Ceramic card production control method and system based on AI real-time feedback

Abstract

The invention provides a ceramic card production control method and system based on AI real-time feedback, and the method comprises the following steps: obtaining multi-dimensional production data in a ceramic card production process, and transmitting the multi-dimensional production data to a data processing center in real time for preprocessing; according to the method, multi-dimensional data are accurately collected through a sensor, data association of each link is analyzed by means of a graph neural network, an accurate scheme is generated, a model is constructed by using a digital twin module to carry out a virtual experiment to provide an optimization strategy, and the quality is predicted by relying on an AI intelligent decision model in combination with real-time data and the optimization scheme. The system automatically and quickly generates an adjustment instruction and feeds back the instruction to the equipment through edge calculation, and meanwhile, the reinforcement learning algorithm is used for dynamically scheduling the equipment to run, so that the problems of raw material errors, manual experience dependence, uneven temperature control and the like are effectively solved, and real-time dynamic adjustment in the production process is realized; the performance consistency, the quality stability and the overall quality of the product are improved, and the production requirements for high quality and high performance of the ceramic card are met.

Description

Technical Field

[0001] The present invention relates to the technical field of ceramic card production, and in particular to a ceramic card production control method and system based on AI real-time feedback. Background Art

[0002] Ceramic cards are a critical component with widespread and critical applications in numerous fields, including electronics and communications. In the electronics field, they are extensively used in various integrated circuit boards. Their excellent insulation properties and stable physical characteristics provide reliable support and electrical isolation for electronic components, ensuring the stable operation of electronic equipment. In the communications field, especially in cutting-edge communication technologies such as 5G base station construction and satellite communications, ceramic cards play a key role in signal transmission and processing. Their superior high-frequency characteristics and low signal loss ensure high-speed, stable signal transmission, greatly improving communication quality and efficiency. For this reason, the production quality of ceramic cards is directly related to the performance and service life of related equipment, while production efficiency influences the industry's development speed and market competitiveness, making them a core focus of the industry. However, many problems still exist in the traditional production process of ceramic cards.

[0003] 1. During the raw material processing stage, the manual adjustment of the raw material mixing ratio has large errors, which affects the consistency of product performance; Second, during the molding process, such as hot pressing, the temperature, pressure and time parameters are set based on manual experience, resulting in significant fluctuations in the quality of different batches of products; 3. During the firing process, the complex and uneven temperature field in the kiln makes the product prone to over- or under-firing problems, resulting in a high defective rate; Fourth, the traditional production model has scattered data collection and delayed analysis, which makes it impossible to adjust parameters in time during the production process and is difficult to meet the current demand for high quality and high performance of ceramic cards; To this end, a ceramic card production control method and system based on AI real-time feedback is proposed. Summary of the Invention

[0004] In view of this, the embodiments of the present invention hope to provide a ceramic card production control method and system based on AI real-time feedback to solve or alleviate the technical problems existing in the prior art and at least provide a beneficial option.

[0005] To solve the above technical problems, a technical solution adopted in this application is: a ceramic card production control method based on AI real-time feedback, comprising the following steps: Step 1: Obtain multi-dimensional production data during the ceramic card production process and transmit it to the data processing center in real time for pre-processing; Step 2: Based on the pre-processed multi-dimensional production data, build and train an AI intelligent decision-making model based on deep reinforcement learning; Step 3: Use graph neural networks to model the production process, analyze the correlation between multi-dimensional production data in each link, and generate a production optimization plan; Step 4: Based on the production optimization plan, a digital twin model of the ceramic card production process is constructed, and virtual experiments are used to provide optimization suggestions and response strategies for actual production; Step 5: Based on the multi-dimensional production data obtained in real time, and according to the production optimization plan, optimization suggestions and response strategies, the AI intelligent decision-making model predicts the quality of the ceramic card and generates a prediction result; Step 6: Based on the prediction results, preset rules, and preset algorithms, the system automatically generates adjustment instructions and uses edge computing technology to feed the adjustment instructions back to the production equipment; Step 7: Based on the adjustment instructions, use the reinforcement learning algorithm to dynamically schedule the operation of production equipment and adjust the equipment scheduling strategy according to the real-time status.

[0006] As a further preferred embodiment of the present technical solution, in step 2, the method for constructing and training an AI intelligent decision-making model based on deep reinforcement learning comprises the following steps: Step 201: Design an AI intelligent decision-making model based on the Actor-Critic architecture and determine the number of nodes in the input layer, hidden layer, and output layer; Step 202: Convert the pre-processed multi-dimensional production data into a format suitable for AI intelligent decision-making model training and divide it into a training set, a validation set, and a test set; Step 203: Design a reward function based on product quality, production efficiency, and production cost; Step 204: Use the training set data to train the AI intelligent decision-making model, and during the training process, regularly use the validation set data to evaluate the performance of the model; Step 205: Use the test set data to evaluate the trained AI intelligent decision-making model.

[0007] As a further preferred embodiment of the present technical solution, in step 3, the production optimization solution includes the ratio of raw material mixing, pressure and temperature setting of the molding equipment, heating curve and time control of the firing process, and process parameter adjustment of the printed circuit; The method for generating a production optimization plan comprises the following steps: Step 301: Construct a production flow chart structure, treating each link of ceramic card production as a node in the graph, and the relationship between each link as an edge in the graph, and assigning corresponding multi-dimensional production data as attributes to the nodes and edges; Step 302: Based on the message passing mechanism of the graph neural network, iteratively update the node features on the production flow chart structure to mine the relevant information of the production links; Step 303: Based on the production link association information learned by the graph neural network, establish a production index evaluation model to measure the overall performance of the production process; Step 304: Through an optimization algorithm, optimize the production link parameters on the premise of meeting production process constraints and actual production conditions; Step 305: Generate a production optimization plan according to the optimized production link parameters.

[0008] As a further optimization of this technical solution, in step four, the method for constructing the digital twin model of the ceramic card production process includes the following steps: Step 401: According to the production optimization plan, abstractly model the production equipment, process flow, and product characteristics of ceramic card production to construct the basic architecture of the digital twin model; Step 402: Establish a real-time data synchronization mechanism to transmit multi-dimensional production data in the production process to the digital twin model in real time; Step 403: Define multiple virtual experiment scenarios in the digital twin model; Step 404: By running virtual experiments, utilize the simulation calculation ability of the digital twin model to predict the production results under virtual experiment scenarios, and generate optimization suggestions and response strategies for actual production.

[0009] As a further optimization of this technical solution, in step five, when the AI intelligent decision-making model predicts the quality of ceramic cards, it adopts the Bayesian network inference algorithm, uses the production optimization plan, optimization suggestions, and response strategies as input variables, and together with the multi-dimensional production data obtained in real time, constructs a quality prediction model and outputs the quality prediction results of ceramic cards under the current production conditions.

[0010] As a further optimization of this technical solution, in step six, the preset algorithm adopts a fuzzy logic-based control algorithm, takes the quality prediction deviation and deviation change rate as inputs, and generates adjustment instructions through the processes of fuzzification, fuzzy inference, and defuzzification.

[0011] As a further optimization of this technical solution, in step one, the multi-dimensional production data is obtained by installing sensors at key links of ceramic card production. The multi-dimensional production data includes raw material weight, raw material composition, green body forming pressure, temperature curve, firing duration, and detection indicators; the preprocessing includes noise reduction processing, normalization processing, and filtering processing.

[0012] To solve the above technical problems, another technical solution adopted in this application is: a ceramic card production control system based on AI real-time feedback. The system includes: a data acquisition and processing module, an AI intelligent decision-making module, a production process optimization module, a digital twin module, a quality prediction module, an instruction generation and feedback module, and a device scheduling module; The data acquisition and processing module is configured to obtain multi-dimensional production data during the production process of ceramic cards and transmit it to the data processing center in real time for preprocessing; The AI intelligent decision-making module is configured to construct and train an AI intelligent decision-making model based on deep reinforcement learning based on the preprocessed multi-dimensional production data; The production process optimization module is configured to model the production process using a graph neural network, analyze the correlation of multi-dimensional production data in each link, and generate a production optimization plan; The digital twin module is configured to construct a digital twin model of the ceramic card production process based on the production optimization plan, and provide optimization suggestions and coping strategies for actual production through virtual experiments; The quality prediction module is configured to predict the quality of ceramic cards based on the real-time obtained multi-dimensional production data, the production optimization plan, as well as the optimization suggestions and coping strategies, and generate a prediction result using the AI intelligent decision-making model; The instruction generation and feedback module is configured to automatically generate adjustment instructions based on the prediction result, based on preset rules and preset algorithms, and use edge computing technology to feedback the adjustment instructions to the production equipment; The device scheduling module is configured to dynamically schedule the operation of production equipment according to the adjustment instructions using a reinforcement learning algorithm, and adjust the device scheduling strategy according to the real-time state.

