Network The weight parameters are: Run the current policy πθ in the environment, sample T time steps of data, and record the state S t 、Action A t , Reward R t , new state S t+1 , calculate the advantage function Y t : ； in, Status The value network output below; is the learning parameter; Using the policy objective function L PPO (θ) Update the policy network weights θ: ； ; Among them, α π is the learning rate; is the ratio of the new and old strategies; To limit the scope of policy updates, is a preset constant; represents the expected function; Use the value network loss function L value Update value network weights : ; ; in, is the learning rate of the value network; Repeat the above steps until the strategy converges.

7. The temperature control intelligent optimization method for adhesive tape production according to claim 1 is characterized in that: The S4 is specifically: The system monitors the ambient temperature Tenv and the ambient humidity Henv, and combines key data of the drying section, including substrate surface temperature and coating speed; The collected environmental data is fuzzified and input into the fuzzy control rule base; fuzzy reasoning is performed through the rule base to determine the gradient adjustment value ΔT that is suitable for the current environment set and hot air adjustment value ΔF set ; The specific parameter adjustment value of the fuzzy output is calculated by the center of gravity method, and the adjustment value is fed back to the drying section temperature control module to complete the compensation adjustment in real time.

8. A temperature control intelligent optimization system for tape production, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, the method specifically executes the steps in the temperature control intelligent optimization method for tape production as described in any one of claims 1 to 7.

Images