Temperature field real-time monitoring method and system in automobile mold injection molding process

Abstract

The invention relates to a real-time monitoring method and system for a temperature field in an automobile mold injection molding process, in particular to the field of temperature field control of automobile mold injection molding, effectively eliminates a monitoring blind area through real-time reconstruction of a complete temperature field, accurately identifies an overheating risk based on prediction data corrected by physical constraints, and improves the accuracy of temperature field monitoring. And finally, active intervention on a cooling system is realized through feedforward-feedback compound control, so that the product quality consistency is remarkably improved, the rejection rate is reduced, meanwhile, the production takt time and energy consumption are optimized, and the intelligent control level and economic benefits of the injection molding process are comprehensively enhanced.

Description

Technical Field

[0001] This invention relates to the field of temperature field control in automotive mold injection molding, and more specifically, to a method and system for real-time monitoring of the temperature field during automotive mold injection molding. Background Technology

[0002] In the automotive manufacturing industry, injection molding is a core process for producing various interior and exterior trim parts, such as instrument panels, door panels, and center consoles. With the automotive industry's increasing pursuit of lightweight, high strength, and high-quality appearance, the design of automotive parts is trending towards larger size, thinner walls, and more complex structures. This results in exceptionally complex cavity structures for injection molds, often containing numerous ribs, deep cavities, and irregular curved surfaces. In actual high-speed, high-cycle continuous production, the high-temperature molten plastic is injected into the mold cavity under high pressure. The shear effect at its flow front generates a large amount of heat, and the layout of the cooling channels inside the mold is further constrained by factors such as temperature and humidity. Due to the physical limitations of mechanical strength and processing technology, it is difficult to achieve perfect conformal cooling to the cavity surface. Therefore, in certain areas of the mold, especially at the end of the melt flow, the root of the rib, and the abrupt change in wall thickness, local overheating is likely to occur due to heat accumulation, or insufficient cooling areas may form because the cooling medium cannot reach the mold effectively. In addition, dynamic factors such as changes in temperature and humidity of the production environment, slight differences in the thermophysical properties of different batches of plastic raw materials, and the gradual decline in the performance of the injection molding machine will further exacerbate the non-uniformity and instability of the temperature field distribution inside the mold, posing a severe challenge to the precise control of the molding process.

[0003] Currently, existing technologies for temperature monitoring and control of injection molds mainly rely on a limited number of contact temperature sensors, such as thermocouples or resistance temperature detectors (RTDs), which are pre-embedded at key points within the mold for measurement. This point-based measurement method, due to the limited number and fixed positions of the sensors, struggles to comprehensively and completely capture all potential temperature anomalies on the complex cavity surface, resulting in significant monitoring blind spots. In terms of control strategies, PID controllers based on classical control theory are commonly used. Their operating mode is essentially static setting and hysteresis feedback; that is, the system only adjusts the valves or flow rate of the cooling system after the sensor actually detects a temperature deviation from the set value. This response mechanism inevitably introduces a time delay, which is problematic for the instantaneous high temperatures and short duration of injection molding. Furthermore, for processes with drastic dynamic changes, lagging control cannot effectively suppress temperature field distortions caused by dynamic factors in each production cycle, nor can it achieve advance prediction and compensation for temperature field changes. Although computer-aided engineering technology can predict temperature distribution through thermal flow simulation analysis to optimize water channel design during the mold design stage, such simulation models are offline and static, and cannot interact and correct with real-time production data. Therefore, the existing technology system has a core defect: it lacks the ability to reconstruct or predict the complete temperature field distribution of the entire mold cavity surface in real time during the production process, especially to accurately infer the temperature in the monitoring blind zone, and based on this prediction result, to perform forward-looking, adaptive, and precise control of the cooling system to fundamentally ensure the uniformity and stability of the temperature field. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a method and system for real-time monitoring of the temperature field during the injection molding process of automotive molds, thereby resolving the issues mentioned in the background section.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: a method for real-time monitoring of the temperature field during the injection molding process of automotive molds, specifically including the following steps: Step S1: Based on the preset mold geometry model and material thermal property parameters, an efficient reduced-order heat transfer model is constructed offline. During the online operation of the injection molding process, the measurement values ​​of multiple temperature sensors installed on the mold are acquired in real time, and the measurement values ​​are used as boundary conditions to input into the reduced-order heat transfer model to drive the model to perform calculations, thereby outputting a complete temperature field distribution data that includes the area on the mold cavity surface where no sensors are installed in real time. Step S2: In the offline preparation stage, process parameter data collected in multiple historical injection molding cycles and corresponding complete temperature field distribution data are used to form a training sample set. A long short-term memory network prediction model is trained using this training sample set. In online operation, the complete temperature field distribution data of the current cycle output in real time from step S1 and the process parameter data related to the current process state are input into the trained long short-term memory network prediction model to predict the predicted value of the temperature field change trend data at future moments. Step S3: The complete temperature field distribution data of the current period output in real time from step S1 is superimposed with the predicted value of the temperature field change trend data of future time predicted by step S2 to obtain a preliminary future temperature field estimate. Subsequently, a physical constraint correction process based on a simplified thermodynamic conservation equation is applied to verify and correct the physical feasibility of the preliminary future temperature field estimate, generating a final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation, and identifying potential overheating risk areas and their risk levels accordingly. Step S4: Using the final future temperature field distribution data and potential overheating risk areas generated in step S3 as feedforward signals, and combining them with a multi-objective optimization function aimed at minimizing temperature tracking error and smoothing control actions, the optimal control parameters for each independent cooling loop are obtained; the optimal control parameters are issued to the actuator in advance to realize adaptive feedforward control of the cooling system, and at the same time, it forms a composite control strategy with traditional feedback control. In a preferred embodiment, the specific process of offline construction of a computationally efficient reduced-order heat transfer model in step S1 is as follows: First, based on the preset mold geometry model and material thermal properties parameters, including the density, specific heat capacity and thermal conductivity of the mold material, a high-fidelity three-dimensional physical model is established to describe the transient heat conduction process inside the mold using partial differential equations. Subsequently, an intrinsic orthogonal decomposition method is applied to the high-fidelity three-dimensional physical model to reduce its order. The order reduction process is as follows: by sampling and solving in the typical working condition parameter space of the high-fidelity three-dimensional physical model, a series of temperature field snapshots describing its dynamic characteristics are obtained, and these temperature field snapshots are combined to construct a snapshot matrix; singular value decomposition is performed on the snapshot matrix to extract the first few eigenmodes with the largest energy to span a low-dimensional subspace. Finally, the partial differential equations of the high-fidelity three-dimensional physical model are projected onto this low-dimensional subspace to obtain a set of low-dimensional ordinary differential equations with a few modal coefficients as state vectors, namely the reduced-order heat transfer model.