[0013] Preferably, as a further improvement of this technical solution, the system further includes a monitoring and warning module. The monitoring and warning module is used to monitor multi-dimensional production data and quality prediction results in real time. If the multi-dimensional production data exceeds the preset range or the quality prediction result shows a quality problem, the warning mechanism will be automatically triggered.

[0014] Preferably, as a further improvement of this technical solution, the data acquisition and processing module has a data encryption and backup function.

[0015] Due to the adoption of the above technical solutions in the embodiments of the present invention, it has the following advantages: 1. The present invention accurately collects multi-dimensional data such as the weight and composition of raw materials by using sensors, and uses a graph neural network to model the production process to generate an accurate raw material mixing ratio plan. The AI intelligent decision-making model performs real-time control based on this data and plan, thereby ensuring the accurate raw material ratio of each batch of products, greatly improving the consistency of product performance, and effectively solving the problem of unstable product quality caused by raw material errors. 2. The present invention uses a graph neural network to deeply analyze the data correlation of each link, generates an accurate pressure and temperature setting plan for the forming equipment, and uses a reinforcement learning algorithm to dynamically schedule the operation of the equipment. It no longer relies on manual experience, makes the forming process more stable, significantly reduces the quality difference between batches, and improves the quality stability of products. 3. The present invention obtains data such as the firing temperature curve and duration in real time, and uses a digital twin module to build a model for virtual experiments to provide optimization strategies for actual production. At the same time, the AI intelligent decision-making model predicts the quality combined with real-time data, and the system automatically generates adjustment instructions to accurately control the firing equipment, effectively improving the temperature control effect in the kiln, greatly reducing the defective rate caused by overfiring or underfiring, and improving the overall quality of products. 4. The present invention predicts the quality through the AI intelligent decision-making model based on real-time data and optimization plans, and the system quickly generates adjustment instructions and feeds them back to the equipment through edge computing, realizing real-time dynamic adjustment of the production process and meeting the high-quality and high-performance production requirements of ceramic cards.

[0016] The above summary is only for the purpose of the specification and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present invention will be readily apparent by reference to the drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of the production control method for ceramic cards based on AI real-time feedback of the present invention; Figure 2 It is a schematic flowchart of the method for constructing and training the AI intelligent decision-making model of the present invention; Figure 3 It is a schematic flowchart of the method for generating a production optimization plan of the present invention; Figure 4Schematic flowchart of the method for constructing a digital twin model of the present invention; Figure 5 Schematic diagram of the functional modules of the ceramic card production control system based on AI real-time feedback of the present invention. Detailed implementation manners

[0019] The following describes the embodiments of the present disclosure in detail with reference to the accompanying drawings.

[0020] It should be clear that the following illustrates the implementation manners of the present disclosure through specific specific examples. Those skilled in the art can easily understand other advantages and effects of the present disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. The present disclosure can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts belong to the scope of protection of the present disclosure.

[0021] It should also be noted that the following describes various aspects of the embodiments within the scope of the appended claims. It should be obvious that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on the present disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, this device can be implemented and this method can be practiced using other structures and / or functions in addition to one or more of the aspects described herein.

[0022] It also needs to be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present disclosure in a schematic manner. Only the components related to the present disclosure are shown in the diagrams, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in its actual implementation can be an arbitrary change, and the component layout type may also be more complex.

[0023] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0024] Figure 1It is a schematic flowchart of the ceramic card production control method based on AI real-time feedback in the embodiments of the present invention. It should be noted that if there are substantially the same results, the method of the present application is not limited to Figure 1 the process sequence shown. For example, Figures 1-4 as shown: The ceramic card production control method based on AI real-time feedback includes the following steps: Step 1: Obtain multi-dimensional production data during the ceramic card production process and transmit it to the data processing center in real time for preprocessing; Specifically, first, deploy various types of sensors at each key link of the ceramic card production; in the raw material storage and feeding area, install a high-precision electronic scale to accurately measure the weight of the raw materials, and use a spectral analyzer to analyze the raw material composition in real time; in the forming stage, equip the hot pressing forming equipment with a pressure sensor to monitor the forming pressure of the green body, install a thermocouple to record the temperature change, and at the same time use a timer to obtain the forming time data; in the firing link, install multiple temperature sensors at different positions inside the kiln to construct a temperature curve, and record the firing duration through a timer; in the detection area, use professional detection equipment to obtain various detection index data of the ceramic card, such as hardness, dielectric constant, etc.; Then, through wired (such as industrial Ethernet) or wireless (such as 5G, Wi-Fi) communication technologies, the data collected by each sensor is packaged; to ensure the accuracy and stability of data transmission, data verification technology (such as CRC verification) is used to perform integrity verification on the transmitted data; the data is transmitted to the data processing center in real time in the form of data packets according to a specific communication protocol (such as Modbus protocol), and during the transmission process, the data is encrypted (such as using the AES encryption algorithm) to prevent data leakage and tampering; After the data processing center receives the multi-dimensional production data, it first performs noise reduction processing; uses the wavelet noise reduction algorithm to remove abnormal fluctuation data caused by factors such as sensor noise and electromagnetic interference, and retains the characteristics of real and effective production data; then, performs normalization processing on the data, maps data with different ranges and dimensions to the [0,1] interval, makes the data comparable, and is convenient for subsequent model training and analysis; for example, maps the pressure data from the actual pressure value (unit: MPa) to the [0,1] interval; finally, uses filtering processing to further optimize the data, adopts the Kalman filtering algorithm, and based on the historical information and current measurement values of the data, performs dynamic filtering on the data to improve the accuracy and stability of the data, providing a reliable data basis for subsequent analysis and decision-making based on these data.

[0025] Step 2: Based on the preprocessed multi-dimensional production data, construct and train an AI intelligent decision-making model based on deep reinforcement learning; Specifically, first, select the Actor-Critic architecture as the structure of the AI intelligent decision-making model. This architecture combines the advantages of policy gradient (Actor) and value function (Critic), enabling more efficient learning and decision-making. At the same time, determine the number of nodes in the input layer, hidden layer, and output layer according to the number of features in the preprocessed multi-dimensional production data. For example, if the preprocessed data includes 10 features such as raw material weight, raw material composition ratio, green body forming pressure, and temperature curve, then the number of input layer nodes is set to 10. Set multiple hidden layers to increase the model's expressive ability. Generally, 2-3 hidden layers can be set, and the number of nodes in each hidden layer can be adjusted according to experience and experiments. For example, set 64 nodes in the first hidden layer and 32 nodes in the second hidden layer. The number of output layer nodes is determined according to the number of decisions the model needs to make. For example, if the model needs to make decisions on 5 aspects such as raw material mixing ratio, forming equipment parameters, and firing parameters, then the number of output layer nodes is set to 5. Then, convert the preprocessed multi-dimensional production data into a format suitable for training the AI intelligent decision-making model. Usually, deep reinforcement learning models use data in tensor (Tensor) format, so the data needs to be converted from the original array or table form to a tensor. Divide the converted data into a training set, a validation set, and a test set, generally in a ratio of 70%-20%-10%, that is, 70% of the data is used to train the model, 20% of the data is used to verify the model's performance during training, and 10% of the data is used to finally test the model's generalization ability. Next, design the reward function according to product quality, production efficiency, and production cost. The design of the reward function should be able to guide the model to make decisions in the direction of improving product quality, enhancing production efficiency, and reducing production cost. Subsequently, input the training set data into the AI intelligent decision-making model, and let the model learn and make decisions according to the input data and the reward function. During the training process, the Actor network generates actions based on the current state, the Critic network evaluates the value of this action, and then updates the parameters of the Actor and Critic networks according to the evaluation results. During the training process, regularly use the validation set data to evaluate the model's performance. Some evaluation metrics such as accuracy and mean square error can be used to measure the model's performance. If the performance on the validation set no longer improves, it means that the model has overfitted, and some measures need to be taken, such as adjusting the model architecture and increasing regularization terms. Finally, after the model training is completed, use the test set data to evaluate the trained AI intelligent decision-making model. Through the evaluation results on the test set, the generalization ability of the model, that is, the performance of the model on unseen data, can be more accurately understood. If the performance on the test set is good, it means that the model has good generalization ability and can be applied to actual production.

