Fan mold cluster type adaptive precision temperature control system

Abstract

The application belongs to the technical field of intelligent manufacturing and industrial automation control, and particularly relates to a fan mold cluster type self-adaptive precise temperature control system. The system comprises a thermal field sensing and modeling module, a dynamic scheduling generation module, a collaborative optimization module, a fault prediction and reconstruction module and a data closed-loop management module. The system constructs a multi-mode cavity thermal coupling topological model through a graph neural network, realizes intelligent distribution of cooling medium in combination with an improved particle swarm algorithm, predicts equipment faults based on an LSTM to trigger redundant switching, and finally realizes safe and efficient data management through an edge-cloud collaborative architecture. The application can improve the temperature control precision and energy efficiency of the cluster mold, and takes into account the local temperature control stability and global operation reliability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and industrial automation control technology, specifically a cluster-type adaptive precision temperature control system for fan molds. Background Technology

[0002] With the development of high-end manufacturing and precision molding processes, the importance of mold temperature control systems in industrial fields such as injection molding and die casting is becoming increasingly prominent. Especially in the production of multi-cavity molds for complex structural components such as fans, higher demands are placed on the uniformity of temperature in each cavity, response speed, and adaptive adjustment capabilities. Although existing technologies offer temperature control solutions for molds or air-cooling systems, they still have significant shortcomings when facing scenarios involving multi-mold cluster collaboration, dynamic heat load changes, and high-precision temperature field control.

[0003] A search revealed a patent, CN119840047B, which discloses a method and system for precise temperature control of mold pipelines, published on July 4, 2025. This patent achieves precise temperature control of the mold pipelines by setting a water tank in the mold's water circuit and adjusting the water temperature online based on the heating time under different pressures, water flow rate, and inlet / outlet water temperature difference. However, this solution mainly relies on empirical data under preset operating conditions (such as the heating time at maximum heating power) for temperature adjustment, lacking a real-time perception and adaptive feedback mechanism for the operating status of multiple mold clusters. Furthermore, its control logic does not consider factors such as thermal interference between molds, environmental disturbances, or changes in production cycle time, making it difficult to achieve dynamic collaborative temperature control between multiple mold cavities and failing to meet the high-precision, high-stability temperature control requirements of fan mold clusters under continuous changing operating conditions.

[0004] A search revealed a flexible duct precision temperature control system and its control method, with publication number CN118816454B, published on August 26, 2025. This patent utilizes flexible ducts embedded in the top of a cold storage facility, combined with temperature and differential pressure sensors, to collaboratively regulate the temperature uniformity within the storage area using air coolers and the cold storage plates, effectively suppressing temperature fluctuations during defrosting. While this system demonstrates innovation in the flexible layout of the air-cooling path and localized cold storage, its application is limited to static storage environments. The control objective is overall spatial temperature uniformity, rather than differentiated, highly dynamic, and precise temperature control for multiple independent heat sources (such as multiple fan molds). Furthermore, the system lacks a cluster scheduling algorithm, making it unable to adaptively allocate airflow and cooling capacity based on the real-time heat load of each mold, thus hindering its applicability to complex thermal management scenarios in industrial mold clusters.

[0005] The above problems indicate that although existing technologies have achieved precise control of single points or single systems in the field of mold temperature control, they still have the following key shortcomings when facing the application requirements of clustered adaptive precision temperature control systems for fan molds: (1) lack of real-time perception and coupled analysis capability of the thermal state of multiple molds; (2) temperature control strategy relies on static parameters and cannot adapt to dynamic changes in production conditions; (3) no intelligent flow distribution mechanism for air / liquid cooling media among multiple branches has been established; (4) lack of clustered collaborative optimization control architecture, making it difficult to balance global energy efficiency and local temperature control accuracy. Therefore, there is an urgent need for a new type of temperature control system that can achieve clustered collaboration, self-sensing, self-adjustment, and high precision for multiple fan molds to support the dual improvement requirements of molding quality and energy efficiency in high-end equipment manufacturing. Summary of the Invention