[0006] In a preferred embodiment, the specific operation of inputting the measured values ​​as boundary conditions into the reduced-order heat transfer model and driving the calculation is as follows: During the online operation of the injection molding process, the measured values ​​of multiple temperature sensors acquired in real time are arranged in a preset order to form a sensor measurement value vector; The sensor measurement vector is used as the input signal and injected into the port defined by the input matrix corresponding to the reduced-order heat transfer model; The modal coefficient vector at the current moment is calculated by solving a system of low-dimensional ordinary differential equations with modal coefficients as state vectors using numerical integration. Finally, by summing the average temperature field calculated offline in advance with all the product results obtained by multiplying each modal coefficient in the modal coefficient vector at the current moment by the corresponding characteristic mode, a complete temperature field distribution data including the area on the surface of the mold cavity where no sensors are deployed is reconstructed.

[0007] In a preferred embodiment, the specific operation of training a long short-term memory network prediction model in step S2 during the offline preparation phase is as follows: A1. Extract data from the historical database for multiple consecutive injection cycles. The data for each cycle includes complete temperature field distribution data obtained from step S1 at different historical time points, as well as process parameter data collected at the same time point as the complete temperature field distribution data to describe the state of the injection process. The process parameter data includes melt temperature, injection speed, holding pressure and cooling water valve opening. A2. Combine the complete temperature field distribution data obtained at the same time point with the process parameter data to form a fused feature vector; A3. A series of continuous fused feature vectors are extracted as model inputs using a preset time window length. The time window length is defined as the number of consecutive historical moments. The change in the complete temperature field distribution data at the end of the time window relative to the beginning of the time window is used as the target output for model training. This constitutes a training sample set. A deep long short-term memory network is trained using the training sample set. The long short-term memory network is configured with an input gate, a forget gate, and an output gate. A3.1 The forgetting gate calculates a forgetting factor based on the fusion feature vector at the current moment and the hidden state at the previous moment, which is used to control the degree of retention of historical information; A3.2 The input gate calculates the input factor based on the fused feature vector at the current time and the hidden state at the previous time, which is used to control the degree of addition of new information; A3.3 The Long Short-Term Memory Network also calculates candidate cell states, which are updated by adding the result of multiplying the cell state of the previous time step by the forgetting factor to the product of the input factor and the candidate cell state. A3.4 The Long Short-Term Memory Network outputs a predicted value of the temperature field change at future times through a fully connected layer; A4. The loss function used during training is the weighted mean square error between the predicted value and the true value; A5. Obtain a trained Long Short-Term Memory (LSTM) network prediction model.

[0008] In a preferred embodiment, the specific operation of predicting the predicted value of the temperature field change trend data at future times during online operation is as follows: In each control cycle, the complete temperature field distribution data of the current cycle is received in real time from step S1, and process parameter data related to the current process state are obtained in the current and past periods, wherein the length of the past period is defined by the preset time window length. The complete temperature field distribution data and process parameter data at the current moment are concatenated to form the fusion feature vector at the current moment, and then combined with the fusion feature vectors of multiple consecutive past moments in chronological order to form an input sequence with the same time window length as preset in the offline training phase. The input sequence is fed into the trained Long Short-Term Memory Network (LSTM) prediction model. The LSM prediction model processes the input sequence through its internal forget gate, input gate, and output gate mechanisms, updates its internal cell states and hidden states, and finally outputs a predicted value of the temperature field change trend data after the future prediction step. This predicted value represents the amount of change in the complete temperature field distribution data at the time after the future prediction step relative to the current control cycle.

[0009] In a preferred embodiment, the specific process of obtaining preliminary future temperature field estimation data in step S3 is as follows: In each control cycle, the complete temperature field distribution data of the current control cycle is received from the real-time output of step S1, and the predicted value of the temperature field change trend data after the future prediction step size is received from step S2. Subsequently, using a preset linear superposition model, the complete temperature field distribution data of the current control cycle is algebraically added to the predicted values ​​of the temperature field change trend data at the time after the future prediction step, and the preliminary future temperature field estimation data is directly calculated. The specific steps for verifying and correcting the physical feasibility of the preliminary future temperature field estimate data by applying a physical constraint correction process based on a simplified thermodynamic conservation equation are as follows: First, a simplified thermodynamic conservation equation for rapid verification is constructed. This equation is based on the principle of energy balance and associates the rate of change of the temperature field per unit time with the heat conduction term and the internal heat source term. The rate of change of the temperature field per unit time is obtained by dividing the difference between the preliminary estimated future temperature field data and the complete temperature field distribution data of the current period by the prediction time step. The heat conduction term is calculated by the thermal conductivity of the mold material and the spatial second derivative of the preliminary estimated future temperature field data. The internal heat source term represents the average heat flux density transferred by the plastic melt per unit volume. Then, the preliminary future temperature field estimate data is substituted into the simplified thermodynamic conservation equation, and the difference between the energy change rate represented by the left side of the equation and the heat flow condition represented by the right side is calculated. This difference represents the degree of energy imbalance, and the difference is compared with a preset physical rationality threshold. When the difference exceeds the physical reasonableness threshold, a correction optimization process is initiated. This process solves for the corrected temperature field by minimizing a physical residual function. The physical residual function includes a first term and a second term. The first term represents the squared norm of the difference between the corrected temperature field and the current periodic complete temperature field distribution data relative to the prediction time step and the difference between the combination of the heat conduction term and the internal heat source term, which is used to measure the degree of satisfaction of the physical conservation law. The second term represents the squared norm of the difference between the corrected temperature field and the preliminary future temperature field estimate data multiplied by a regularization parameter, which is used to measure the fidelity to the original prediction value. The corrected temperature field that minimizes the physical residual function is solved by an iterative optimization algorithm, thereby generating the final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation.

[0010] In a preferred embodiment, the specific process of identifying potential overheating risk areas and their risk levels is as follows: The final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation is compared point by point with a preset temperature threshold matrix related to material processing to identify all spatial locations where the temperature value exceeds the corresponding temperature threshold. Next, using the connected component analysis algorithm in image processing, adjacent overheating locations in space are merged and connected, and marked as several consecutive potential overheating risk areas; Then, a comprehensive risk index is calculated for each identified potential overheating risk area. The calculation process of the comprehensive risk index is as follows: obtain the highest temperature value in the risk area, calculate the difference between the highest temperature value and the corresponding threshold in the temperature threshold matrix, and divide the difference by a preset reference temperature rise value for normalization processing to obtain the normalized overheating amplitude. At the same time, the size of the risk area is obtained, and the proximity index between the risk area and the known vulnerable critical areas in the mold is assessed; Finally, using preset first weighting coefficients, second weighting coefficients, and third weighting coefficients, the normalized overtemperature amplitude, the area size of the risk region, and the proximity are weighted and summed to obtain the comprehensive risk index of the risk region. Finally, based on the magnitude of the comprehensive risk index of each risk area, it is divided into different risk level ranges. When the comprehensive risk index is less than the first preset threshold, the corresponding risk area is defined as low risk level; when the comprehensive risk index is greater than or equal to the first preset threshold and less than the second preset threshold, the corresponding risk area is defined as medium risk level; and when the comprehensive risk index is greater than or equal to the second preset threshold, the corresponding risk area is defined as high risk level.

