Precise intelligent temperature control system of injection molding machine

Abstract

The invention belongs to the technical field of intelligent manufacturing, and particularly relates to a precise intelligent temperature control system of an injection molding machine. Comprising an acquisition layer, a control layer and an execution layer. The acquisition layer is used for monitoring the temperature of the injection molding machine cavity in real time; the control layer comprises a central controller and is used for receiving the temperature digital signal, executing logical operation according to a pre-trained deep Q network algorithm and outputting a control signal; the execution layer comprises a heating device and a cooling device, the heating device is used for heating a material cavity of the injection molding machine, the cooling device is used for cooling the material cavity of the injection molding machine, and based on deep learning and experience playback of the DQN algorithm, the algorithm can process a complex environment with a continuous state space, and in a precise intelligent temperature control system of the injection molding machine, the temperature of the injection molding machine can be accurately controlled. The DQN algorithm is used for selecting the optimal control action according to the current temperature state so as to achieve accurate control over the temperature, defects of injection molding products caused by temperature fluctuation are reduced, and the machining quality and the quality consistency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent manufacturing, and in particular, relates to a precise intelligent temperature control system for an injection molding machine. Background Art

[0002] In the injection molding process, temperature control is a key factor in ensuring product quality and production efficiency. Traditional injection molding machine temperature control systems usually use PID control or simple feedback control methods. Although these methods can achieve certain control effects under steady-state conditions, they often show obvious deficiencies in accuracy when facing complex and changeable injection molding environments, such as changes in material properties and ambient temperature.

[0003] Traditional temperature control methods have difficulty handling complex environments with continuous state spaces and cannot adapt to nonlinear interference such as changes in material properties and ambient temperature fluctuations during the injection molding process, resulting in insufficient temperature control accuracy and poor product quality consistency. Summary of the invention

[0004] In order to solve the technical problem that traditional temperature control methods are difficult to handle complex environments with continuous state spaces and cannot adapt to nonlinear changes in the injection molding process, resulting in insufficient temperature control accuracy, the present invention provides a precise intelligent temperature control system for an injection molding machine.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A precise intelligent temperature control system for an injection molding machine comprises a collection layer, a control layer and an execution layer; the collection layer comprises a temperature sensor and a signal converter; the temperature sensor is used to monitor the temperature of the injection molding machine cavity in real time; the signal converter is used to convert the temperature current signal into a temperature digital signal; The control layer includes a central controller that receives temperature digital signals, performs logic operations based on a pre-trained deep Q network algorithm, and outputs control signals; The execution layer includes a heating device and a cooling device. The heating device is used to heat the injection molding machine cavity according to the control instructions of the central controller; the cooling device is used to cool the injection molding machine cavity according to the control instructions of the central controller.

[0006] Preferably, the construction process of the deep Q network algorithm includes: Build a deep Q network algorithm to define the state space, action space and reward function in the injection molding machine temperature control system; Preferably, the state space includes the current temperature, the temperature error and the temperature change rate; the action space includes increasing the temperature, decreasing the temperature and no compensation; the reward function is the feedback returned by the system after executing the action, including positive reward, small positive reward and negative reward; Use a neural network to gradually approximate the optimal Q-value function, which represents the expected cumulative reward for performing an action in a specific state; the calculation formula of the Q-value function is: ; In the formula, is the state of the system at time t; is the action taken by the system at time t; It is the system that is performing the action Immediate rewards after is a discount factor used to balance immediate rewards and future rewards; Among them, the neural network includes an input layer, a hidden layer and an output layer. The input layer is used to input the current state space of the system, and the number of state spaces is the number of neurons in the input layer; The hidden layer is used to perform nonlinear transformation on the input data and extract data features; The output layer is used to output the Q value of each action.

[0007] Preferably, the pre-training process of the deep Q network algorithm includes: S1) Initialize the deep Q network and randomly select the connection weights of the neural network; S2) In each sampling period t, the state space of the acquisition system , including current temperature, target temperature and temperature change rate; S3) Based on the current Q value, use the ε-greedy strategy to select an action ; S4) The system performs the selected action , and return the reward and the next state ; S5) , , , ) is stored in the experience pool; S6) After a batch of injection molding process is completed, a batch of data is randomly selected from the experience pool to update the connection weights of the neural network; S7) Repeat steps S2-S6) until the algorithm converges and the optimal Q value function is found.