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] Therefore, this invention provides a cluster-type adaptive precision temperature control system for fan molds, which solves the problems of existing mold temperature control systems lacking real-time perception and coupling analysis capabilities of thermal state in multi-fan mold cluster operation, relying on static parameters for temperature control strategies that cannot respond to dynamic operating condition changes, lacking intelligent flow distribution mechanism for air / liquid cooling media among multiple branches, and lacking a cluster collaborative optimization control architecture that makes it difficult to balance global energy efficiency and local temperature control accuracy.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a cluster-type adaptive precision temperature control system for fan molds, comprising, The thermal field sensing and modeling module deploys a distributed temperature-pressure-flow sensor array to collect real-time thermal parameters of each mold cavity and cooling circuit, and constructs a multi-cavity thermal coupling topology model based on graph neural networks. The thermal parameters include the mold cavity surface temperature, the inlet and outlet temperatures of the cooling medium, the flow rate, the pressure, and the environmental disturbance factor. The dynamic scheduling generation module calculates the heat load weight of each mold cavity based on the thermal coupling topology model and generates an initial cooling medium allocation strategy in combination with the production cycle signal. The initial cooling medium distribution strategy includes the flow rate setting value and start / stop sequence of each cooling branch; The collaborative optimization module encodes the initial cooling medium allocation strategy into individual particles, introduces a thermal interference suppression factor and an energy efficiency constraint to construct a multi-objective fitness function, and performs iterative optimization through an improved particle swarm algorithm to output the optimal cooling scheduling scheme. The optimal cooling scheduling scheme includes the precise flow rate, start / stop status, and medium temperature settings for each cooling branch in each time slice; The fault prediction and reconstruction module predicts the status of the cooling pump, valve and sensor based on the long short-term memory network. When the predicted value exceeds the preset threshold, it triggers cooling path reconstruction and redundancy switching. The data closed-loop management module transmits optimization results and operation logs to the edge server via industrial Ethernet. The data is stored in a time-series database and key parameters are encrypted using the national cryptographic SM4 algorithm before being uploaded to the cloud management platform.

[0009] As a preferred embodiment of the fan mold cluster adaptive precision temperature control system of the present invention, the following steps are included: deploying a distributed temperature-pressure-flow sensor array to collect real-time thermal parameters of each mold cavity and cooling circuit, and constructing a multi-cavity thermal coupling topology model based on a graph neural network: PT100 platinum resistance temperature sensors are embedded in the critical areas of the mold cavity of each fan mold, turbine flow meters and piezoresistive pressure sensors are installed at the inlet and outlet of the cooling water circuit, and infrared thermal imagers are deployed on the environmental side. Each mold is defined as a graph node, and the node feature vector consists of the average cavity temperature, cooling temperature difference, flow rate fluctuation rate, and temperature rise gradient of adjacent molds. The Euclidean distance is used to calculate the thermal influence intensity between any two nodes. When the thermal influence intensity is greater than a set threshold, a directed edge is established between the two nodes, and the edge weight is the product of the thermal conductivity coefficient and the spatial distance. The constructed graph structure is input into the graph convolutional network, and the neighborhood thermal state information is aggregated through two layers of graph convolution operations to output the thermal coupling embedding vector of each node. The edge weights are updated using the cosine similarity between the embedded vectors to form a dynamic thermally coupled topology model.

[0010] The beneficial effects of this invention are: the system constructs a multi-cavity thermal coupling topology model through a graph neural network, combines an improved particle swarm optimization algorithm to achieve intelligent allocation of cooling medium, and uses LSTM to predict equipment failures to trigger redundancy switching. Finally, it achieves safe and efficient data management through an edge-cloud collaborative architecture. This application can improve the temperature control accuracy and energy efficiency of cluster molds, while taking into account both local temperature control stability and global operational reliability. Attached Figure Description

[0011] Figure 1 This is a system architecture block diagram of the present invention; Figure 2 This is a flowchart illustrating the collaborative optimization process of the present invention. Figure 3 This is a flowchart of the data closed-loop management process of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Specific implementation examples are given below.