[0011] In a preferred embodiment, in step S4, by combining a multi-objective optimization function aimed at minimizing temperature tracking error and smoothing control actions, the optimal control parameters for each independent cooling loop are obtained. Specifically, the operation is as follows: First, the final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation generated in step S3 is received as the prediction target of the feedforward control. At the same time, the potential overheating risk areas identified in step S3 and their corresponding risk level information are received, and the control parameter vectors of each cooling loop at the current moment and the target temperature field data pre-set by the process requirements are obtained. Next, a multi-objective optimization function to be minimized is constructed. This function consists of two terms. The first term is the temperature tracking error term, which is calculated by integrating the spatial weight function over the entire three-dimensional spatial domain of the mold with the square of the difference between the predicted target temperature field data and the target temperature field data. The value of the spatial weight function is dynamically set according to the spatial point position, specifically: If the spatial point is located within a potential overheating risk area, its weight value is equal to a preset base weight plus a risk weight coefficient multiplied by the risk level of the risk area; If the spatial point does not belong to any risk area, its weight value is equal to the basic weight; The second term is the control action smoothing term, which is obtained by calculating the square of the Euclidean distance between the control parameter vector to be solved and the control parameter vector at the previous moment, and multiplying it by a control action smoothing weight coefficient. Then, a local linear influence model of the control parameter vector on the final future temperature field distribution data is established. This model indicates that the final future temperature field distribution data is approximately equal to the product of a control gain matrix and a control parameter vector plus a bias vector. The control gain matrix and the bias vector are determined by the thermal response characteristics of the mold through offline simulation. By substituting the local linear influence model into the multi-objective optimization function, the original nonlinear optimization problem is transformed into a quadratic programming problem with boundary constraints. Finally, the sequential quadratic programming algorithm is used, and the interior point method is employed to solve the quadratic programming problem to obtain the optimal control parameter vector.

[0012] In a preferred embodiment, the specific operation of issuing the optimal control parameters to the actuator in advance to achieve adaptive feedforward control of the cooling system, while forming a composite control strategy with traditional feedback control, is as follows: The optimal control parameter vector obtained by the solution is used as the feedforward control quantity and issued to the actuators of each independent cooling circuit in advance within the current control cycle; Simultaneously, a conventional feedback controller is run in parallel. This feedback controller takes the deviation between the complete temperature field distribution data of the current period output in real time by step S1 and the preset target temperature field data as input to calculate the feedback control correction amount. Finally, the feedforward control quantity and the feedback control correction quantity are linearly superimposed to form the final composite control signal.

[0013] This application also provides a real-time temperature field monitoring system for automotive mold injection molding process, specifically including: The real-time temperature field reconstruction module is used to trigger the online operation of the injection molding process by a preset mold geometry model and thermophysical parameters. It uses an offline-constructed, computationally efficient reduced-order heat transfer model to obtain the measured values ​​of multiple temperature sensors placed on the mold in real time as boundary conditions. This drives the reduced-order heat transfer model to perform calculations and outputs a complete temperature field distribution data that includes the area on the mold cavity surface where no sensors are placed. The temperature field change trend prediction module is used to trigger the training sample set composed of process parameter data and corresponding temperature field data collected in multiple historical injection molding cycles during the offline preparation stage. It trains a long short-term memory network prediction model and, during online operation, is triggered by the complete temperature field distribution data of the current cycle output in real time in step S1 and the process parameter data related to the current process state. The data is then input into the trained long short-term memory network prediction model to predict the temperature field change trend data at future moments. The hotspot risk identification module is triggered by the complete temperature field distribution data of the current period output in real time in step S1 and the temperature field change trend data of future time predicted in step S2. By superimposing the two, a preliminary future temperature field estimate is obtained. Then, a physical constraint correction process based on a simplified thermodynamic conservation equation is applied to verify and correct the physical feasibility of the preliminary future temperature field estimate, generating a final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation, and identifying potential overheating risk areas and their risk levels accordingly. The adaptive feedforward composite control module is used as a feedforward signal to trigger the final future temperature field distribution data and potential overheating risk areas generated in step S3. It combines a multi-objective optimization function with the goals of minimizing temperature tracking error and smoothing control actions to solve for the optimal control parameters for each independent cooling loop. The optimal control parameters are then issued to the actuator in advance to realize adaptive feedforward control of the cooling system. At the same time, it forms a composite control strategy with traditional feedback control.

[0014] The beneficial effects of this invention are: the method effectively eliminates the monitoring blind spot by reconstructing the complete temperature field in real time, accurately identifies overheating risks based on the predicted data corrected by physical constraints, and finally realizes active intervention in the cooling system through feedforward-feedback composite control, thereby significantly improving product quality consistency, reducing scrap rate, optimizing production cycle and energy consumption, and comprehensively enhancing the intelligent control level and economic benefits of the injection molding process. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0018] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0019] Example 1 This embodiment provides, for example Figure 1 The method for real-time monitoring of temperature field during automotive mold injection molding, as shown, specifically includes the following steps: Step S1: Based on the preset mold geometry model and material thermophysical parameters, an efficient reduced-order heat transfer model is constructed offline. During the online operation of the injection molding process, the measurement values ​​of multiple temperature sensors deployed on the mold are acquired in real time, and the measurement values ​​are used as boundary conditions to input into the reduced-order heat transfer model to drive the model to perform calculations. This results in the real-time output of complete temperature field distribution data, including the area on the mold cavity surface where no sensors are deployed. Step S1 creatively transforms the three-dimensional thermal model, which cannot be calculated in real time, into a low-dimensional dynamic system that can be solved in real time by introducing a model reduction technique based on intrinsic orthogonal decomposition. It uses limited sensor data as a driver and solves the low-dimensional state vector to finally reconstruct a complete temperature field without blind spots. The "complete temperature field distribution data" output in this step is the only authoritative temperature state basis upon which all subsequent prediction and control behaviors of this application depend, ensuring the consistency and accuracy of the data source for the entire method. Step S2: In the offline preparation stage, process parameter data collected in multiple historical injection molding cycles and corresponding complete temperature field distribution data are used to form a training sample set. A long short-term memory network prediction model is trained using this training sample set. In online operation, the complete temperature field distribution data of the current cycle output in real time from step S1 and the process parameter data related to the current process state are input into the trained long short-term memory network prediction model to predict the predicted value of the temperature field change trend data at future moments. Step S2 introduces an LSTM network trained on historical data with long-term memory capabilities to apply sequence learning technology to the dynamic prediction of the injection molding temperature field. This step successfully digitizes and models the complex relationship between process disturbances and temperature field evolution that is difficult to describe precisely with physical equations, providing crucial forward-looking information for the adaptive control of the entire method. Step S3: The complete temperature field distribution data for the current period, output in real time from Step S1, is superimposed with the predicted values ​​of the temperature field change trend data for future moments, obtained from Step S2, to obtain a preliminary estimate of the future temperature field. Subsequently, a physical constraint correction process based on a simplified thermodynamic conservation equation is applied to verify and correct the physical feasibility of this preliminary future temperature field estimate, generating a final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation. Based on this, potential overheating risk areas and their risk levels are identified. Step S3 introduces a "preliminary estimation - physical correction - risk identification" process. The sequential processing logic of "separate" constructs a post-processing pipeline for prediction that integrates data and physics. Its effects are as follows: by using simplified physical conservation laws as constraints, the pure data prediction results are corrected through optimization algorithms, ensuring the physical credibility of the prediction results; furthermore, through a multi-factor risk quantification model, accurate and graded positioning of overheating risks is achieved. The output of step S3 not only enhances the foresight capability of the entire method, but also upgrades quality control from "abnormal alarm" to "risk quantification", providing scientific and reliable input for intelligent decision-making and control in step S4, reflecting the logical progression from "perception and prediction" to "assessment and decision-making". Step S4: Using the final future temperature field distribution data generated in Step S3 and the potential overheating risk area as feedforward signals, and combining a multi-objective optimization function aimed at minimizing temperature tracking error and smoothing control actions, the optimal control parameters for each independent cooling loop are obtained. The optimal control parameters are then issued to the actuator in advance to achieve adaptive feedforward control of the cooling system. At the same time, it forms a composite control strategy with traditional feedback control. Step S4 constructs a multi-objective optimization function that integrates risk-weighted temperature tracking and control action smoothing, transforming the risk quantification information provided in Step S3 into a spatial weighting factor, thereby realizing differentiated feedforward control based on risk level. By transforming the nonlinear optimization problem into a real-time solvable quadratic programming problem and combining it with the feedforward-feedback composite architecture, Step S4 not only improves the accuracy and foresight of the control but also ensures the stability of the system. This step perfectly realizes the closed loop from "data perception" to "intelligent decision-making" to "precise execution".