[0026] Step 3: Use graph neural networks to model the production process, analyze the correlation between multi-dimensional production data in each link, and generate a production optimization plan; Specifically, we first used a graph neural network (GNN) to model the ceramic card production process. We considered each link in the production process (such as raw material mixing, molding, firing, and printed circuits) as nodes in the graph, and the mutual influence between each link (such as the raw material mixing ratio affecting the molding effect, and the molding parameters affecting the firing process) as edges in the graph. We then assigned multi-dimensional production data to these nodes and edges as attributes. Then, GNN analyzes the graph structure through specific algorithms (such as message passing algorithms) to explore the complex relationships between multi-dimensional production data at each stage. For example, it can discover that a certain raw material mixing ratio leads to a higher probability of defective products in the subsequent molding and firing processes, and that different molding parameters affect the temperature distribution during firing. Finally, based on these analysis results, a comprehensive production optimization plan is generated, including adjusting the raw material mixing formula, optimizing the pressure and temperature parameters of the molding equipment, and improving the heating curve of the firing process to improve overall production efficiency and product quality.

[0027] Step 4: Based on the production optimization plan, a digital twin model of the ceramic card production process is constructed, and virtual experiments are used to provide optimization suggestions and response strategies for actual production; Specifically, based on the production optimization plan, a comprehensive and detailed abstraction of the physical system of ceramic card production was conducted, covering multiple aspects such as production equipment, process flow, and product characteristics. These abstracted components were then integrated to build the infrastructure of the digital twin model. This defined the model's hierarchical structure, data interfaces, and interaction mechanisms to ensure that the model accurately reflected the operation of the physical system. Then, sensors are rationally deployed at key locations in the ceramic card production site to collect multi-dimensional production data in real time. This data includes but is not limited to the weight and composition of raw materials, pressure and temperature during the molding process, and temperature curve and time during the firing process. At the same time, a stable and reliable data transmission network is established, such as wired Ethernet or wireless Wi-Fi or 5G networks, to transmit the data collected by the sensors to the digital twin model in real time and accurately. Next, based on the actual production process and experience, the variation range of key parameters in each production link was determined. Virtual experimental scenarios with different parameter combinations were designed to analyze the types of failures that may occur in ceramic card production equipment. At the same time, considering the impact of external environmental factors on ceramic card production, different external environmental conditions were set to simulate the impact of changes in environmental factors on the production process and product quality. Subsequently, according to the designed virtual experiment scenario, the experiment is run in the digital twin model. The model simulates the operation of the production process based on the internal physical model and logical relationships, outputs the corresponding production results, and uses data analysis methods to deeply analyze the virtual experiment results, find out the variation laws of the production process under different parameter combinations, fault scenarios and environmental conditions, as well as the influence mechanism on product quality and production efficiency; Finally, based on the analysis of the virtual experiment results, corresponding optimization suggestions are put forward for different scenarios, and strategies to deal with various abnormal situations are designed, including emergency treatment plans for equipment failures, response measures for external environmental changes, etc.

[0028] Step 5: Based on the multi-dimensional production data obtained in real time, according to the production optimization plan, optimization suggestions and response strategies, the AI intelligent decision-making model predicts the quality of ceramic cards and generates prediction results; Specifically, first, continuously collect multi-dimensional production data from sensors at each key link of ceramic card production, covering raw material weight, composition, green body forming pressure, temperature curve, firing duration, and detection indicators, etc. These data reflect the real-time state of the current production process; and combine the production optimization plan, optimization suggestions and response strategies with the real-time production data. The production optimization plan stipulates the ideal parameter settings for each production link, the optimization suggestions are adjustment measures for different situations, and the response strategies are methods to handle abnormal conditions; Then, extract the features closely related to the quality of ceramic cards from the integrated data. For example, extract features such as the highest temperature, heating rate, and holding time from the temperature curve; extract features such as the content ratio of key chemical components from the raw material composition data; and use feature selection algorithms (such as correlation analysis, principal component analysis, etc.) to select the features that have a greater impact on quality prediction, reduce redundant information, and improve the calculation efficiency and prediction accuracy of the model; Next, input the extracted and selected feature data into the trained AI intelligent decision-making model. This model can be a model based on deep reinforcement learning, or other machine learning or deep learning models; the model predicts the quality of ceramic cards under the current production conditions based on the input data, combined with the relationship between the production data and quality learned by itself. The prediction content includes whether the pass rate of the product, key performance indicators (such as hardness, dielectric constant, etc.) meet the standards, etc.; Finally, present the prediction results in a quantitative manner. For example, the qualified rate of the ceramic card is predicted to be 85%, and the probability of the hardness reaching 80 HRA is 90%. These quantitative results can intuitively reflect the product quality under the current production conditions. Then, display the prediction results in visual forms such as charts and reports, which is convenient for production management personnel to intuitively understand the product quality situation. For example, use a line chart to show the change trend of the qualified rate of products in different batches, and use a bar chart to compare the compliance of different performance indicators. In addition, it is necessary to regularly evaluate the accuracy of the prediction results. By comparing with the product quality inspection results in actual production, calculate the prediction error. If the prediction error is large, it is necessary to analyze the reasons, further optimize the model, or there may be new situations in the production process.

[0029] Step 6: Based on the prediction results, the system automatically generates adjustment instructions according to preset rules and algorithms, and uses edge computing technology to feedback the adjustment instructions to the production equipment. Specifically, first, the system receives the quality prediction results of the ceramic card from Step 5, deeply analyzes the key information therein, such as whether the product quality meets the standard, the specific types of quality defects, and the predicted values of relevant performance indicators. Then, compare the prediction results with the preset rules. For example, if the prediction results show that the hardness of the ceramic card is lower than the standard value, and the preset rule stipulates that when the hardness deviation is between 5% - 10%, the firing temperature needs to be increased; if the deviation exceeds 10%, in addition to increasing the firing temperature, the forming pressure also needs to be adjusted. These rules are formulated based on the process knowledge, historical experience, and quality standards of ceramic card production. At the same time, according to the prediction results and the matching rules, trigger the corresponding preset algorithm. In this embodiment, the preset algorithm adopts a fuzzy logic-based control algorithm, which takes the quality prediction deviation and the deviation change rate as inputs, and generates precise adjustment instructions through the processes of fuzzification, fuzzy reasoning, and defuzzification. For example, if the quality prediction deviation is "large" and the deviation change rate is "rising", the algorithm outputs an instruction to "increase the firing temperature by 10°C and increase the forming pressure by 5 MPa". Then, the system sends the generated adjustment instructions to the edge computing device. After receiving the instructions, the edge computing device converts the instructions into signals that the device can recognize according to the communication protocol of the production equipment (such as Modbus, OPCUA, etc.), and quickly feedbacks the adjustment instructions to the corresponding production equipment through wired (such as industrial Ethernet) or wireless (such as 5G, Wi-Fi) communication methods. After receiving the adjustment instructions, the production equipment immediately executes the corresponding operations. For example, the firing equipment adjusts the heating power according to the instructions to change the firing temperature; the forming equipment adjusts the pressure parameters according to the instructions to ensure that the blank forming process meets the optimized requirements. Through this real-time feedback mechanism, the deviation in the production process can be quickly corrected, ensuring the production quality of the ceramic card. Among them, edge computing technology can perform data processing and analysis on the edge side close to the data source or user, with advantages such as low latency, high bandwidth, and privacy protection. In the production of ceramic cards, production equipment is distributed at various positions in the workshop. If data is transmitted to a remote data center for processing and then the instructions are returned, it will cause significant latency and affect production efficiency. Edge computing devices (such as industrial gateways, intelligent sensors, etc.) can directly process data at the production site and quickly generate adjustment instructions.