[0014] The specific implementation of the fan mold cluster adaptive precision temperature control system of the present invention is as follows; Figure 1 As shown, the overall architecture of the system includes a thermal field perception and modeling module, a dynamic scheduling generation module, a collaborative optimization module, a fault prediction and reconstruction module, a data closed-loop management module, a fan mold cluster, an edge controller, an edge server, and a cloud management platform. The fan mold cluster consists of multiple parallel-arranged injection molding fan molds. Each mold has an independent cooling water system connected to a central cooling pump station via cooling medium pipes. Each mold's cooling branch outlet and inlet are equipped with an electric proportional-integral valve, whose control terminal is connected to the edge controller via a 4–20mA analog output module. Simultaneously, PT100 platinum resistance temperature sensors are embedded in key areas of each mold cavity to collect the mold cavity surface temperature. Turbine flow meters and piezoresistive pressure sensors are installed at the cooling water inlet and outlet, respectively. An infrared thermal imager is deployed on the environmental side. All the above sensing devices are connected to the fieldbus module via an M12 aviation connector and transmit the collected real-time thermal parameters to the edge controller using the PROFINET protocol. The thermal field sensing and modeling module is deployed in the edge controller, and its input is connected to the aforementioned distributed sensing array. This sensing array includes various temperature, pressure, flow rate, and infrared imaging devices to acquire real-time thermal parameters of each mold cavity and cooling circuit, specifically including mold cavity surface temperature, cooling medium inlet and outlet temperatures, flow rate, pressure, and environmental disturbance factors. During system operation, each fan mold is abstracted as a graph node, and the node feature vector consists of the average mold cavity temperature, cooling temperature difference, flow rate fluctuation rate, and temperature rise gradient of adjacent molds. The thermal influence intensity between any two graph nodes is calculated using Euclidean distance. When the intensity exceeds a preset threshold, a directed edge is established between the two nodes, with the edge weight defined as the product of the thermal conductivity coefficient and the reciprocal of the spatial distance. Subsequently, the constructed graph structure is input into a graph convolutional network, which contains two layers of graph convolutional operation units. Each layer performs weighted aggregation of the features of neighboring nodes, and finally outputs the thermally coupled embedding vector corresponding to each graph node. Based on this, the original edge weights are dynamically updated using the cosine similarity between the embedding vectors, thereby forming a dynamic thermally coupled topology model. This model is stored in the memory of the edge controller in the form of a data structure and serves as the basis for the generation of subsequent scheduling strategies. The dynamic scheduling generation module is also integrated into the edge controller. Its input receives the production cycle signal from the MES system and the dynamic thermal coupling topology model output by the thermal field perception and modeling module. In actual operation, the holding pressure duration and the start time of the cooling window in the current injection cycle are first extracted from the MES system. Then, principal component analysis is performed on the thermal coupling embedding vector corresponding to each graph node, and the first principal component is taken as the thermal sensitivity index of the cavity. Next, the thermal sensitivity index is multiplied by the cavity volume and material specific heat capacity of the corresponding mold to obtain the basic heat load value. Then, a correction coefficient based on the change of ambient temperature and a compensation term caused by thermal interference from adjacent molds are introduced to comprehensively calculate the comprehensive heat load weight of each cavity. Then, the total cooling medium flow rate is proportionally allocated according to the weight to generate the initial flow rate setting value of each cooling branch. Finally, according to the start time of the cooling window, the flow rate setting value of each branch is mapped to the start and stop timing sequence to form the initial cooling medium allocation strategy. This strategy is temporarily stored in the buffer of the edge controller in the form of a time-flow rate lookup table. The collaborative optimization module is located in the core processing unit of the edge controller. Its input is connected to the initial cooling medium allocation strategy output by the dynamic scheduling generation module. In this embodiment, the strategy is encoded as the position vector of each particle in the particle swarm. Each particle position vector is composed of the flow rate setpoint (real part) of each cooling branch and the timing switch state (Boolean part). The multi-objective fitness function consists of a weighted sum of three terms: the reciprocal of the standard deviation of the cavity temperature, the negative value of the total pump power consumption, and the negative value of the absolute value of the temperature difference between adjacent cavities. A thermal interference suppression term is introduced into the particle velocity update formula. The value of this term is equal to the product of the difference in adjacent particle positions and the weight of the corresponding thermal coupling edge. At the same time, an energy efficiency constraint boundary is set. If the pump power corresponding to a certain particle position exceeds 85% of the rated value, a penalty term is applied in the fitness calculation. The particle swarm is initialized using a chaotic mapping method to generate the initial population. In each iteration, the top 10% of the particles in terms of fitness are selected as the guides, and the remaining particles choose to fly towards the guides or the global optimal position with a certain probability. When the number of iterations reaches 200 or the fitness improvement is less than 10⁻ for 50 consecutive generations, the algorithm is called to define the optimal fitness function. 4 When the iteration process is terminated, the optimal cooling scheduling scheme is output. This scheme includes the precise flow rate, start / stop status and medium temperature setpoint of each cooling branch in each time slice, and is directly sent to each electric proportional integral valve and cooling pump control system. The fault prediction and reconfiguration module is also deployed in the edge controller. Its input is connected to the historical operation database of the cooling pump, valves, and sensors. During continuous system operation, it collects cooling pump current signals, vibration spectrum data, valve opening feedback values, and temperature sensor drift as historical operation data. A three-layer LSTM state prediction unit is constructed with an input window length of 30 minutes and an output of the state prediction value for the next 10 minutes. The Bayesian optimization algorithm is used to automatically adjust the number of hidden units and the learning rate of the LSTM network to minimize the Huber loss function on the validation set. The dynamic warning threshold judge sets the dynamic warning threshold based on three times the standard deviation of the predicted value plus the historical mean. When the predicted value of any device exceeds its corresponding warning threshold, the cooling path reconfiguration actuator immediately starts the path reconfiguration process: shutting down the electric proportional integral valve corresponding to the faulty branch, redistributing its original allocated flow to the adjacent healthy branch according to the heat load weight, and activating the redundancy switching controller to enable the standby pump group. At the same time, the fault diagnosis report generator automatically generates a fault diagnosis report containing the abnormal parameter type, occurrence time, switching log, and suggested maintenance measures, and writes it to the operation log file. The data closed-loop management module consists of an edge controller, an edge server, and a cloud management platform. The edge controller has a built-in OPC UA server that encapsulates key data such as the optimal cooling scheduling scheme, dynamic thermal coupling topology model, and fault logs into JSON format and pushes them to the edge server via gigabit industrial Ethernet. The edge server uses the InfluxDB time-series database to index and store the received data by timestamp. For sensitive information such as mold cavity temperature setpoints, flow distribution matrices, and equipment status codes, the national cryptographic algorithm SM4 is used for block encryption. The encryption key is dynamically generated by the hardware security module HSM and rotated periodically. The encrypted data is uploaded to the cloud management platform via the MQTT protocol. A two-way SSL / TLS tunnel is established between the cloud management platform and the edge server to ensure the integrity and confidentiality of the transmission process. After receiving the data, the platform parses the content and drives the WebGL 3D visualization interface to display the temperature distribution, cooling medium flow path, and equipment operating status of each mold cavity in the fan mold cluster in real time. In the practical application scenario of this embodiment, it