[0008] Preferably, the specific implementation process of step S6) includes: Calculate the weight update amount based on the Q value update formula and the gradient descent formula; Among them, the Q value update formula is: ; The gradient descent formula is: ; In the formula, is the learning rate; is the discount factor; is the weight update amount; W is the neural network weight matrix; is the partial derivative of the loss function L with respect to the weight W; Use the calculated weight update amount to update the connection weights of the neural network.

[0009] Preferably, the positive reward is that the temperature error is less than the maximum allowable error, then a positive reward (+1) is obtained; the small positive reward is that the temperature error is greater than the maximum allowable error but less than the temperature error of the previous time step, then a small positive reward (+0.1) is obtained; in other cases, a negative reward is obtained.

[0010] Preferably, the process of the central controller outputting a control signal includes: Deploy the trained deep Q network to the central controller. Obtain the parameters of the system's current state space in real time; The parameters of the state space are input into the trained deep Q network, and the deep Q network outputs the Q value of each action; The action with the largest Q value is selected as the control instruction to perform temperature compensation.

[0011] Preferably, the process of the central controller outputting the control signal further includes: The optimal compensation time is determined using an iterative learning control (ILC) algorithm; The central controller outputs control instructions to the execution layer at the optimal compensation time based on the control instructions and the learned optimal compensation time to achieve accurate temperature compensation.

[0012] Preferably, the specific process of determining the optimal compensation time using an iterative learning control (ILC) algorithm includes: In the production of injection molding machines, each pressurized injection molding process is recorded as batch k; Set the maximum time error eT allowed by the temperature control system to achieve the temperature change rate; In the kth batch, the temperature control system performs operations according to the temperature change rate of the current batch; and obtains the actual temperature value of the batch; After the kth batch is finished, the actual temperature curve is compared with the target temperature curve to obtain the temperature error curve of the entire batch; The temperature error curve is analyzed by ILC algorithm, and the time error et is calculated; Construct an ILC algorithm. The objective function of the ILC algorithm minimizes the square of the temperature error and the weighted square of the compensation change, thus achieving the optimal calculation of temperature control in time. Among them, the calculation formula of the temperature error term is: ; The calculation formula for the compensation change term is: ; The calculation formula of the objective function is: ; In the formula, k is the batch number; t is the time variable; is the target temperature curve; is the actual temperature curve of the kth batch; is the temperature error curve; is the temperature change rate of the kth batch; is the temperature change rate of the k+1th batch; To compensate for the change in quantity; is the weight factor, which is used to balance the temperature error term and the compensation amount change term; Compare the time error et of the current batch with the preset maximum allowable time error eT; If et>eT, iterative learning continues; if et≤eT, it is considered that the ILC algorithm has converged and the optimal compensation time has been found.

[0013] Beneficial effects of the present invention: 1. The present invention is based on the deep learning and experience playback of the DQN algorithm, which enables the algorithm to handle complex environments with continuous state spaces. In the precision intelligent temperature control system of the injection molding machine, the DQN algorithm is used to select the optimal control action (heating, cooling or no compensation) according to the current temperature state to achieve precise temperature control, reduce injection molding product defects caused by temperature fluctuations, and improve processing quality and quality consistency.

[0014] 2. The ILC algorithm can adaptively adjust the control signal according to the error information of the previous batch to adapt to the production time differences of different batches (such as changes in material properties, changes in ambient temperature, etc.), avoiding the control failure problem caused by fixed parameters of traditional PID control, reducing manual debugging time and costs, avoiding unnecessary energy consumption, and significantly improving production efficiency, thereby increasing the output of injection molding products. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 This is a structural framework diagram of a precise intelligent temperature control system for an injection molding machine according to the present invention.

[0017] Figure 2 This is a pre-training flow chart of a deep Q network algorithm in a precise intelligent temperature control system for an injection molding machine of the present invention.