[0020] In this embodiment, the specific process of building a computationally efficient reduced-order heat transfer model offline in step S1 is as follows: First, based on the preset mold geometry model and material thermophysical parameters, including the mold material's density, specific heat capacity, and thermal conductivity, a high-fidelity three-dimensional physical model is established to describe the transient heat conduction process inside the mold using partial differential equations. The partial differential equation formula is as follows: ; in, Density indicates the density of the mold material; it measures the mass of a unit volume of mold, and the unit is usually kilograms per cubic meter (kg / m³). 3 The higher the density, the greater the material's ability to store heat, and the slower the temperature changes. This refers to the specific heat capacity at constant pressure of the mold material. It measures the amount of heat required to raise the temperature per unit mass of the material by one unit. The unit is joules per kilogram per Kelvin (J / (kg·K)). The higher the specific heat capacity, the more difficult it is for the material to heat up or cool down. "Indicates constant pressure" The temperature field is represented by TY(x,y,z,t), which is a function of space (three-dimensional coordinates x,y,z) and time (t). It represents the temperature value at every point inside the mold at a given moment, in Kelvin (K) or degrees Celsius (°C), where t represents time in seconds (s). The equation describes the temperature field. How does it evolve over time (t)? , Represents the temperature field The partial derivative with respect to time t intuitively represents the rate of temperature change. The thermal conductivity of a mold material measures its ability to conduct heat, and is measured in watts per meter per Kelvin (W / (m·K)). The higher the value, the faster the heat is conducted in the material. Let a vector differential operator be defined in the formula: This represents the temperature gradient, with the direction pointing towards the direction of the fastest temperature increase. According to Fourier's law of heat conduction, heat flux density is represented by a vector, which is the amount of heat passing through a unit area per unit time. The divergence of the heat flux density vector (i.e., the heat conduction term) is physically represented as the difference between the heat flowing into a unit volume and the heat flowing out of a unit volume per unit time; that is, the net rate of heat accumulation within that volume. This refers to the intensity of the internal heat source (i.e., the internal heat source term), which is the heat generated per unit time and per unit volume, measured in watts per cubic meter (W / m³). 3 In the context of injection molding, this mainly comes from the heat transferred to the mold by the injected high-temperature molten plastic. Subsequently, an intrinsic orthogonal decomposition method is applied to the high-fidelity three-dimensional physical model to reduce its order. The order reduction process is as follows: by sampling and solving in the typical working condition parameter space of the high-fidelity three-dimensional physical model, a series of temperature field snapshots describing its dynamic characteristics are obtained, and these temperature field snapshots are combined to construct a snapshot matrix; singular value decomposition is performed on the snapshot matrix to extract the first few eigenmodes with the largest energy to span a low-dimensional subspace. Finally, projecting the partial differential equations of the high-fidelity three-dimensional physical model onto this low-dimensional subspace yields a set of low-dimensional ordinary differential equations with a few modal coefficients as state vectors, i.e., a reduced-order heat transfer model. The formulas for the low-dimensional ordinary differential equations are as follows: ; in, The state vector, also known as the modal coefficient vector, represents the reduced-order heat transfer model. It is a ra-dimensional column vector. (superscript " "" indicates transpose, where each component These are time-varying coefficients, which can be understood as the weights or projections of the "complete temperature field" onto the corresponding "characteristic modes." This is the most crucial variable in the entire order reduction method, used to represent the temperature field in all subsequent steps. `ra` represents the dimension of the reduced heat transfer model, i.e., the number of characteristic modes retained. Typically, `ra` is much smaller than the number of grid points in the original model, which significantly reduces computational complexity, achieving a leap from "unable to be computed in real time" to "able to be computed in real time." State vector The derivative with respect to time t, i.e., the rate of change of the modal coefficients with time. This represents the reduced-order system matrix, with dimensions ra×ra. It is obtained by projecting from the higher-dimensional system and determines the internal dynamic characteristics of the reduced-order system (such as energy decay, oscillation, etc.). Let represent the input matrix, with dimensions ra × ma. It establishes the relationship between ma sensor inputs and a ra-dimensional state vector, defining how external inputs affect the system state. The input vector is composed of a finite number of sensor measurements. (superscript " "" indicates transpose), and the unit is Kelvin (K). It serves as a known condition to drive the calculation of the entire reduced-order model. The purpose of this formula is to transform the computationally intensive three-dimensional spatial distribution problem into a fast calculation problem that only requires solving a system of ra-dimensional ordinary differential equations, thus laying the foundation for real-time simulation. The specific operation of inputting the measured values ​​as boundary conditions into the reduced-order heat transfer model and driving the calculation is as follows: During the online operation of the injection molding process, the measured values ​​of multiple temperature sensors acquired in real time are arranged in a preset order to form a sensor measurement value vector; The sensor measurement vector is used as the input signal and injected into the port defined by the input matrix corresponding to the reduced-order heat transfer model; The modal coefficient vector at the current moment is calculated by solving a system of low-dimensional ordinary differential equations with modal coefficients as state vectors using numerical integration. Finally, by summing all the products obtained by multiplying the average temperature field calculated offline in advance with each modal coefficient in the modal coefficient vector at the current moment by its corresponding characteristic mode, a complete temperature field distribution data including the area on the mold cavity surface where no sensors are deployed is reconstructed. The formula is as follows: ; in, This represents the complete temperature field distribution data, i.e., at three-dimensional spatial location (x, y, z) and time. The temperature value is used as the sole and authoritative data basis for all subsequent steps (prediction, identification, and control), enabling temperature monitoring of the mold surface without blind spots, thus overcoming the limitations of traditional limited-point measurements. The average temperature field is a static temperature field that does not change over time. It is usually obtained by averaging all temperature field snapshots from the offline sampling phase and is used as a baseline or reference state for temperature field calculation. It represents an average thermal equilibrium point of the system, and the reconstructed temperature field is superimposed on this baseline with dynamically changing conditions. This represents the modal coefficients or state vectors, which is the first... The modality at time... The weights or magnitudes of the modal coefficients constitute the state vector. This is the core of the reduced-order heat transfer model, which represents the high-dimensional temperature field using a low-dimensional vector, making real-time and fast calculations possible. The system dynamics are entirely described by the evolution of this low-dimensional vector. Indicates the first The first characteristic mode, also known as the basis function or POD mode, is a set of fixed spatial distribution functions obtained from offline analysis, representing the most important modes of temperature field change. It is equivalent to a set of "standard temperature change models". The actual temperature field at any time can be regarded as these standard models with different weights ( The result of linear superposition is shown in the figure. Here, 'ra' represents the dimension of the reduced-order heat transfer model. It is an integer, and 'ra' is much smaller than the number of grid points in the original model. This determines the complexity and computational speed of the reduced-order model. The more modes retained, the higher the reconstruction accuracy, but the greater the computational cost. A trade-off between accuracy and efficiency is necessary. Representing three-dimensional spatial coordinates, used to locate any point inside the mold, and to clarify that the temperature field is a function defined in three-dimensional space, rather than just a few discrete measurement points. Indicates the first The discrete time points (or control cycles) are used to emphasize that the entire reconstruction process is performed in real time and discretely, with each control cycle outputting the latest complete temperature field. The summation symbol represents the linear superposition of modes from the first to the rath mode, describing how the contributions of each mode are combined to form a complete dynamic temperature change component.