[0030] Step 7: According to the adjustment instructions, use the reinforcement learning algorithm to dynamically schedule the operation of the production equipment and adjust the equipment scheduling strategy according to the real-time state; Specifically, first, determine the state space that describes the real-time state of the production equipment, which includes multi-dimensional information such as the operating parameters of the equipment (such as temperature, pressure, rotation speed, etc.), the working state of the equipment (such as idle, running, faulty, etc.), the progress of the production task (such as the output already completed, the remaining production time, etc.), and the quality prediction results. For example, for a molding equipment, its state can be expressed as (current pressure, current temperature, number of molded products, predicted quality pass rate); at the same time, clarify the set of actions that can be taken on the production equipment, and these actions include adjusting the operating parameters of the equipment (such as changing the raw material feeding speed, adjusting the molding pressure, modifying the firing temperature curve, etc.), starting and stopping the equipment, task allocation of the equipment, etc. For example, for a group of firing equipment, there can be actions such as "raising the firing temperature of equipment A by 5°C" and "assigning the next production task to equipment B"; and design a reasonable reward function to evaluate the pros and cons of each action. The reward function should comprehensively consider factors such as product quality, production efficiency, and production cost. For example, if the product quality improves, the production efficiency increases, and the production cost decreases after the adjustment instructions, a positive reward is given; conversely, if the product quality deteriorates, the production efficiency decreases, or the production cost increases, a negative reward is given; specifically, if the product pass rate increases by 5%, a reward of +10 points can be given; if the equipment energy consumption increases by 10%, a reward of -5 points is given. Then, select a suitable reinforcement learning algorithm, such as the Deep Q-Network (DQN), policy gradient algorithms (such as A2C, A3C), etc. These algorithms can learn the optimal strategy in a complex environment. For example, the DQN algorithm approximates the action value function through a neural network and can handle high-dimensional state spaces and action spaces; at the same time, initialize the parameters of the agent, including the weights of the neural network, learning rate, discount factor, etc. These parameters will affect the learning speed and performance of the agent. For example, set the learning rate to 0.001 and the discount factor to 0.9 to balance short-term rewards and long-term rewards. Next, the agent continuously perceives the state of the production equipment and selects an action from the action space according to the current state. For example, when the agent perceives that the pressure of the molding equipment is lower than the pressure required by the adjustment instruction, it selects the action of "increasing the molding pressure by 2 MPa"; it sends the selected action to the production equipment, and the equipment makes corresponding adjustments according to the action. For example, after the molding equipment receives the instruction to increase the pressure, it adjusts the hydraulic system to raise the pressure to the specified value; after executing the action, the agent obtains the reward and the new equipment state feedback from the environment. The reward reflects the impact of this action on the production process, and the new state provides a basis for the next decision-making. For example, after increasing the molding pressure, the qualified rate of the product increases, and the agent obtains a positive reward, and at the same time perceives the new equipment state (new pressure value, new number of molded products, etc.); Subsequently, based on the obtained reward and new state, the agent updates its own policy using the reinforcement learning algorithm. For example, in the DQN algorithm, the agent stores the current state, action, reward, and new state in the experience replay buffer, and then randomly samples a batch of data from the buffer for learning to update the weights of the neural network to improve the estimation accuracy of the action value function; as the learning progresses, the agent continuously adjusts its policy so that it can select better actions in different states. For example, through multiple learning, the agent learns how to adjust the operating parameters of the equipment to obtain the maximum reward under different quality prediction results and equipment states; Finally, continuously monitor the real-time state of the production equipment and changes in the production environment, such as equipment failures, abnormal raw material supply, changes in market demand, etc.; when new situations occur, the agent re-evaluates and adjusts the scheduling policy according to the real-time state. For example, if a certain piece of equipment fails, the agent needs to quickly reassign the production tasks on this equipment to other available equipment and adjust the operating parameters of the relevant equipment to ensure the continuity of production and product quality.

[0031] In one embodiment, specifically, in step two, the method for constructing and training an AI intelligent decision-making model based on deep reinforcement learning includes the following steps: Step 201: Design an AI intelligent decision-making model based on the Actor-Critic architecture and determine the number of nodes in the input layer, hidden layer, and output layer; Among them, the Actor-Critic architecture combines the advantages of policy gradient (Actor) and value function (Critic) and can learn and make decisions more effectively; the Actor network is responsible for generating action policies, that is, determining what operations to take according to the current production state; the Critic network evaluates the value of the actions generated by the Actor network and guides the update of the Actor network; The number of nodes in the input layer depends on the number of features of the multi-dimensional production data after preprocessing. For example, if the preprocessed data contains 15 features such as raw material weight, raw material composition ratio, forming pressure, forming temperature, firing time, firing temperature curve, etc., then the number of nodes in the input layer is set to 15; Generally, 2 - 3 hidden layers are set. The number of nodes in the first hidden layer can be set to 2 - 3 times the number of nodes in the input layer, and the number of nodes in subsequent hidden layers gradually decreases. For example, if there are 15 nodes in the input layer, the first hidden layer can be set to 32 nodes, and the second hidden layer to 16 nodes. The hidden layer uses a non-linear activation function (such as ReLU) to increase the expressive power of the model; The number of nodes in the output layer is determined by the number of decisions the model needs to make. If the model needs to make decisions on 6 aspects such as raw material mixing ratio, forming equipment parameters (pressure, temperature), firing equipment parameters (heating curve, holding time), etc., then the number of nodes in the output layer is set to 6.

[0032] Step 202: Convert the preprocessed multi-dimensional production data into a format suitable for training the AI intelligent decision-making model, and divide it into a training set, a validation set, and a test set; Specifically, the preprocessed data is usually structured tabular data, which needs to be converted into a tensor format suitable for training deep learning models. In Python, libraries such as NumPy or PyTorch can be used for conversion. For example, use NumPy to convert the data into a multi-dimensional array, and then use PyTorch to convert it into a tensor; According to the common ratio, the converted data is divided into a training set (70%), a validation set (20%), and a test set (10%). The train_test_split function in the Scikit-learn library can be used for division. First, the data is divided into a training set and a temporary set in a 7:3 ratio, and then the temporary set is divided into a validation set and a test set in a 2:1 ratio.

[0033] Step 203: Design a reward function based on product quality, production efficiency, and production cost; Among them, the reward function should be able to guide the model to make decisions in the direction of improving product quality, enhancing production efficiency, and reducing production cost. By giving different reward values to different production results, the model can learn the optimal decision-making strategy; A specific design example of the reward function is as follows: Product quality: If all performance indicators of the produced ceramic cards meet the high-quality standards (such as hardness, dielectric constant, etc. within the specified range), a positive reward (such as +20 points) is given; if there are defective or waste products, a negative reward (such as -30 points) is given; Production efficiency: A positive reward is given for an increase in production efficiency (e.g., an increase in the number of products produced per unit time). The reward value can be calculated based on the proportion of the efficiency increase. For example, if the production efficiency increases by 10%, a reward of +15 points is given; a negative reward is given for a decrease in production efficiency. Production cost: A positive reward is given for a decrease in production cost (e.g., a reduction in raw material consumption or energy consumption). The reward value is related to the extent of the cost reduction. For example, if the production cost is reduced by 15%, a reward of +12 points is given; a negative reward is given for an increase in production cost. The final reward value is the weighted sum of the reward values in three aspects: product quality, production efficiency, and production cost. The weights can be adjusted according to the focus of actual production.

[0034] Step 204: Use the training set data to train the AI intelligent decision-making model, and during the training process, regularly use the validation set data to evaluate the performance of the model. Specifically, first, input the training set data into the AI intelligent decision-making model and use Stochastic Gradient Descent (SGD) or its improved algorithm (such as Adam) for training. In each training step, the Actor network generates actions based on the current production status, the Critic network evaluates the value of the action, and then updates the parameters of the Actor and Critic networks according to the evaluation results. Then, every certain number of training rounds (e.g., 10 rounds), use the validation set data to evaluate the performance of the model. The evaluation metrics can include the average reward value, prediction accuracy, etc. If the performance on the validation set no longer improves, it indicates that the model is overfitting, and some measures need to be taken, such as adjusting the model architecture, adding regularization terms (such as L1 or L2 regularization), etc.

[0035] Step 205: Use the test set data to evaluate the trained AI intelligent decision-making model. Specifically, after the model training is completed, first use the test set data to conduct a final evaluation of the trained AI intelligent decision-making model. The test set data is data that the model has never seen during the training process and can more realistically reflect the generalization ability of the model. Then, evaluate the model using the same evaluation metrics as the validation set (such as the average reward value, prediction accuracy). If the performance on the test set is good, it indicates that the model has good generalization ability and can be applied to actual production; if the performance is poor, further analysis is required to identify the reasons and make adjustments and improvements to the model.

[0036] In one embodiment, specifically, in step three, the production optimization plan includes the proportion of raw material mixing, the pressure and temperature settings of the forming equipment, the heating curve and time control during the firing process, and the adjustment of the process parameters of the printed circuit. The method for generating a production optimization plan includes the following steps: Step 301: Construct the production flow chart structure. Consider each link in the ceramic card production as a node in the graph, and the relationships between each link as the edges in the graph. Assign corresponding multi-dimensional production data as attributes to the nodes and edges. Specifically, first, comprehensively sort out the ceramic card production process. Set each independent production link such as raw material mixing, forming, firing, printed circuit, and quality inspection as nodes in the graph. These nodes represent the key steps in the production process, and each node carries specific production tasks and data information. Then, according to the sequence and logical relationship of the production process, determine the connecting edges between each node. For example, after the raw material mixing link is completed, it enters the forming link, which constructs a directed edge from the raw material mixing node to the forming node to represent the sequence and dependency relationship between production links. Finally, for each node, assign corresponding multi-dimensional production data. For example, the attributes of the raw material mixing node cover the ratio of various raw materials, mixing time, operating parameters of the mixing equipment, etc.; the attributes of the forming node include forming pressure, temperature, time, and the size specifications of the green body, etc.; for the attributes of the edge, assign data such as the time of material transfer between adjacent links, logistics cost, and delay time of information transfer, so as to completely describe the state and mutual relationship of each link in the production process.