is assumed that the fan mold cluster includes 8 injection molds arranged in parallel, each mold is equipped with an independent cooling branch, the cooling medium is deionized water, and the circulation power is provided by a main cooling pump and a standby pump. After the system is powered on, the distributed sensor array begins to collect the thermal parameters of each mold cavity and cooling circuit, and uploads them to the edge controller via the PROFINET protocol. The thermal field perception and modeling module constructs an initial dynamic thermal coupling topology model based on this. The dynamic scheduling generation module combines the current production cycle signal provided by the MES system to generate an initial cooling medium allocation strategy. The collaborative optimization module calls the improved particle swarm optimization algorithm to perform multi-objective optimization on this strategy and outputs the optimal cooling scheduling scheme. This scheme controls each electric proportional integral valve to adjust the flow rate through the 4-20mA analog output module. The fault prediction and reconstruction module continuously monitors the equipment status, and triggers path reconstruction and redundancy switching once a potential fault is detected. All operating data is stored locally and synchronized with the cloud through the data closed-loop management module to form a complete data closed loop. The entire system achieves adaptive, accurate, and efficient temperature control of the fan mold cluster without manual intervention. To enable those skilled in the art to fully understand and implement this invention, the following supplements the specific implementation principle of this invention in conjunction with a specific application scenario; In the actual operation scenario of a wind turbine mold cluster containing eight injection molds arranged in parallel, after the system starts up, the distributed sensor array first completes the full-dimensional acquisition of thermal parameters. Among them, the PT100 platinum resistance temperature sensor embedded on the surface of each mold cavity provides real-time feedback on the local temperature rise trend, the turbine flow meter and piezoresistive pressure sensor at the inlet and outlet of the cooling water channel synchronously record the changes in medium flow rate and pressure drop, and the ambient infrared thermal imager captures the radiative heat exchange information between adjacent molds. The above data is connected to the fieldbus module through the M12 aviation connector and transmitted to the edge controller via the PROFINET protocol, providing raw input for the thermal field perception and modeling module. The thermal field sensing and modeling module is based on Figure 2 The process described above constructs a dynamic thermal coupling topology model: Each mold is abstracted as a graph node, and its node feature vector integrates four dimensions: average cavity temperature, cooling temperature difference, flow fluctuation rate, and temperature rise gradient of neighboring molds; the thermal influence intensity between any two graph nodes is calculated using Euclidean distance. When this value exceeds a preset threshold (e.g., 0.35 W / m·K), a directed edge is established, with the edge weight set as the product of the thermal conductivity coefficient and the inverse of the spatial distance; then, this graph structure is input into a graph convolutional network, and the neighborhood thermal state is weighted and aggregated through two layers of graph convolution operations, outputting the thermal coupling embedding vector of each node; further, the original edge weights are dynamically corrected using the cosine similarity between the embedding vectors, thus forming a dynamic thermal coupling topology model that can evolve with the working conditions. This model accurately characterizes the asymmetric thermal interference relationship between multiple mold cavities, providing a physical constraint basis for subsequent scheduling; The dynamic scheduling generation module receives the current production cycle signal from the MES system, extracts the holding time and the start time of the cooling window, and performs initial scheduling based on the dynamic thermal coupling topology model: principal component analysis is performed on the thermal coupling embedding vector corresponding to each graph node, and the first principal component is taken as the thermal sensitivity index, which reflects the response sensitivity of the mold cavity to cooling disturbances; this index is multiplied by the mold cavity volume and the specific heat capacity of the material to obtain the basic heat load value; then the ambient temperature correction coefficient (obtained by fitting infrared thermal imager data) and the thermal interference compensation term of adjacent molds (based on the weighted summation of directed edges) are superimposed to finally generate the comprehensive heat load weight of each mold cavity; based on this, the total flow provided by the central cooling pump station is proportionally allocated to 8 independent cooling branches to form the initial flow setpoint, and mapped to a time-flow reference table according to the start time of the cooling window, i.e., the initial cooling medium allocation strategy; The collaborative optimization module encodes the initial cooling medium allocation strategy into position vectors of individual particles in the particle swarm. The real part corresponds to the flow rate setpoint of each branch, and the Boolean part represents the valve opening and closing status in each time slice. A multi-objective fitness function is constructed, with three components corresponding to cavity temperature uniformity (measured by the reciprocal of the standard deviation of the eight cavity temperatures), system energy efficiency (characterized by the negative value of the total pump power consumption), and thermal interference suppression capability (quantified by the negative value of the absolute value of the temperature difference between adjacent cavities). During the iteration of the improved particle swarm algorithm, a thermal interference suppression term is introduced into the particle velocity update. This term is equal to the product of the flow rate difference between adjacent particles in the corresponding branch and the weight of the thermal coupling edge, thereby guiding particles to avoid overcooling in the strong thermal coupling region in the search space. At the same time, a hard constraint is set that the pump power does not exceed 85% of the rated value. Particles exceeding the limit are subject to exponential penalties in the fitness calculation. The population initialized by chaotic mapping converges within 200 generations and outputs the optimal cooling scheduling scheme, which precisely specifies the opening command and cooling medium temperature setpoint of each electric proportional integral valve in each second. The fault prediction and reconfiguration module continuously monitors the health status of the equipment: The LSTM status prediction unit 20 uses the cooling pump current, vibration spectrum, valve opening feedback, and PT100 drift within a 30-minute sliding window as inputs to predict the evolution trend of key parameters in the next 10 minutes; the dynamic early warning threshold judge sets an adaptive threshold based on the historical data mean and three times the standard deviation to avoid false alarms under varying operating conditions with a fixed threshold; when the predicted flow rate of a certain branch continuously deviates from the set value by more than the threshold, the cooling path reconfiguration actuator immediately closes the electric proportional integral valve of that branch and redistributes the originally allocated flow rate to the adjacent healthy branch according to the heat load weight of the adjacent graph nodes in the dynamic thermal coupling topology model; the redundancy switching controller synchronously starts the standby pump group to ensure the stability of the total flow rate; the fault diagnosis report generator records the anomaly type, occurrence time, flow redistribution ratio, and suggested maintenance measures, and writes them to the local log; The data closed-loop management module realizes end-to-end data flow: the edge controller encapsulates the optimal cooling scheduling scheme, dynamic thermal coupling topology model and fault log into JSON format through the built-in OPC UA server, and pushes it to the edge server via gigabit industrial Ethernet; the edge server uses InfluxDB to store time-series data by timestamp index, and calls the SM4 key generated by the hardware security module HSM to encrypt sensitive fields such as mold cavity temperature setpoint and flow distribution matrix in groups; the encrypted data is uploaded to the cloud management platform through the MQTT protocol, and the platform decrypts it through the SSL / TLS tunnel and drives the WebGL engine to render the three-dimensional temperature cloud map, cooling medium flow vector and equipment operating status identifier of each mold cavity in the fan mold cluster in real time, supporting remote process optimization and preventive maintenance decisions; Through the above steps, this system has achieved technical effects such as mold cavity temperature standard deviation ≤1.2℃, pump group comprehensive energy efficiency improvement of 18.7%, and fault response delay <8 seconds in the continuous variable cycle production of 8-mold cluster. It effectively solves the problem of difficulty in coordinating and optimizing temperature control accuracy and global energy efficiency under multi-mold thermal interference coupling.