[0021] In this embodiment, it is specifically necessary to explain the specific operation of training a long short-term memory network prediction model in step S2 during the offline preparation stage: A1. Extract data from the historical database for multiple consecutive injection cycles. The data for each cycle includes complete temperature field distribution data obtained from step S1 at different historical time points, as well as process parameter data collected at the same time point as the complete temperature field distribution data to describe the state of the injection process. The process parameter data includes melt temperature, injection speed, holding pressure and cooling water valve opening. A2. Combine the complete temperature field distribution data obtained at the same time point with the process parameter data to form a fused feature vector; A3. A series of continuous fused feature vectors are extracted as model inputs using a preset time window length. The time window length is defined as the number of consecutive historical moments. The change in the complete temperature field distribution data at the end of the time window relative to the beginning of the time window is used as the target output for model training. This forms a training sample set with a large number of training sample pairs. A deep long short-term memory network is trained using the training sample set. The long short-term memory network is configured with an input gate, a forget gate, and an output gate. A3.1 The forgetting gate calculates the forgetting factor based on the fused feature vector at the current time step and the hidden state at the previous time step, which is used to control the degree of retention of historical information; A3.2 The input gate calculates the input factor based on the fused feature vector at the current time step and the hidden state at the previous time step, which is used to control the degree of addition of new information; A3.3 Long Short-Term Memory Networks also calculate candidate cell states, which are updated by adding the product of the input factor and the candidate cell state to the result of multiplying the cell state of the previous time step by the forgetting factor. A3.4 Long Short-Term Memory (LSTM) networks output predicted values ​​of future temperature field changes through fully connected layers; A4. The loss function used during training is the weighted mean square error between the predicted value and the true value, and higher weights are given to high-temperature regions in the complete temperature field distribution data so that the model can focus more on predicting regions prone to thermal defects. A5. Obtain a trained Long Short-Term Memory (LSTM) network prediction model; In online operation, the specific steps to predict the future temperature field change trend data are as follows: In each control cycle, the complete temperature field distribution data of the current cycle is received in real time from step S1, and process parameter data related to the current process state are obtained in the current and past periods, wherein the length of the past period is defined by the preset time window length. The complete temperature field distribution data and process parameter data at the current moment are concatenated to form the fusion feature vector at the current moment, and then combined with the fusion feature vectors of multiple consecutive past moments in chronological order to form an input sequence with the same time window length as preset in the offline training phase. The input sequence is fed into the trained Long Short-Term Memory (LSTM) network prediction model. This LSM network prediction model processes the input sequence through its internal forget gate, input gate, and output gate mechanisms, updates its internal cell states and hidden states, and finally outputs a predicted value of the temperature field change trend data after the next prediction step. This predicted value represents the change in the complete temperature field distribution data at the time after the next prediction step relative to the current control cycle. The next prediction step (i.e., the next n steps) is a preset positive integer prediction step (e.g., n=1 means predicting the next control cycle, n=2 means predicting the next two control cycles). The length of the control cycle depends on the system sampling rate (e.g., one cycle per second).

[0022] In this embodiment, the specific process of obtaining a preliminary future temperature field estimate in step S3 is as follows: In each control cycle, the complete temperature field distribution data of the current control cycle is received from the real-time output of step S1, and the predicted value of the temperature field change trend data after the future prediction step size is received from step S2. Subsequently, using a preset linear superposition model, the complete temperature field distribution data of the current control cycle and the predicted values ​​of the temperature field change trend data after the future prediction step are algebraically added together to directly calculate the preliminary future temperature field estimation data, thereby completing the preliminary extrapolation from the current state to the future state, which is the basis for subsequent physical correction. Its effect is to seamlessly integrate the "current holographic perception" in step S1 with the "future trend prediction" in step S2. The specific steps for verifying and correcting the physical feasibility of the preliminary future temperature field estimate data by applying a physical constraint correction process based on a simplified thermodynamic conservation equation are as follows: First, a simplified thermodynamic conservation equation for rapid verification is constructed. This equation is based on the principle of energy balance and correlates the rate of change of the temperature field per unit time with the heat conduction term and the internal heat source term. The rate of change of the temperature field per unit time is obtained by dividing the difference between the preliminary estimated future temperature field data and the complete temperature field distribution data of the current period by the prediction time step (the prediction time step is equal to the future prediction step multiplied by the control period duration). The heat conduction term is calculated by the thermal conductivity of the mold material and the spatial second derivative of the preliminary estimated future temperature field data. The internal heat source term represents the average heat flux density transferred by the plastic melt per unit volume. Then, the preliminary future temperature field estimate data is substituted into the simplified thermodynamic conservation equation, and the difference between the energy change rate represented by the left side of the equation and the heat flow condition represented by the right side is calculated. This difference represents the degree of energy imbalance, and the difference is compared with a preset physical rationality threshold. When the difference exceeds the physical rationality threshold (which is preset through statistical analysis of historical process data to ensure that no correction is needed under more than 95% of normal operating conditions), the correction optimization process is initiated. This process solves for the corrected temperature field by minimizing a physical residual function. The physical residual function includes a first term and a second term. The first term represents the norm square of the difference between the corrected temperature field and the current period's complete temperature field distribution data relative to the predicted time step and the difference between the combination of the heat conduction term and the internal heat source term. It is used to measure the degree to which the physical conservation law is satisfied. The second term represents the norm square of the difference between the corrected temperature field and the preliminary future temperature field estimate data multiplied by a regularization parameter. It is used to measure the fidelity to the original prediction value. The corrected temperature field that minimizes the physical residual function is solved through an iterative optimization algorithm, thereby generating the final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation. The specific process for identifying potential overheating risk areas and their risk levels is as follows: The final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation is compared point by point with a preset temperature threshold matrix related to material processing to identify all spatial locations where the temperature value exceeds the corresponding temperature threshold. Next, using the connected component analysis algorithm in image processing, adjacent overheating locations in space are merged and connected, and marked as several consecutive potential overheating risk areas; Then, a comprehensive risk index is calculated for each identified potential overheating risk area. The calculation process of the comprehensive risk index is as follows: obtain the highest temperature value in the risk area, calculate the difference between the highest temperature value and the corresponding threshold in the temperature threshold matrix, and divide the difference by a preset reference temperature rise value for normalization to obtain the normalized overheating amplitude. At the same time, the size of the risk area is obtained, and the proximity index between the risk area and the known vulnerable critical areas in the mold is assessed; Finally, using the preset first weighting coefficient, second weighting coefficient, and third weighting coefficient, the normalized overtemperature amplitude, the area size of the risk region, and the proximity are weighted and summed to obtain the comprehensive risk index of the risk region. Finally, based on the magnitude of the comprehensive risk index of each risk area, it is divided into different risk level intervals. When the comprehensive risk index is less than the first preset threshold, the corresponding risk area is defined as low risk level; when the comprehensive risk index is greater than or equal to the first preset threshold and less than the second preset threshold, the corresponding risk area is defined as medium risk level; and when the comprehensive risk index is greater than or equal to the second preset threshold, the corresponding risk area is defined as high risk level. This completes the quantitative risk classification from simple over-temperature judgment to multi-factor coupling, and provides a direct basis for differentiated and precise control in subsequent steps.