[0037] Step 302: Based on the message passing mechanism of the graph neural network, perform iterative update of node features on the production flow chart structure to mine the associated information of production links. Specifically, first, in the graph neural network, set the initial values for the initial feature vectors of each node. These initial values are based on the multi-dimensional production data assigned to the nodes in Step 301. For example, encode the raw material mixing ratio data as part of the initial feature vector of the raw material mixing node. Then, according to the message passing mechanism of the graph neural network, at each iteration, each node will collect the messages passed from its adjacent nodes. These messages contain the feature information of the adjacent nodes. The node will fuse these messages with its own original features. For example, the forming node will receive the messages passed from the raw material mixing node and the firing node, and then update its own feature vector according to a specific fusion rule (such as weighted summation). Finally, after multiple iterative updates, the feature vectors of the nodes will continuously evolve and gradually contain more associated information about the production links. For example, through this iterative update, it is found that there is a potential association between the temperature parameter in the firing link and the content of a certain additive in the raw material mixing link. This associated information is not obvious in the initial data but is mined through the learning of the graph neural network.

[0038] Step 303: Based on the production link association information learned by the graph neural network, establish a production index evaluation model to measure the overall performance of the production process; Specifically, first, select a suitable model architecture to build the production index evaluation model, such as a multi-layer perceptron (MLP) or other regression models. The input of the model is the feature vectors of each node learned by the graph neural network, and these feature vectors have integrated the association information of the production links; Then, clarify the production indexes to be measured, mainly including product quality, production efficiency, and production cost. The product quality index can be measured by the qualified rate of the product, key performance indicators (such as the insulation performance and signal transmission loss of ceramic cards, etc.); the production efficiency index can be reflected by the output per unit time, equipment utilization rate, etc.; the production cost index covers raw material cost, energy consumption cost, equipment maintenance cost, etc.; Finally, use historical production data to train the production index evaluation model. By adjusting the parameters of the model (such as weights and biases), make the output of the model close to the actual production index value. During the training process, adopt methods such as cross-validation to ensure the generalization ability of the model and avoid overfitting, so as to accurately measure the overall performance of the production process.

[0039] Step 304: Through an optimization algorithm, optimize the production link parameters on the premise of meeting the production process constraints and actual production conditions; Specifically, first, select a suitable optimization algorithm, such as a genetic algorithm, particle swarm optimization algorithm, or simulated annealing algorithm, etc., to optimize the production link parameters. These algorithms can search for the optimal solution in a complex solution space and meet the requirements of optimizing production link parameters; Then, set the constraint conditions of the optimization algorithm according to the production process requirements of ceramic cards and actual production conditions. For example, in the raw material mixing link, the ratio of raw materials must be within the safe and process-feasible range; the pressure and temperature in the forming link cannot exceed the rated value of the equipment; the heating rate and holding time in the firing link need to meet the characteristics requirements of ceramic materials, etc. At the same time, the resource limitations in actual production, such as the production capacity of equipment and the allocation of manpower, etc., also need to be considered; Finally, take the production index evaluation model as the objective function of the optimization algorithm. The algorithm continuously adjusts the production link parameters on the premise of meeting the constraint conditions. For example, the genetic algorithm iteratively optimizes the production link parameters by simulating the process of natural selection and genetic variation. Each iteration will generate a new set of parameter combinations, and calculate the corresponding production index value through the production index evaluation model, and select the parameter combination with a better index value for the next iteration until the preset optimization termination condition is reached (such as the index value no longer improves or the improvement amplitude is extremely small).

[0040] Step 305: Generate a production optimization plan based on the optimized production process parameters. Specifically, first, extract the optimized production process parameters from the optimal solution obtained by the optimization algorithm, and organize these parameters in the order of the production process to form a detailed parameter list. For example, determine the precise ratio of various raw materials in the raw material mixing process, the optimal pressure and temperature settings of the molding equipment, the optimized heating curve and time arrangement during the firing process, and the process parameters of the printed circuit process, etc. Then, based on the organized parameters, write a complete production optimization plan. The plan should not only include the parameter settings of each production process but also explain the basis and expected effects of adjusting these parameters. For example, in the plan, explain that adjusting the raw material mixing ratio is to improve the stability of the product, and it is expected that the defective rate of the product can be reduced by a certain percentage through this adjustment; optimizing the parameters of the molding equipment is to improve production efficiency, and it is expected that the output per unit time will increase by how much, etc. Finally, conduct a small-scale verification test on the generated production optimization plan in the actual production environment. By comparing the changes in production indicators before and after the test, evaluate the effectiveness of the plan. If it is found that there is a deviation between the actual effect and the expectation, analyze the reasons and make corresponding adjustments to the plan to ensure that the production optimization plan can effectively improve the production quality and efficiency of ceramic cards and reduce production costs.

[0041] In one embodiment, specifically, in step four, the method for constructing a digital twin model of the ceramic card production process includes the following steps: Step 401: Abstractly model the production equipment, process flow, and product characteristics of ceramic card production according to the production optimization plan to construct the basic framework of the digital twin model. Specifically, first, conduct a detailed study of various types of equipment involved in ceramic card production, such as raw material mixers, molding presses, firing kilns, etc. Thoroughly analyze the mechanical structure, electrical system, and working principle of the equipment, and extract key features and parameters. For example, for a raw material mixer, abstract parameters such as the shape, size, and rotation speed of its mixing paddles, and the volume of the mixing chamber. Based on the working principle and historical operation data of the equipment, establish a behavior model of the equipment. Taking a firing kiln as an example, according to the power of its heating elements and heat transfer characteristics, establish a mathematical model of the temperature inside the kiln changing with time and heating power to simulate the temperature distribution in the kiln under different working conditions. Then, according to the process flow in the production optimization plan, break down the ceramic card production process into multiple specific links, such as raw material preparation, forming, firing, surface treatment, etc., and clarify the inputs, outputs, and operation requirements of each link; determine the sequence, parallel relationship, and dependency relationship between production links, and construct a logical model of the process flow; for example, the forming link must be carried out after the raw material mixing is completed, and some surface treatment processes need to be carried out after the firing is completed and the product cools to a certain temperature; Next, analyze the physical properties of the ceramic card, such as size, density, hardness, dielectric constant, etc., and establish a relationship model between the product physical properties and production process parameters according to the product design requirements and actual production data; for example, study the influence laws of forming pressure and temperature on the density and hardness of the ceramic card; consider the functional requirements of the ceramic card in actual applications, such as signal transmission performance, insulation performance, etc., and establish a mapping relationship between the product functional characteristics and production process parameters, so as to simulate the functional performance of the product under different production conditions in the digital twin model; Finally, integrate the above models of production equipment, process flow, and product characteristics to construct the basic architecture of the digital twin model, define the hierarchical structure, data interface, and interaction mechanism of the model, and ensure that each part of the model can work together to accurately reflect the physical system of ceramic card production.

[0042] Step 402, establish a real-time data synchronization mechanism to transmit multi-dimensional production data in the production process to the digital twin model in real time; Specifically, first, deploy sensors at various key positions in the ceramic card production site to collect multi-dimensional data in the production process in real time; for example, install weight sensors on the raw material storage tanks to monitor the raw material weight, install pressure and temperature sensors on the forming equipment to obtain forming parameters, install thermocouples in the kiln to measure the temperature curve, etc.; at the same time, build a stable and reliable data transmission network to transmit the data collected by the sensors to the data processing center in real time. Wired networks (such as Ethernet) or wireless networks (such as Wi-Fi, 5G) can be used for data transmission to ensure the timeliness and accuracy of data transmission; Then, in the data processing center, clean and preprocess the collected raw data, remove noise, outliers, and duplicate data to improve the data quality; and convert the processed data into a format that the digital twin model can recognize and process. According to the interface requirements of the digital twin model, perform operations such as encoding and normalization on the data to ensure the consistency and compatibility of the data; Next, adopt real-time data push technology to send the processed data to the digital twin model in a timely manner to update the model status. Technologies such as message queues and data buses can be used to achieve efficient data push and distribution. And according to the characteristics of the production process and actual requirements, set a reasonable data synchronization frequency. For parameters that change relatively quickly (such as furnace temperature), a higher synchronization frequency can be set; for parameters that change relatively slowly (such as raw material inventory), the synchronization frequency can be appropriately reduced.