[0018] Figure 3 This is a flow chart of using an ILC algorithm to determine the optimal compensation time in a precise intelligent temperature control system for an injection molding machine of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1-Figure 3 As shown, a precise intelligent temperature control system for an injection molding machine includes a collection layer, a control layer and an execution layer; The acquisition layer includes a temperature sensor and a signal converter; the temperature sensor is used to monitor the temperature of the injection molding machine cavity in real time; the signal converter is used to convert the temperature current signal into a temperature digital signal; The control layer includes a central controller that receives temperature digital signals, performs logic operations based on a pre-trained deep Q network algorithm, and outputs control signals; The execution layer includes a heating device and a cooling device. The heating device is used to heat the injection molding machine cavity according to the control instructions of the central controller; the cooling device is used to cool the injection molding machine cavity according to the control instructions of the central controller.

[0021] In the specific implementation process, the acquisition layer is responsible for real-time monitoring of the temperature of the injection molding machine cavity and converting the temperature signal into a digital signal for use by the control layer. This includes the selection and installation of temperature sensors, such as thermocouples (type K, type J, etc.) or thermistors (PT100, PT1000, etc.). The temperature sensor is installed at the key position of the injection molding machine cavity to ensure that the temperature change of the cavity can be accurately reflected. And the selection of a high-precision analog-to-digital converter (ADC), such as a 16-bit or 24-bit ADC module, ensures accurate conversion of the temperature signal. The signal converter converts the analog signal output by the temperature sensor into a digital signal. The converted digital signal is transmitted to the control layer through a communication interface (such as I2C, SPI or RS485).

[0022] The central controller of the control layer gives priority to high-performance embedded processors (such as ARM Cortex-A series or Intel Atom series) to ensure that complex deep Q network algorithms can be processed in real time. It also supports the training and reasoning of deep Q networks by configuring sufficient memory (such as 4GB RAM) and storage space (such as 32GB eMMC).

[0023] The execution layer starts the heating device or cooling device according to the control signal to adjust the temperature of the material cavity. The heating device selects efficient heating elements (such as electric heating tubes, electric heating wires, etc.) to ensure that the material cavity can be heated quickly and evenly. Pulse width modulation (PWM) or solid-state relays (SSR) are used to control the power of the heating element to achieve precise temperature regulation. The cooling device selects efficient cooling methods (such as water cooling or refrigerant cooling) to ensure that the temperature of the material cavity can be quickly reduced. Proportional valves or inverters are used to control the flow or speed of the cooling medium to achieve precise temperature regulation.

[0024] Specifically, the precision intelligent temperature control system for injection molding machines of the present invention can achieve precise control of the temperature of the material cavity through the collaborative work of the acquisition layer, the control layer and the execution layer. The acquisition layer is responsible for real-time temperature monitoring, the control layer generates control signals based on the deep Q network algorithm, and the execution layer adjusts the temperature according to the control signals. The implementation process of the system covers multiple links such as hardware selection, algorithm deployment, control signal generation and execution actions, ensuring the stability, real-time and accuracy of the system.

[0025] Furthermore, the construction process of the deep Q network algorithm includes: Build a deep Q network algorithm to define the state space, action space and reward function in the injection molding machine temperature control system; Furthermore, the state space includes the current temperature, the temperature error and the temperature change rate; the action space includes increasing the temperature, decreasing the temperature and no compensation; the reward function is the feedback returned by the system after executing the action, including positive reward, small positive reward and negative reward; Use a neural network to gradually approximate the optimal Q-value function, which represents the expected cumulative reward for performing an action in a specific state; the calculation formula of the Q-value function is: ; In the formula, is the state of the system at time t; is the action taken by the system at time t; It is the system that is performing the action Immediate rewards after is a discount factor used to balance immediate rewards and future rewards; Among them, the neural network includes an input layer, a hidden layer and an output layer. The input layer is used to input the current state space of the system, and the number of state spaces is the number of neurons in the input layer; The hidden layer is used to perform nonlinear transformation on the input data and extract data features; The output layer is used to output the Q value of each action.

[0026] Specifically, the Deep Q Network (DQN) is a reinforcement learning algorithm that combines Q learning and neural networks. It approximates the Q-value function through a neural network to learn the expected benefits of taking different actions in a specific state. This enables the algorithm to handle complex environments with continuous state spaces. In the precision intelligent temperature control system of the injection molding machine, the DQN algorithm is used to select the optimal control action (heating, cooling, or no compensation) according to the current temperature state to achieve precise temperature control.