[0015] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A cluster-type adaptive precision temperature control system for fan molds, characterized in that: include, The thermal field sensing and modeling module deploys a distributed temperature-pressure-flow sensor array to collect real-time thermal parameters of each mold cavity and cooling circuit, and constructs a multi-cavity thermal coupling topology model based on graph neural networks. The thermal parameters include the mold cavity surface temperature, the inlet and outlet temperatures of the cooling medium, the flow rate, the pressure, and the environmental disturbance factor. The dynamic scheduling generation module calculates the heat load weight of each mold cavity based on the thermal coupling topology model and generates an initial cooling medium allocation strategy in combination with the production cycle signal. The initial cooling medium distribution strategy includes the flow rate setting value and start / stop sequence of each cooling branch; The collaborative optimization module encodes the initial cooling medium allocation strategy into individual particles, introduces thermal interference suppression factors and energy efficiency constraints to construct a multi-objective fitness function, and performs iterative optimization through an improved particle swarm algorithm to output the optimal cooling scheduling scheme. The optimal cooling scheduling scheme includes the precise flow rate, start / stop status, and medium temperature settings for each cooling branch in each time slice; The fault prediction and reconstruction module predicts the status of the cooling pump, valve and sensor based on the long short-term memory network. When the predicted value exceeds the preset threshold, it triggers cooling path reconstruction and redundancy switching. The data closed-loop management module transmits optimization results and operation logs to the edge server via industrial Ethernet. The data is stored in a time-series database and key parameters are encrypted using the national cryptographic SM4 algorithm before being uploaded to the cloud management platform.

2. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The deployment of a distributed temperature-pressure-flow sensor array to collect real-time thermal parameters of each mold cavity and cooling circuit, and the construction of a multi-cavity thermally coupled topology model based on a graph neural network, includes the following steps: PT100 platinum resistance temperature sensors are embedded in the critical areas of the mold cavity of each fan mold, turbine flow meters and piezoresistive pressure sensors are installed at the inlet and outlet of the cooling water circuit, and infrared thermal imagers are deployed on the environmental side. Each mold is defined as a graph node, and the node feature vector consists of the average cavity temperature, cooling temperature difference, flow rate fluctuation rate, and temperature rise gradient of adjacent molds. The Euclidean distance is used to calculate the thermal influence intensity between any two nodes. When the thermal influence intensity is greater than a set threshold, a directed edge is established between the two nodes, and the edge weight is the product of the thermal conductivity coefficient and the spatial distance. The constructed graph structure is input into the graph convolutional network, and the neighborhood thermal state information is aggregated through two layers of graph convolution operations to output the thermal coupling embedding vector of each node. The edge weights are updated using the cosine similarity between the embedded vectors to form a dynamic thermally coupled topology model.

3. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The step of calculating the thermal load weight of each mold cavity based on the thermal coupling topology model and generating the initial cooling medium distribution strategy in combination with the production cycle signal includes the following steps: Obtain the current production cycle signal from the MES system and extract the holding pressure duration and cooling window in the injection molding cycle; Principal component analysis was performed on the thermal coupling embedding vector of each cavity node, and the first principal component was taken as the thermal sensitivity index. The basic heat load value is obtained by multiplying the heat sensitivity index by the cavity volume and the specific heat capacity of the material. Introduce an ambient temperature correction factor and a thermal interference compensation term for adjacent molds to calculate the overall heat load weight; The total cooling medium flow rate is proportionally allocated according to the comprehensive heat load weight, and the initial flow rate setpoint for each branch is generated. Based on the start time of the cooling window, the flow setpoint is mapped to the start-stop timing sequence to form the initial cooling medium distribution strategy.

4. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The process of encoding the initial cooling medium allocation strategy as individual particles, introducing a thermal interference suppression factor and an energy efficiency constraint to construct a multi-objective fitness function, and iteratively optimizing using an improved particle swarm algorithm to output the optimal cooling scheduling scheme includes the following steps: The flow rate setpoint of each cooling branch is combined with the timing switch state to form a real-Boolean hybrid code, which constitutes the particle position vector; The multi-objective fitness function is defined as a weighted sum of three terms: the reciprocal of the standard deviation of the cavity temperature, the negative of the total pump power consumption, and the negative of the absolute value of the temperature difference between adjacent cavities. A thermal interference suppression term is introduced into the particle velocity update formula. This term is proportional to the positional difference between adjacent particles and the weight of the thermally coupled edge. Set energy efficiency constraint boundaries and apply a penalty term when the pump power corresponding to the particle position exceeds 85% of the rated value; The particle swarm is initialized using chaotic mapping. In each generation, the top 10% of the particles with the best fitness are selected as guides, and the remaining particles fly toward the guides or the global optimal position according to probability. When the number of iterations reaches 200 or the fitness improvement is less than 10⁻ for 50 consecutive generations. 4 When the iteration ends, the optimal cooling schedule is output.

5. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The process of predicting the status of cooling pumps, valves, and sensors based on historical operating data from long short-term memory networks, and triggering cooling path reconstruction and redundancy switching when the predicted value exceeds a preset threshold, includes the following steps: Collect historical operating data such as cooling pump current, vibration spectrum, valve opening feedback, and temperature sensor drift. Construct a three-layer LSTM network, with the input being 30 minutes of historical data within a sliding window, and the output being the state prediction value for the next 10 minutes; Bayesian optimization is used to adjust the number of hidden units and the learning rate of the LSTM to minimize the Huber loss on the validation set; Set a dynamic early warning threshold, which is three times the standard deviation of the predicted value plus the historical mean; When the predicted value of any device exceeds its corresponding warning threshold, path reconfiguration is initiated: the faulty branch is shut down, its flow is redistributed to the adjacent healthy branch according to the heat load weight, and the standby pump group is activated. At the same time, a fault diagnosis report is generated, and abnormal parameters and switching logs are recorded.

6. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The process of transmitting optimization results and operation logs to the edge server via industrial Ethernet, storing them in a time-series database, encrypting key parameters using the national cryptographic SM4 algorithm, and then uploading them to the cloud management platform includes the following steps: Configure an OPC UA server in the edge controller to encapsulate the optimal cooling scheduling scheme, thermally coupled topology model, and fault logs into JSON format; Data is pushed to an edge server via gigabit industrial Ethernet, and the server uses InfluxDB to store data indexed by timestamps. The mold cavity temperature setpoint, flow distribution matrix and equipment status code are encrypted using the SM4 algorithm, and the key is dynamically generated by the hardware security module HSM. The encrypted data is uploaded to the cloud management platform via the MQTT protocol. After being parsed, the platform drives the WebGL 3D visualization interface to display the temperature control status.

7. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The distributed temperature-pressure-flow sensor array is connected to the fieldbus module via an M12 aviation connector. The fieldbus module is connected to the edge controller using the PROFINET protocol.

8. The fan mold cluster adaptive precision temperature control system as described in claim 1, characterized in that: The flow regulating valve of the cooling branch is an electric proportional integral valve, and its control signal is driven by the edge controller through a 4–20mA analog output module.

9. The fan mold cluster adaptive precision temperature control system as described in claim 2, characterized in that: The training data for the graph neural network comes from an offline simulation platform. This platform is based on ANSYS Fluent to build a CFD model of a multi-cavity shared cooling system and generate thermal coupling datasets under different operating conditions.

10. The fan mold cluster adaptive precision temperature control system as described in claim 6, characterized in that: A two-way SSL / TLS tunnel is established between the edge server and the cloud management platform to ensure the integrity and confidentiality of data transmission.

Images