[0043] Step 403: Define multiple virtual experiment scenarios in the digital twin model; Specifically, first, according to the actual situation and experience of the production process, determine the change range of key parameters in each production link. For example, the change range of the raw material mixing ratio can be within ±5%, and the change range of the forming pressure can be within ±10%. And design virtual experiment scenarios with different parameter combinations to simulate various changes in process parameters during the production process. For example, set scenarios with different raw material mixing ratios such as A:B:C = 3:2:1, 3.1:2:0.9, 2.9:2.1:1, etc., to study the impact of different ratios on product quality. Then, analyze the types of faults that occur in the ceramic card production equipment, such as the failure of the stirring paddle of the raw material mixer, the unstable pressure of the forming press, and the damage of the heating element in the furnace. And simulate the equipment failure scenarios in the digital twin model, setting the time, location, and severity of the failure. For example, simulate the scenario where a heating element in the furnace fails at the 30th minute during the firing process, resulting in a local temperature drop. Finally, consider the impact of external environmental factors on ceramic card production, such as temperature, humidity, air pressure, etc., and analyze the change range of these environmental factors in different seasons and different regions. Set different external environmental conditions in the digital twin model to simulate the impact of environmental factor changes on the production process and product quality. For example, simulate the impact of increased raw material humidity on the forming and firing processes in a high-temperature and high-humidity environment.

[0044] Step 404: Through running virtual experiments, utilize the simulation calculation ability of the digital twin model to predict the production results in the virtual experiment scenarios, and generate optimization suggestions and coping strategies for actual production; Specifically, first, according to the defined virtual experiment scenarios, set the corresponding experimental parameters in the digital twin model, including process parameters, equipment status, external environmental conditions, etc. Start the simulation calculation function of the digital twin model, and simulate the production process according to the set experimental parameters. The model calculates the production results in different scenarios, such as product quality indicators, production efficiency, production cost, etc., based on its internal physical model and logical relationships. Then, evaluate the production results obtained from the virtual experiment, compare various production indicators under different scenarios. For example, analyze the pass rate of the product under different raw material mixing ratio scenarios, the decline in production efficiency under equipment failure scenarios, etc.; and use data analysis methods such as statistical analysis and machine learning algorithms to deeply explore the laws and trends in the production result data. For example, find the quantitative relationship between process parameters and product quality through regression analysis; Finally, based on the production result analysis, for different virtual experiment scenarios, put forward corresponding optimization suggestions. For example, if the product quality is significantly improved under a certain combination of process parameters, it is recommended to adopt this parameter combination in actual production; if it is found that equipment failure will cause serious production losses, it is recommended to strengthen equipment maintenance and fault warning; at the same time, design strategies to deal with various abnormal situations, including emergency handling plans for equipment failures and response measures for external environment changes. For example, formulate a rapid replacement plan for kiln heating elements when they fail, and measures to adjust the raw material drying process in high-temperature and high-humidity environments; feedback these optimization suggestions and response strategies to the actual production process to guide production decision-making and operation adjustment.

[0045] In one embodiment, specifically, in step five, when the AI intelligent decision-making model predicts the quality of the ceramic card, it uses the Bayesian network inference algorithm. Taking the production optimization plan, optimization suggestions, and response strategies as input variables, together with the multi-dimensional production data obtained in real time, it constructs a quality prediction model and outputs the quality prediction result of the ceramic card under the current production conditions; Among them, the Bayesian network is a probabilistic graphical model. It represents the dependence relationship between variables through a directed acyclic graph (DAG) and uses a conditional probability table (CPT) to quantify these relationships; in the quality prediction of ceramic cards, the inference ability of the Bayesian network will be used to calculate the probability distribution of the quality of the ceramic card according to the known input variables; The specific way to output the quality prediction result of the ceramic card under the current production conditions is as follows: First, determine the input variables, including the production optimization plan, optimization suggestions and response strategies from the virtual experiment of the digital twin model, and the production process data collected in real time through sensors; Among them, the production optimization plan includes the optimized parameters of the production links, such as the raw material mixing ratio, the pressure and temperature settings of the forming equipment, the heating curve and time of the firing process, etc. These parameters are the basis for ensuring product quality and have a direct impact on the quality of the ceramic card; The optimization suggestions and response strategies from the virtual experiment of the digital twin model, such as adjustment measures in certain abnormal situations (such as equipment failure, raw material fluctuations), can help the model better adapt to different production scenarios and improve the accuracy of quality prediction; The production process data collected in real time by sensors, including the actual weight and composition of raw materials, the real-time pressure and temperature during the forming process, and the temperature curve during the firing process, reflect the actual situation of the current production and are an important basis for quality prediction; Then, construct a Bayesian network structure, defining the input variables and the quality situation of the ceramic card as nodes in the Bayesian network. For example, the raw material mixing ratio, forming pressure, firing temperature, etc. are used as input nodes, and the quality indicators of the ceramic card (such as hardness, dielectric constant, pass rate, etc.) are used as output nodes; and according to the causal relationship between variables, determine the directed edges between nodes. For example, the raw material mixing ratio will affect the forming process and thus affect the quality of the ceramic card. Therefore, there is a directed edge from the raw material mixing ratio node to the forming pressure node and then to the quality indicator node; Next, use historical production data to statistically calculate the occurrence probability of the quality indicators of the ceramic card under different combinations of input variable values. For example, statistically calculate the probability that the hardness of the ceramic card is qualified under different raw material mixing ratios and forming pressures; and use methods such as maximum likelihood estimation to estimate the conditional probability table of each node in the Bayesian network according to the statistical data. The conditional probability table describes the probability that the node takes different values given the values of its parent nodes; Subsequently, take the multi-dimensional production data, production optimization plans, optimization suggestions, and countermeasures obtained in real time as evidence and input them into the Bayesian network. These evidences will update the probability distributions of the nodes in the network; use Bayesian network inference algorithms (such as variable elimination method, belief propagation algorithm, etc.) to calculate the posterior probability distribution of the quality indicators of the ceramic card according to the input evidence and conditional probability table. For example, calculate the probability that the hardness of the ceramic card is qualified under the current production conditions; Finally, output the quality prediction results. The quality prediction model outputs the probability distribution of the quality indicators of the ceramic card. For example, predict that the probability of the hardness of the ceramic card being qualified is 80%, and the probability of the dielectric constant being qualified is 85%; according to the quality prediction results, production managers can make corresponding decisions. If the predicted probability of unqualified quality is relatively high, they can timely adjust production parameters or take countermeasures to improve product quality.