[0027] In the specific implementation process, the state space of the temperature control system is the current environment description of the system, which is used to represent the temperature state of the injection molding machine cavity. The parameters are expressed as follows: Current temperature: the current batch temperature value of the cavity.

[0028] Temperature error: The difference between the current batch temperature and the target temperature.

[0029] Temperature change rate: The rate of change between the current batch temperature and the previous batch temperature.

[0030] The action space is the set of operations that the system can perform. In the temperature control system, the action space includes: heating, cooling, or no compensation. The reward function is used to evaluate the effect of the system after performing the action and guide the algorithm to learn the optimal strategy.

[0031] The deep Q network uses a neural network to approximate the Q value function. The network structure design includes: Input layer: receiving parameters of the state space (current temperature, temperature error, temperature change rate), the number of neurons in the input layer is 3. Hidden layer: contains several fully connected layers, using activation functions (such as ReLU) for nonlinear transformation. The number of hidden layers and the number of neurons are adjusted according to the complexity of the problem, for example: Layer 1: 64 neurons, ReLU activation. Layer 2: 32 neurons, ReLU activation. Output layer: outputs the Q value of each action, the number of neurons is 3 (corresponding to three actions).

[0032] Furthermore, the positive reward is that the temperature error is less than the maximum allowable error, then a positive reward (+1) is obtained; the small positive reward is that the temperature error is greater than the maximum allowable error but less than the temperature error of the previous time step, then a small positive reward (+0.1) is obtained; in other cases, a negative reward is obtained.

[0033] This scheme realizes intelligent optimization of temperature control system through deep Q learning. Its working principle is: comprehensively capture system state information through three-dimensional state space. Then, use deep neural network to approximate Q value function and evaluate the long-term benefits of different actions. According to the experience replay mechanism, break the data correlation and improve learning efficiency. Finally, in the update and iteration process, balance exploration and utilization based on ε-greedy strategy, and gradually converge to the optimal control strategy. This scheme has strong adaptability and generalization ability, and can effectively handle nonlinear characteristics and uncertainties in temperature control systems.

[0034] Furthermore, the pre-training process of the deep Q network algorithm includes: S1) Initialize the deep Q network and randomly select the connection weights of the neural network; S2) In each sampling period t, the state space of the acquisition system , including current temperature, target temperature and temperature change rate; S3) Based on the current Q value, use the ε-greedy strategy to select an action ; S4) The system performs the selected action , and return the reward and the next state ; S5) , , , ) is stored in the experience pool; S6) After a batch of injection molding process is completed, a batch of data is randomly selected from the experience pool to update the connection weights of the neural network; S7) Repeat steps S2-S6) until the algorithm converges and the optimal Q value function is found.

[0035] In the specific implementation process, the deep Q network structure is initialized. First, the three neurons in the input layer correspond to the current temperature, target temperature, and temperature change rate respectively; secondly, the hidden layer adopts a multi-layer structure to extract deep features; finally, the three neurons in the output layer correspond to heating, cooling, and uncompensated actions respectively. The initial weights are filled with random decimals to ensure that the network has a good starting point for training. And set the hyperparameters required for training, such as: discount factor γ (usually 0.9 to 0.99), exploration rate ϵ (the initial value is usually 1, gradually decaying), learning rate (such as 0.001), batch size (such as 32 or 64).

[0036] Next, calculate the reward and state transition: After the system performs the selected action, it needs to evaluate the effect of the action. According to the size of the temperature error, the system returns the corresponding reward value. At the same time, the new state after the action is performed is recorded to form a complete state transition sample.

[0037] And create an experience pool to store the experience data (state, action, reward, next state) collected during training. And set the experience playback buffer size (such as 10,000-50,000 experiences) to ensure the diversity and stability of the data.

[0038] As a more preferred implementation method, the ε-greedy strategy strikes a balance between exploration and exploitation in the random action selection mechanism in the experience pool. When the random number is greater than ε, the system selects the action with the largest Q value; when the random number is less than ε, the system randomly selects an action. The ε value gradually decreases during the training process, causing the system to gradually shift from exploration to exploitation.