[0023] In this embodiment, it is particularly important to explain step S4, where a multi-objective optimization function aimed at minimizing temperature tracking error and smoothing control actions is used to solve for the optimal control parameters for each independent cooling loop. The specific operation is as follows: First, the final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation generated in step S3 is received as the prediction target of feedforward control. At the same time, the potential overheating risk areas identified in step S3 and their corresponding risk level information are received. The control parameter vectors of each cooling loop at the current moment (such as the set value of the previous cycle) and the target temperature field data (which is uniformly distributed or based on the ideal temperature of the product requirements) are obtained in advance by the process requirements. Next, a multi-objective optimization function to be minimized is constructed. This function consists of two terms. The first term is the temperature tracking error term, which measures the difference between the final future temperature field distribution data generated in step S3 and the preset target temperature field data. It is calculated by integrating the spatial weight function over the entire three-dimensional spatial domain of the mold with the square of the difference between the predicted target temperature field data and the target temperature field data. The value of the spatial weight function is dynamically set according to the spatial point position, specifically: If the spatial point is located within a potential overheating risk area, its weight value is equal to a preset base weight plus a risk weight coefficient multiplied by the risk level of the risk area; If the spatial point does not belong to any risk area, its weight value is equal to the basic weight; The second term is the control action smoothing term, which measures the change in the control parameter vector to be solved and the control parameter vector at the previous time step. It is obtained by calculating the square of the Euclidean distance between the control parameter vector to be solved and the control parameter vector at the previous time step, and multiplying it by a control action smoothing weight coefficient. The first term of this multi-objective optimization function is the risk-weighted temperature tracking error term, which drives the future temperature field to approach the target temperature field and strengthens the tracking of high-risk areas. The second term is the control action smoothing term, which avoids drastic fluctuations in control parameters and improves system stability. Its effect is to transform discrete risk levels into continuous spatial weights, so that the optimization problem can directly respond to risk information. Then, a local linear influence model of the control parameter vector on the final future temperature field distribution data is established. This model indicates that the final future temperature field distribution data is approximately equal to the product of a control gain matrix and a control parameter vector plus a bias vector. The control gain matrix and the bias vector are determined by the thermal response characteristics of the mold through offline simulation or system identification. By substituting the local linear influence model into the multi-objective optimization function, the original nonlinear optimization problem is transformed into a quadratic programming problem with boundary constraints, where the constraints are the physical upper and lower limits of the control parameters of each independent cooling loop. Finally, the sequential quadratic programming algorithm is used, and the interior point method or effective set method is employed to solve the quadratic programming problem to obtain the optimal control parameter vector. The optimal control parameters are pre-issued to the actuator to achieve adaptive feedforward control of the cooling system. This, combined with traditional feedback control, forms a composite control strategy. The specific operation is as follows: The optimal control parameter vector obtained by the solution is used as the feedforward control quantity and issued to the actuators of each independent cooling circuit in advance within the current control cycle; Simultaneously, a conventional feedback controller is run in parallel. This feedback controller takes the deviation between the complete temperature field distribution data of the current period output in real time by step S1 and the preset target temperature field data as input to calculate the feedback control correction amount. Finally, the feedforward control quantity and the feedback control correction quantity are linearly superimposed to form the final composite control signal, which is then applied to the actuator of the cooling system, thereby achieving an effective combination of feedforward control based on future temperature field prediction and feedback control based on current temperature field measurement.

[0024] Example 2 This embodiment provides, for example Figure 2 The system shown is a real-time temperature field monitoring system for the injection molding process of automotive molds, specifically including: The real-time temperature field reconstruction module is used in the online operation of the injection molding process. It is triggered by the preset mold geometry model and thermophysical parameters, and uses the offline constructed calculation-efficient reduced-order heat transfer model to obtain the measurement values ​​of multiple temperature sensors placed on the mold in real time as boundary conditions to drive the reduced-order heat transfer model to perform calculations, thereby outputting a complete temperature field distribution data including the area on the surface of the mold cavity where no sensors are placed. The temperature field change trend prediction module is used to train a long short-term memory network prediction model by using process parameter data and corresponding temperature field data collected in multiple historical injection molding cycles as a training sample set during the offline preparation stage. During online operation, the complete temperature field distribution data of the current cycle and process parameter data related to the current process state, which are output in real time by step S1, are input into the trained long short-term memory network prediction model to predict the temperature field change trend data at future moments. The hotspot risk identification module is triggered by the complete temperature field distribution data of the current period output in real time in step S1 and the temperature field change trend data of future time predicted in step S2. By superimposing the two, a preliminary future temperature field estimate is obtained. Then, a physical constraint correction process based on a simplified thermodynamic conservation equation is applied to verify and correct the physical feasibility of the preliminary future temperature field estimate, generating a final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation, and identifying potential overheating risk areas and their risk levels accordingly. The adaptive feedforward composite control module is used as a feedforward signal to trigger the final future temperature field distribution data and potential overheating risk areas generated in step S3. It combines a multi-objective optimization function with the goals of minimizing temperature tracking error and smoothing control actions to solve for the optimal control parameters for each independent cooling loop. The optimal control parameters are then issued to the actuator in advance to realize adaptive feedforward control of the cooling system. At the same time, it forms a composite control strategy with traditional feedback control.