[0046] In one embodiment, specifically, in step six, the preset algorithm adopts a fuzzy logic-based control algorithm, taking the quality prediction deviation and the deviation change rate as inputs, and generating adjustment instructions through the processes of fuzzification, fuzzy inference, and defuzzification; Among them, the fuzzy logic control algorithm is an intelligent control method based on fuzzy set theory and fuzzy inference. It does not require an accurate mathematical model and can handle complex non-linear systems and uncertainty problems; in the production of ceramic cards, taking the quality prediction deviation and the deviation change rate as inputs, generating adjustment instructions through the fuzzy logic control algorithm to achieve real-time dynamic adjustment of the production process; The specific method for generating adjustment instructions is as follows: First, use the quality prediction deviation and the deviation change rate as inputs; Among them, the quality prediction deviation refers to the difference between the actual quality of the ceramic card and the quality predicted by the quality prediction model. For example, if the quality prediction model predicts the hardness of the ceramic card to be 80 HRA, and the actual detected hardness is 78 HRA, then the quality prediction deviation is -2 HRA; The deviation change rate is the change rate of the quality prediction deviation over time, which reflects the change trend of the quality deviation. For example, if the quality prediction deviations detected twice in a row are -2 HRA and -3 HRA respectively, then the deviation change rate is -1 HRA / time; Then, define fuzzy sets for the quality prediction deviation and the deviation change rate respectively. For example, for the quality prediction deviation, fuzzy sets such as "Negative Big (NB)", "Negative Medium (NM)", "Negative Small (NS)", "Zero (ZO)", "Positive Small (PS)", "Positive Medium (PM)", "Positive Big (PB)" can be defined; for the deviation change rate, similar fuzzy sets can also be defined; and determine the membership function for each fuzzy set, which is used to describe the degree to which the input variable belongs to a certain fuzzy set. Common membership functions include triangular membership functions, trapezoidal membership functions, etc. For example, for the "Negative Small (NS)" fuzzy set of the quality prediction deviation, its membership function can be defined as a triangle. When the quality prediction deviation is between -3 and -1, the membership degree linearly increases from 0 to 1; when it is between -1 and 1, the membership degree linearly decreases from 1 to 0; at the same time, according to the input quality prediction deviation and deviation change rate, calculate their membership degrees belonging to each fuzzy set. For example, if the quality prediction deviation is -2 HRA, according to the membership function of the "Negative Small (NS)" fuzzy set, calculate its membership degree to be 0.8; Next, based on the experience and knowledge of ceramic card production, formulate fuzzy rules. For example, Rule 1: If the quality prediction deviation is "Negative Big (NB)" and the deviation change rate is "Negative Medium (NM)", then the adjustment instruction is "Greatly increase the proportion of raw material A"; Rule 2: If the quality prediction deviation is "Positive Small (PS)" and the deviation change rate is "Zero (ZO)", then the adjustment instruction is "Slightly reduce the forming pressure", etc.; and match the fuzzy rules according to the membership degrees obtained by fuzzyfication. For example, if the membership degree of the quality prediction deviation belonging to "Negative Small (NS)" is 0.8, and the membership degree of the deviation change rate belonging to "Positive Small (PS)" is 0.6, find the fuzzy rules containing "Negative Small (NS)" and "Positive Small (PS)", and calculate the output fuzzy set of the rules according to the premise conditions and membership degrees of the rules; synthesize the output fuzzy sets of all matching rules to obtain the final output fuzzy set. Common synthesis methods include the maximum-minimum method, the maximum-product method, etc.; Finally, an exact adjustment instruction value is obtained from the final output fuzzy set. Common defuzzification methods include the centroid method, the maximum membership degree method, etc. For example, using the centroid method, calculate the centroid position of the output fuzzy set and use it as the exact value of the adjustment instruction; and according to the exact value obtained by defuzzification, combined with the actual production situation, generate specific adjustment instructions. For example, if the result obtained by defuzzification is that the proportion of raw material A needs to be increased by 5%, then generate an adjustment instruction of "increase the proportion of raw material A by 5%".

[0047] In one embodiment, specifically, in step one, multi-dimensional production data is obtained by installing sensors at key links in the production of ceramic cards. The multi-dimensional production data includes raw material weight, raw material composition, green body forming pressure, temperature curve, firing duration, and detection indicators. Among them, the key links include the raw material link, the green body forming link, the firing link, and the detection link. In the raw material link, by installing high-precision weight sensors on the raw material storage tanks, the weight change of the raw materials is monitored in real time, so that the input amount of the raw materials can be accurately grasped to ensure the accurate proportion of the raw materials in the production process; at the same time, equipment such as a spectral analyzer is used to analyze the raw material composition. The spectral analyzer can quickly and accurately determine the content of various chemical components in the raw materials, providing important basic data for subsequent production links. In the green body forming link, by installing pressure sensors on the forming equipment, the pressure during the green body forming process can be accurately measured. Appropriate forming pressure is crucial for the density and strength of the green body. By monitoring the pressure in real time, the equipment parameters can be adjusted in time to ensure the quality of the green body; at the same time, temperature sensors are arranged to obtain the temperature curve. The temperature curve reflects the change of temperature with time during the forming process. Different ceramic materials and forming processes have different requirements for the temperature curve. Accurately grasping the temperature curve helps to optimize the forming process. In the firing link, by installing multiple thermocouples in the kiln furnace to measure the temperature curve, the temperature control during the firing process is one of the key factors affecting the quality of ceramic cards. Through multiple thermocouples, the temperature changes at different positions in the kiln furnace can be comprehensively and accurately monitored to ensure that the firing process meets the process requirements; at the same time, a timer is used to record the firing duration. The firing duration interacts with factors such as temperature and raw materials and jointly affects the performance of ceramic cards. Accurately recording the firing duration can provide a basis for subsequent quality analysis and process optimization. In the detection link, professional detection equipment is used to obtain detection indicators. These detection indicators include physical property indicators such as the hardness, density, and dielectric constant of ceramic cards, as well as quality indicators such as appearance defects. The detection indicators are the direct basis for evaluating the quality of ceramic cards and can timely detect problems existing in the production process.

[0048] The preprocessing includes noise reduction processing, normalization processing, and filtering processing. Among them, wavelet denoising algorithm is used for noise reduction processing of data. The wavelet denoising algorithm can effectively separate signals and noise according to the different characteristics of signals and noise in the wavelet transform domain, and retain the main features of the signals. For example, for the temperature data collected by the temperature sensor, after wavelet denoising processing, the high-frequency noise generated by electromagnetic interference can be removed, making the temperature curve smoother and better reflecting the actual temperature change situation. The normalization processing adopts the minimum-maximum normalization method to map the data to the interval [0,1]. The formula is: ; Where is the original data, and are the minimum and maximum values of this type of data respectively, is the normalized data. For example, for the raw material weight data, it is normalized to the interval [0,1] through this formula, making the weight data of different raw materials comparable. The filtering processing adopts the Kalman filtering algorithm for filtering. The Kalman filtering algorithm is a recursive optimal estimation method. It uses the state equation and observation equation of the system, and uses the current observation value and the estimated value of the previous moment to make an optimal estimate of the state of the system. For example, for the pressure data collected by the pressure sensor, after Kalman filtering processing, a more stable and accurate pressure value can be obtained, reducing the pressure fluctuation caused by sensor measurement errors and environmental interference.

[0049] In summary, the ceramic card production control method based on AI real-time feedback provided by the embodiments of the present invention, through a series of operations such as multi-dimensional data collection and preprocessing, constructing and training an AI intelligent decision-making model, using a graph neural network to generate an optimization plan, simulating experiments with the digital twin model, predicting quality based on real-time data, automatically generating adjustment instructions, and using a reinforcement learning algorithm to schedule equipment, effectively solves the problems of large raw material allocation errors, forming parameters relying on manual experience, high firing defective product rate, and lagging data processing in the traditional ceramic card production process, realizes the real-time dynamic adjustment of the production process, ensures the accurate raw material ratio, stable forming process, and reliable firing quality of each batch of products, significantly improves the product performance consistency, quality stability, and overall quality, meets the strict production requirements for high-quality and high-performance ceramic cards in the fields of electronics, communication, etc., and provides a strong technical support for the intelligent upgrading of the ceramic card production industry, with important practical significance and application value.

[0050] Figure 5 is the schematic diagram of the functional modules of the ceramic card production control system based on AI real-time feedback in the embodiments of the present application, as Figure 5As shown in the figure, a ceramic card production control system based on AI real-time feedback, the system includes: a data acquisition and processing module, an AI intelligent decision-making module, a production process optimization module, a digital twin module, a quality prediction module, an instruction generation and feedback module, and a device scheduling module; The data acquisition and processing module is configured to acquire multi-dimensional production data during the production process of ceramic cards and transmit it to the data processing center in real time for preprocessing; The AI intelligent decision-making module is configured to construct and train an AI intelligent decision-making model based on deep reinforcement learning based on the preprocessed multi-dimensional production data; The production process optimization module is configured to model the production process using a graph neural network, analyze the correlation of multi-dimensional production data in each link, and generate a production optimization plan; The digital twin module is configured to construct a digital twin model of the ceramic card production process based on the production optimization plan, and provide optimization suggestions and coping strategies for actual production through virtual experiments; The quality prediction module is configured to predict the quality of ceramic cards based on the real-time acquired multi-dimensional production data, according to the production optimization plan, optimization suggestions and coping strategies, and the AI intelligent decision-making model, and generate a prediction result; The instruction generation and feedback module is configured to automatically generate adjustment instructions based on the prediction result, based on preset rules and preset algorithms, and use edge computing technology to feedback the adjustment instructions to the production equipment; The device scheduling module is configured to dynamically schedule the operation of production equipment according to the adjustment instructions using a reinforcement learning algorithm, and adjust the device scheduling strategy according to the real-time state.

[0051] In one embodiment, specifically, the system further includes a monitoring and warning module. The monitoring and warning module is used to monitor multi-dimensional production data and quality prediction results in real time. If the multi-dimensional production data exceeds the preset range or the quality prediction result shows a quality problem, the warning mechanism is automatically triggered.

[0052] In one embodiment, specifically, the data acquisition and processing module has a data encryption and backup function; during the data acquisition process, the acquired multi-dimensional production data is encrypted to prevent the data from being stolen or tampered with during transmission. At the same time, the preprocessed data is regularly backed up and stored in a secure storage device to cope with data loss or damage, ensuring the security and integrity of the system data.