[0039] Finally, in the weight update process, the network weights are updated based on the temporal difference error. The system first calculates the target Q value, and then calculates the error between the predicted Q value and the target Q value. Through the back-propagation algorithm, the system calculates the gradient of the weights and updates the network parameters.

[0040] Convergence judgment criteria: The convergence of the algorithm is comprehensively judged through multiple indicators: first, observe whether the average reward value tends to be stable; second, check whether the change of Q value is lower than the preset threshold; finally, verify whether the temperature control effect meets the accuracy requirements.

[0041] Specifically, the pre-training process continuously improves the control performance of the system through iterative optimization, and finally converges to a stable control strategy. The whole process makes full use of the characteristics of deep learning and realizes the intelligent optimization of the temperature control system.

[0042] Furthermore, the specific implementation process of step S6) includes: Calculate the weight update amount based on the Q value update formula and the gradient descent formula; Among them, the Q value update formula is: ; The gradient descent formula is: ; In the formula, is the learning rate; is the discount factor; is the weight update amount; W is the neural network weight matrix; is the partial derivative of the loss function L with respect to the weight W; Use the calculated weight update amount to update the connection weights of the neural network.

[0043] Specifically, the process of updating the Q value is: sampling an experience from the experience pool; using the target network to calculate the next state s t+1 The Q value of all actions is updated. The maximum Q value is selected to update the target Q value.

[0044] Then, the update process of the connection weights of the neural network is: use the loss function to measure the difference between the current Q value and the target Q value; use the gradient descent method to calculate the weight update amount.

[0045] In the update phase of the pre-training process, the target Q value is calculated through the Q value update formula, the difference between the predicted Q value and the target Q value is measured through the loss function, and the weight of the neural network is updated through the gradient descent method, so that the Q value function is close to the optimal value. This process is achieved through steps such as forward propagation, back propagation, and weight update, ensuring that the deep Q network can learn the optimal control strategy.

[0046] Furthermore, the process of the central controller outputting the control signal includes: Deploy the trained deep Q network to the central controller. Obtain the parameters of the system's current state space in real time; The parameters of the state space are input into the trained deep Q network, and the deep Q network outputs the Q value of each action; The action with the largest Q value is selected as the control instruction to perform temperature compensation.

[0047] In the specific implementation process, real-time temperature data is obtained from the acquisition layer through a communication interface (such as RS485 or Ethernet). Then, the current state space parameters are calculated, including the current temperature, temperature error, and temperature change rate. The state space parameters are converted into tensor format and input into the deep Q network. The deep learning framework is used for reasoning and the Q value of each action is output. The Q values ​​of the three actions are compared and the action with the largest Q value is selected. Finally, a control signal is generated according to the selected action, for example: increase the temperature: generate a heating control signal to start the heating device.

[0048] Lower the temperature: Generate a cooling control signal to start the cooling device.

[0049] No compensation: No control signal is generated and the current state is maintained.

[0050] By acquiring temperature data in real time, running the deep Q network algorithm, generating control signals and outputting them to the execution layer, the central controller achieves precise control of the temperature of the injection molding machine cavity. Optimization and debugging measures further improve the real-time performance, stability and safety of the system, ensuring the efficient operation of the temperature control system.

[0051] Furthermore, the process of the central controller outputting the control signal also includes: The optimal compensation time is determined using an iterative learning control (ILC) algorithm; The central controller outputs control instructions to the execution layer at the optimal compensation time based on the control instructions and the learned optimal compensation time to achieve accurate temperature compensation.

[0052] Specifically, iterative learning control (ILC) is an advanced control method suitable for repetitive tasks, especially for periodic processes such as injection molding machine temperature control systems. ILC continuously learns the control error of the previous batch and optimizes the control strategy of the current batch, thereby gradually approaching the optimal compensation time.

[0053] In the actual injection molding process, the production time of different batches may vary, for example: Variations in material properties: Different batches of raw materials may result in different temperature response times.

[0054] Ambient temperature changes: Changes in ambient temperature may affect the heating and cooling rates of the cavity.

[0055] Changes in equipment status: The aging or maintenance status of the equipment may affect the temperature control performance.

[0056] ILC can adaptively adjust the control signal based on the error information of the previous batch to adapt to the time difference between different batches. After each batch, the control error ek(t) and control signal uk(t) are recorded. Using the error information of the previous batch, the control signal of the current batch is updated. Through iterative learning, ILC can reduce the temperature control difference between batches and improve the consistency of product quality.

[0057] Furthermore, the specific process of determining the optimal compensation time using the iterative learning control (ILC) algorithm includes: In the production of injection molding machines, each pressurized injection molding process is recorded as batch k; Set the maximum time error eT allowed by the temperature control system to achieve the temperature change rate; In the kth batch, the temperature control system performs operations according to the temperature change rate of the current batch; and obtains the actual temperature value of the batch; After the kth batch is finished, the actual temperature curve is compared with the target temperature curve to obtain the temperature error curve of the entire batch; The temperature error curve is analyzed by ILC algorithm, and the time error et is calculated; Construct an ILC algorithm. The objective function of the ILC algorithm minimizes the square of the temperature error and the weighted square of the compensation change, thus achieving the optimal calculation of temperature control in time. Among them, the calculation formula of the temperature error term is: ; The calculation formula for the compensation change term is: ; The calculation formula of the objective function is: ; In the formula, k is the batch number; t is the time variable; is the target temperature curve; is the actual temperature curve of the kth batch; is the temperature error curve; is the temperature change rate of the kth batch; is the temperature change rate of the k+1th batch; To compensate for the change in quantity; is the weight factor, which is used to balance the temperature error term and the compensation amount change term; Compare the time error et of the current batch with the preset maximum allowable time error eT; If et>eT, iterative learning continues; if et≤eT, it is considered that the ILC algorithm has converged and the optimal compensation time has been found.

[0058] Using the iterative learning control (ILC) algorithm to determine the optimal compensation time can effectively deal with the time differences between different batches in the injection molding process. By continuously learning the error information of the previous batch, ILC gradually optimizes the control signal to improve the accuracy and consistency of temperature control. Ultimately, the precise control of the injection molding machine cavity temperature is achieved. The application of ILC not only reduces manual debugging time, but also improves product quality and production efficiency.

[0059] In summary, the present invention proposes a precise intelligent temperature control system for an injection molding machine, which makes the system easy to expand and maintain through the modular design of the acquisition layer, control layer and execution layer. Based on the deep learning and experience playback of the DQN algorithm, it can quickly converge to the optimal control strategy; intelligently select heating or cooling actions to achieve precise temperature control. In addition, in order to cope with the time differences between different production batches in the actual injection molding process, the ILC algorithm can adaptively adjust the control signal according to the error information of the previous batch to adapt to the production time differences between different batches (such as changes in material properties, changes in ambient temperature, etc.). It reduces the time and cost of manual debugging, avoids unnecessary energy consumption, and significantly improves production efficiency. It has promoted the intelligent and green development of the injection molding industry.

[0060] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A precise intelligent temperature control system for an injection molding machine, characterized in that: It includes an acquisition layer, a control layer and an execution layer; the acquisition layer includes a temperature sensor and a signal converter; the temperature sensor is used to monitor the temperature of the injection molding machine cavity in real time; the signal converter is used to convert the temperature current signal into a temperature digital signal; The control layer includes a central controller that receives temperature digital signals, performs logical operations based on a pre-trained deep Q network algorithm, and outputs control signals; The construction process of the deep Q network algorithm includes: Build a deep Q network algorithm to define the state space, action space and reward function in the temperature control system of the injection molding machine; Use a neural network to gradually approximate the optimal Q-value function; the neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to input the current state space of the system, and the number of state spaces is used as the number of neurons in the input layer. The hidden layer is used to perform nonlinear transformation on the input data and extract data features; The output layer is used to output the Q value of each action; The execution layer includes a heating device and a cooling device. The heating device is used to heat the injection molding machine cavity according to the control instructions of the central controller; the cooling device is used to cool the injection molding machine cavity according to the control instructions of the central controller.

2. According to claim 1, the precise intelligent temperature control system for injection molding machine is characterized in that: The Q-value function represents the expected cumulative reward for performing an action in a specific state; the calculation formula of the Q-value function is: ； In the formula, is the state of the system at time t; is the action taken by the system at time t; It is the system that is performing the action Immediate rewards after is a discount factor that balances immediate rewards with future rewards.