[0053] Regarding other details of the implementation technical solutions of each module in the above-mentioned ceramic card production control system based on AI real-time feedback, reference can be made to the description in the above-mentioned ceramic card production control method based on AI real-time feedback in the above-mentioned embodiment, which will not be elaborated here.

[0054] It should be noted that the embodiments in this specification are all described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For related parts, reference can be made to the description of the method embodiments.

[0055] The basic principles of the present disclosure have been described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, benefits, effects, etc. mentioned in the present disclosure are only examples and not limitations. It cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present disclosure. In addition, the above-mentioned specific details are only for the purpose of illustration and easy understanding, rather than limitations. The above details do not limit the present disclosure to necessarily adopt the above specific details to implement.

[0056] In the present disclosure, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, equipment, and systems involved in the present disclosure are only illustrative examples and do not intend to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any way. Words such as "including", "comprising", "having", etc. are open-ended words, meaning "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the word "and / or", and can be used interchangeably with it, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0057] In addition, as used herein, the "or" used in the listing of items starting with "at least one" indicates a separate listing, so that for example, the listing of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (that is, A and B and C). Furthermore, the term "exemplary" does not mean that the described examples are preferred or better than other examples.

[0058] It should also be pointed out that in the systems and methods of the present disclosure, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present disclosure.

[0059] Various changes, substitutions, and alterations to the technology described herein can be made without departing from the teachings defined by the appended claims. Additionally, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and acts described above. Processes, machines, manufactures, compositions of events, means, methods, or acts that are currently available or later to be developed that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Accordingly, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or acts within their scope.

[0060] The foregoing description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0061] The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present disclosure to the form disclosed herein. Although several example aspects and embodiments have been discussed above, those skilled in the art will recognize some of their variations, modifications, alterations, additions, and subcombinations.

Claims

1. A ceramic card production control method based on real-time AI feedback, characterized in that The following steps are involved: Step 1: Obtain multi-dimensional production data during the ceramic card production process and transmit it to the data processing center in real time for pre-processing; Step 2: Based on the pre-processed multi-dimensional production data, build and train an AI intelligent decision-making model based on deep reinforcement learning; Step 3: Use graph neural networks to model the production process, analyze the correlation between multi-dimensional production data in each link, and generate a production optimization plan; Step 4: Based on the production optimization plan, a digital twin model of the ceramic card production process is constructed, and virtual experiments are used to provide optimization suggestions and response strategies for actual production; Step 5: Based on the multi-dimensional production data obtained in real time, and according to the production optimization plan, optimization suggestions and response strategies, the AI intelligent decision-making model predicts the quality of the ceramic card and generates a prediction result; Step 6: Based on the prediction results, preset rules, and preset algorithms, the system automatically generates adjustment instructions and uses edge computing technology to feed the adjustment instructions back to the production equipment; Step 7: Based on the adjustment instructions, use the reinforcement learning algorithm to dynamically schedule the operation of production equipment and adjust the equipment scheduling strategy according to the real-time status.

2. The ceramic card production control method based on AI real-time feedback according to claim 1, wherein In step 2, the method for building and training an AI intelligent decision-making model based on deep reinforcement learning includes the following steps: Step 201: Design an AI intelligent decision-making model based on the Actor-Critic architecture and determine the number of nodes in the input layer, hidden layer, and output layer; Step 202: Convert the pre-processed multi-dimensional production data into a format suitable for AI intelligent decision-making model training and divide it into a training set, a validation set, and a test set; Step 203: Design a reward function based on product quality, production efficiency, and production cost; Step 204: Use the training set data to train the AI intelligent decision-making model, and during the training process, regularly use the validation set data to evaluate the performance of the model; Step 205: Use the test set data to evaluate the trained AI intelligent decision-making model.

3. The ceramic card production control method based on AI real-time feedback according to claim 1, wherein, In step three, the production optimization plan includes the mixing ratio of raw materials, the pressure and temperature setting of the molding equipment, the heating curve and time control of the firing process, and the process parameter adjustment of the printed circuit; The method for generating a production optimization plan comprises the following steps: Step 301: Construct a production flow chart structure, treating each link of ceramic card production as a node in the graph, and the relationship between each link as an edge in the graph, and assigning corresponding multi-dimensional production data as attributes to the nodes and edges; Step 302: Based on the message passing mechanism of the graph neural network, iteratively update the node features on the production flow chart structure to mine the relevant information of the production links; Step 303: Based on the production link correlation information learned by the graph neural network, a production indicator evaluation model is established to measure the overall performance of the production process; Step 304: Optimize the production process parameters by using an optimization algorithm while satisfying the production process constraints and actual production conditions; Step 305: Generate a production optimization plan based on the optimized production link parameters.

4. The ceramic card production control method based on AI real-time feedback according to claim 1, characterized in that, In step 4, the method for constructing a digital twin model of the ceramic card production process includes the following steps: Step 401: According to the production optimization plan, abstractly model the production equipment, process flow, and product characteristics of ceramic card production to construct the infrastructure of the digital twin model; Step 402: Establish a real-time data synchronization mechanism to transmit multi-dimensional production data during the production process to the digital twin model in real time; Step 403: Define multiple virtual experiment scenarios in the digital twin model; Step 404: By running virtual experiments, utilize the simulation calculation ability of the digital twin model to predict the production results under virtual experiment scenarios, and generate optimization suggestions and coping strategies for actual production.

5. The ceramic card production control method based on AI real-time feedback according to claim 1, characterized in that: In Step 5, when the AI intelligent decision-making model predicts the quality of ceramic cards, it adopts the Bayesian network inference algorithm. Using the production optimization plan, optimization suggestions, and coping strategies as input variables, together with the multi-dimensional production data obtained in real time, it constructs a quality prediction model and outputs the quality prediction results of ceramic cards under the current production conditions.

6. The ceramic card production control method based on AI real-time feedback according to claim 1, characterized in that: In Step 6, the preset algorithm adopts a control algorithm based on fuzzy logic. Using the quality prediction deviation and deviation change rate as inputs, it generates adjustment instructions through the processes of fuzzification, fuzzy inference, and defuzzification.

7. The ceramic card production control method based on AI real-time feedback according to claim 1, characterized in that: In Step 1, the multi-dimensional production data is obtained by installing sensors at key links in ceramic card production. The multi-dimensional production data includes raw material weight, raw material composition, green body forming pressure, temperature curve, firing duration, and detection indicators; the preprocessing includes noise reduction processing, normalization processing, and filtering processing.

8. A ceramic card production control system based on AI real-time feedback, which is applied to the ceramic card production control method based on AI real-time feedback according to any one of claims 1-7, characterized in that, The system includes: a data acquisition and processing module, an AI intelligent decision-making module, a production process optimization module, a digital twin module, a quality prediction module, an instruction generation and feedback module, and a device scheduling module; The data acquisition and processing module is configured to obtain multi-dimensional production data during the production process of ceramic cards and transmit it to the data processing center for preprocessing in real time; The AI intelligent decision-making module is configured to construct and train an AI intelligent decision-making model based on deep reinforcement learning based on the preprocessed multi-dimensional production data; The production process optimization module is configured to model the production process using a graph neural network, analyze the correlation of multi-dimensional production data in each link, and generate a production optimization plan; The digital twin module is configured to construct a digital twin model of the ceramic card production process based on the production optimization plan, and provide optimization suggestions and coping strategies for actual production through virtual experiments; The quality prediction module is configured to predict the quality of ceramic cards based on the multi-dimensional production data obtained in real time, according to the production optimization plan, optimization suggestions, and coping strategies, and the AI intelligent decision-making model, and generate prediction results; The instruction generation and feedback module is configured to automatically generate adjustment instructions based on the prediction results, based on preset rules and preset algorithms, and use edge computing technology to feedback the adjustment instructions to the production equipment; The device scheduling module is configured to dynamically schedule the operation of production equipment according to the adjustment instructions using a reinforcement learning algorithm, and adjust the device scheduling strategy according to the real-time state.

9. The ceramic card production control system based on AI real-time feedback according to claim 8, characterized in that: The system further includes a monitoring and warning module, which is used to monitor multi-dimensional production data and quality prediction results in real time. If the multi-dimensional production data exceeds the preset range or the quality prediction results indicate quality problems, the warning mechanism will be automatically triggered.

10. The ceramic card production control system based on AI real-time feedback according to claim 8, characterized in that: The data acquisition and processing module has the functions of data encryption and backup.

Images