3. The precise intelligent temperature control system for an injection molding machine according to claim 1, characterized in that: The state space includes the current temperature, temperature error and temperature change rate; the action space includes increasing the temperature, lowering the temperature and no compensation; the reward function is the feedback returned by the system after executing the action, including positive reward, small positive reward and negative reward.

4. The precise intelligent temperature control system for an injection molding machine according to claim 3, characterized in that: The positive reward is obtained when the temperature error is less than the maximum allowable error; the small positive reward is obtained when the temperature error is greater than the maximum allowable error but less than the temperature error of the previous time step; a negative reward is obtained in other cases.

5. The precise intelligent temperature control system for an injection molding machine according to claim 1, characterized in that: The pre-training process of the deep Q network algorithm includes: S1) Initialize the deep Q network and randomly select the connection weights of the neural network; S2) In each sampling period t, the state space of the acquisition system , including current temperature, target temperature and temperature change rate; S3) Based on the current Q value, use the ε-greedy strategy to select an action ; S4) The system executes the selected action , and return the reward and the next state ; S5) , , , ) is stored in the experience pool; S6) After a batch of injection molding process is completed, a batch of data is randomly selected from the experience pool to update the connection weights of the neural network; S7) Repeat steps S2-S6) until the algorithm converges and the optimal Q value function is found.

6. The precise intelligent temperature control system for an injection molding machine according to claim 5, characterized in that: The specific implementation process of step S6) includes: Calculate the weight update amount based on the Q value update formula and the gradient descent formula; Among them, the Q value update formula is: ； The gradient descent formula is: ; In the formula, is the learning rate; is the discount factor; is the weight update amount; W is the neural network weight matrix; is the partial derivative of the loss function L with respect to the weight W; Use the calculated weight update amount to update the connection weights of the neural network.

7. The precise intelligent temperature control system for an injection molding machine according to claim 1, characterized in that: The process of the central controller outputting a control signal includes: Deploy the trained deep Q network to the central controller. Obtain the parameters of the system's current state space in real time; The parameters of the state space are input into the trained deep Q network, and the deep Q network outputs the Q value of each action; The action with the largest Q value is selected as the control instruction to perform temperature compensation.

8. The precise intelligent temperature control system for an injection molding machine according to claim 7, characterized in that: The process of the central controller outputting the control signal also includes: Use iterative learning control algorithm to determine the optimal compensation time; The central controller outputs control instructions to the execution layer at the optimal compensation time based on the control instructions and the learned optimal compensation time to achieve accurate temperature compensation.

9. The precise intelligent temperature control system for an injection molding machine according to claim 8, characterized in that: The specific process of using the iterative learning control algorithm to determine the optimal compensation time includes: In the production of injection molding machines, each pressurized injection molding process is recorded as batch k; Set the maximum time error eT allowed by the temperature control system to achieve the temperature change rate; In the kth batch, the temperature control system performs operations according to the temperature change rate of the current batch; and obtains the actual temperature value of the batch; After the kth batch is finished, the actual temperature curve is compared with the target temperature curve to obtain the temperature error curve of the entire batch; The temperature error curve is analyzed by iterative learning control algorithm, and the time error et is calculated; Compare the time error et of the current batch with the preset maximum allowable time error eT; If et>eT, the iterative learning continues; if et≤eT, it is considered that the iterative learning control algorithm converges and finds the optimal compensation time.

10. The precise intelligent temperature control system for an injection molding machine according to claim 9, characterized in that: The specific process of analyzing the temperature error curve by the iterative learning control algorithm includes: constructing an iterative learning control algorithm, wherein the objective function of the iterative learning control algorithm realizes the time optimization calculation of the temperature control by minimizing the square of the temperature error and the weighted square sum of the compensation amount change; Among them, the calculation formula of the temperature error term is: ; The calculation formula for the compensation change term is: ; The objective function is calculated as follows: ; In the formula, k is the batch number; t is the time variable; is the target temperature curve; is the actual temperature curve of the kth batch; is the temperature error curve; is the temperature change rate of the kth batch; is the temperature change rate of the k+1th batch; To compensate for the change in quantity; is the weight factor used to balance the temperature error term and the compensation change term.

Images