[0025] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0026] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0027] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0028] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0029] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0030] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0031] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for real-time monitoring of the temperature field during the injection molding process of an automotive mold, characterized in that, Specifically, the following steps are included: Step S1: Based on the preset mold geometry model and material thermal property parameters, an efficient reduced-order heat transfer model is constructed offline. During the online operation of the injection molding process, the measurement values ​​of multiple temperature sensors installed on the mold are acquired in real time, and the measurement values ​​are used as boundary conditions to input into the reduced-order heat transfer model to drive the model to perform calculations, thereby outputting a complete temperature field distribution data that includes the area on the mold cavity surface where no sensors are installed in real time. Step S2: In the offline preparation stage, process parameter data collected in multiple historical injection molding cycles and corresponding complete temperature field distribution data are used to form a training sample set. A long short-term memory network prediction model is trained using this training sample set. In online operation, the complete temperature field distribution data of the current cycle output in real time from step S1 and the process parameter data related to the current process state are input into the trained long short-term memory network prediction model to predict the predicted value of the temperature field change trend data at future moments. Step S3: The complete temperature field distribution data of the current period output in real time from step S1 is superimposed with the predicted value of the temperature field change trend data of future time predicted by step S2 to obtain a preliminary future temperature field estimate. Subsequently, a physical constraint correction process based on a simplified thermodynamic conservation equation is applied to verify and correct the physical feasibility of the preliminary future temperature field estimate, generating a final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation, and identifying potential overheating risk areas and their risk levels accordingly. Step S4: Using the final future temperature field distribution data and potential overheating risk areas generated in step S3 as feedforward signals, and combining them with a multi-objective optimization function aimed at minimizing temperature tracking error and smoothing control actions, the optimal control parameters for each independent cooling loop are obtained. The optimal control parameters are then issued to the actuator in advance to achieve adaptive feedforward control of the cooling system, while forming a composite control strategy with traditional feedback control.

2. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 1, characterized in that: In step S1, the specific process of offline construction of a computationally efficient reduced-order heat transfer model is as follows: First, based on the preset mold geometry model and material thermal properties parameters, including the density, specific heat capacity and thermal conductivity of the mold material, a high-fidelity three-dimensional physical model is established to describe the transient heat conduction process inside the mold using partial differential equations. Subsequently, an intrinsic orthogonal decomposition method is applied to the high-fidelity three-dimensional physical model to reduce its order. The order reduction process is as follows: by sampling and solving in the typical working condition parameter space of the high-fidelity three-dimensional physical model, a series of temperature field snapshots describing its dynamic characteristics are obtained, and these temperature field snapshots are combined to construct a snapshot matrix; singular value decomposition is performed on the snapshot matrix to extract the first few eigenmodes with the largest energy to span a low-dimensional subspace. Finally, the partial differential equations of the high-fidelity three-dimensional physical model are projected onto this low-dimensional subspace to obtain a set of low-dimensional ordinary differential equations with a few modal coefficients as state vectors, namely the reduced-order heat transfer model.

3. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 2, characterized in that: The specific operation of inputting the measured values ​​as boundary conditions into the reduced-order heat transfer model and driving the calculation is as follows: During the online operation of the injection molding process, the measured values ​​of multiple temperature sensors acquired in real time are arranged in a preset order to form a sensor measurement value vector; The sensor measurement vector is used as the input signal and injected into the port defined by the input matrix corresponding to the reduced-order heat transfer model; The modal coefficient vector at the current moment is calculated by solving a system of low-dimensional ordinary differential equations with modal coefficients as state vectors using numerical integration. Finally, by summing the average temperature field calculated offline in advance with all the product results obtained by multiplying each modal coefficient in the modal coefficient vector at the current moment by the corresponding characteristic mode, a complete temperature field distribution data including the area on the surface of the mold cavity where no sensors are deployed is reconstructed.

4. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 3, characterized in that: In step S2, the specific operation of training a long short-term memory network prediction model in the offline preparation phase is as follows: A1. Extract data from the historical database for multiple consecutive injection cycles. The data for each cycle includes complete temperature field distribution data obtained from step S1 at different historical time points, as well as process parameter data collected at the same time point as the complete temperature field distribution data to describe the state of the injection process. The process parameter data includes melt temperature, injection speed, holding pressure and cooling water valve opening. A2. Combine the complete temperature field distribution data obtained at the same time point with the process parameter data to form a fused feature vector; A3. A series of continuous fused feature vectors are extracted as model inputs using a preset time window length. The time window length is defined as the number of consecutive historical moments. The change in the complete temperature field distribution data at the end of the time window relative to the beginning of the time window is used as the target output for model training. This constitutes a training sample set. A deep long short-term memory network is trained using the training sample set. The long short-term memory network is configured with an input gate, a forget gate, and an output gate. A3.1 The forgetting gate calculates a forgetting factor based on the fusion feature vector at the current moment and the hidden state at the previous moment, which is used to control the degree of retention of historical information; A3.2 The input gate calculates the input factor based on the fused feature vector at the current time and the hidden state at the previous time, which is used to control the degree of addition of new information; A3.3 The Long Short-Term Memory Network also calculates candidate cell states, which are updated by adding the result of multiplying the cell state of the previous time step by the forgetting factor to the product of the input factor and the candidate cell state. A3.4 The Long Short-Term Memory Network outputs a predicted value of the temperature field change at future times through a fully connected layer; A4. The loss function used during training is the weighted mean square error between the predicted value and the true value; A5. Obtain a trained Long Short-Term Memory (LSTM) network prediction model.

5. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 4, characterized in that: The specific operation for predicting the future temperature field change trend data during online operation is as follows: In each control cycle, the complete temperature field distribution data of the current cycle is received in real time from step S1, and process parameter data related to the current process state are obtained in the current and past periods, wherein the length of the past period is defined by the preset time window length. The complete temperature field distribution data and process parameter data at the current moment are concatenated to form the fusion feature vector at the current moment, and then combined with the fusion feature vectors of multiple consecutive past moments in chronological order to form an input sequence with the same time window length as preset in the offline training phase. The input sequence is fed into the trained Long Short-Term Memory Network (LSTM) prediction model. The LSM prediction model processes the input sequence through its internal forget gate, input gate, and output gate mechanisms, updates its internal cell states and hidden states, and finally outputs a predicted value of the temperature field change trend data after the future prediction step. This predicted value represents the amount of change in the complete temperature field distribution data at the time after the future prediction step relative to the current control cycle.

6. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 5, characterized in that: In step S3, the specific process of obtaining preliminary future temperature field estimation data is as follows: In each control cycle, the complete temperature field distribution data of the current control cycle is received from the real-time output of step S1, and the predicted value of the temperature field change trend data after the future prediction step size is received from step S2. Subsequently, using a preset linear superposition model, the complete temperature field distribution data of the current control cycle is algebraically added to the predicted values ​​of the temperature field change trend data at the time after the future prediction step, and the preliminary future temperature field estimation data is directly calculated. The specific steps for verifying and correcting the physical feasibility of the preliminary future temperature field estimate data by applying a physical constraint correction process based on a simplified thermodynamic conservation equation are as follows: First, a simplified thermodynamic conservation equation for rapid verification is constructed. This equation is based on the principle of energy balance and associates the rate of change of the temperature field per unit time with the heat conduction term and the internal heat source term. The rate of change of the temperature field per unit time is obtained by dividing the difference between the preliminary estimated future temperature field data and the complete temperature field distribution data of the current period by the prediction time step. The heat conduction term is calculated by the thermal conductivity of the mold material and the spatial second derivative of the preliminary estimated future temperature field data. The internal heat source term represents the average heat flux density transferred by the plastic melt per unit volume. Then, the preliminary future temperature field estimate data is substituted into the simplified thermodynamic conservation equation, and the difference between the energy change rate represented by the left side of the equation and the heat flow condition represented by the right side is calculated. This difference represents the degree of energy imbalance, and the difference is compared with a preset physical rationality threshold. When the difference exceeds the physical reasonableness threshold, a correction optimization process is initiated. This process solves for the corrected temperature field by minimizing a physical residual function. The physical residual function includes a first term and a second term. The first term represents the squared norm of the difference between the corrected temperature field and the current periodic complete temperature field distribution data relative to the prediction time step and the difference between the combination of the heat conduction term and the internal heat source term, which is used to measure the degree of satisfaction of the physical conservation law. The second term represents the squared norm of the difference between the corrected temperature field and the preliminary future temperature field estimate data multiplied by a regularization parameter, which is used to measure the fidelity to the original prediction value. The corrected temperature field that minimizes the physical residual function is solved by an iterative optimization algorithm, thereby generating the final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation.

7. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 6, characterized in that: The specific process for identifying potential overheating risk areas and their risk levels based on this is as follows: The final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation is compared point by point with a preset temperature threshold matrix related to material processing to identify all spatial locations where the temperature value exceeds the corresponding temperature threshold. Next, using the connected component analysis algorithm in image processing, adjacent overheating locations in space are merged and connected, and marked as several consecutive potential overheating risk areas; Then, a comprehensive risk index is calculated for each identified potential overheating risk area. The calculation process of the comprehensive risk index is as follows: obtain the highest temperature value in the risk area, calculate the difference between the highest temperature value and the corresponding threshold in the temperature threshold matrix, and divide the difference by a preset reference temperature rise value for normalization processing to obtain the normalized overheating amplitude. At the same time, the size of the risk area is obtained, and the proximity index between the risk area and the known vulnerable critical areas in the mold is assessed; Finally, using preset first weighting coefficients, second weighting coefficients, and third weighting coefficients, the normalized overtemperature amplitude, the area size of the risk region, and the proximity are weighted and summed to obtain the comprehensive risk index of the risk region. Finally, based on the magnitude of the comprehensive risk index of each risk area, it is divided into different risk level ranges. When the comprehensive risk index is less than the first preset threshold, the corresponding risk area is defined as low risk level; when the comprehensive risk index is greater than or equal to the first preset threshold and less than the second preset threshold, the corresponding risk area is defined as medium risk level; and when the comprehensive risk index is greater than or equal to the second preset threshold, the corresponding risk area is defined as high risk level.

8. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 7, characterized in that: In step S4, a multi-objective optimization function aimed at minimizing temperature tracking error and smoothing control actions is used to solve for the optimal control parameters for each independent cooling loop. The specific operation is as follows: First, the final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation generated in step S3 is received as the prediction target of the feedforward control. At the same time, the potential overheating risk areas identified in step S3 and their corresponding risk level information are received, and the control parameter vectors of each cooling loop at the current moment and the target temperature field data pre-set by the process requirements are obtained. Next, a multi-objective optimization function to be minimized is constructed. This function consists of two terms. The first term is the temperature tracking error term, which is calculated by integrating the spatial weight function over the entire three-dimensional spatial domain of the mold with the square of the difference between the predicted target temperature field data and the target temperature field data. The value of the spatial weight function is dynamically set according to the spatial point position, specifically: If the spatial point is located within a potential overheating risk area, its weight value is equal to a preset base weight plus a risk weight coefficient multiplied by the risk level of the risk area; If the spatial point does not belong to any risk area, its weight value is equal to the basic weight; The second term is the control action smoothing term, which is obtained by calculating the square of the Euclidean distance between the control parameter vector to be solved and the control parameter vector at the previous moment, and multiplying it by a control action smoothing weight coefficient. Then, a local linear influence model of the control parameter vector on the final future temperature field distribution data is established. This model indicates that the final future temperature field distribution data is approximately equal to the product of a control gain matrix and a control parameter vector plus a bias vector. The control gain matrix and the bias vector are determined by the thermal response characteristics of the mold through offline simulation. By substituting the local linear influence model into the multi-objective optimization function, the original nonlinear optimization problem is transformed into a quadratic programming problem with boundary constraints. Finally, the sequential quadratic programming algorithm is used, and the interior point method is employed to solve the quadratic programming problem to obtain the optimal control parameter vector.

9. The method for real-time monitoring of temperature field during automotive mold injection molding process according to claim 8, characterized in that: The specific operation of issuing the optimal control parameters to the actuator in advance to achieve adaptive feedforward control of the cooling system, while forming a composite control strategy with traditional feedback control, is as follows: The optimal control parameter vector obtained by the solution is used as the feedforward control quantity and issued to the actuators of each independent cooling circuit in advance within the current control cycle; Simultaneously, a conventional feedback controller is run in parallel. This feedback controller takes the deviation between the complete temperature field distribution data of the current period output in real time by step S1 and the preset target temperature field data as input to calculate the feedback control correction amount. Finally, the feedforward control quantity and the feedback control correction quantity are linearly superimposed to form the final composite control signal.

10. A real-time temperature field monitoring system for automotive mold injection molding process is applied to a real-time temperature field monitoring method for automotive mold injection molding process as described in any one of claims 1-9, characterized in that: Specifically, it includes: The real-time temperature field reconstruction module is used in the online operation of the injection molding process. It is triggered by the preset mold geometry model and thermophysical parameters, and uses an offline-constructed, computationally efficient reduced-order heat transfer model to obtain the measured values ​​of multiple temperature sensors placed on the mold in real time as boundary conditions. This drives the reduced-order heat transfer model to perform calculations, thereby outputting a complete temperature field distribution data that includes the area on the mold cavity surface where no sensors are placed. The temperature field change trend prediction module is used to trigger the training sample set composed of process parameter data and corresponding temperature field data collected in multiple historical injection molding cycles during the offline preparation stage. It trains a long short-term memory network prediction model and, during online operation, is triggered by the complete temperature field distribution data of the current cycle output in real time in step S1 and the process parameter data related to the current process state. The data is then input into the trained long short-term memory network prediction model to predict the temperature field change trend data at future moments. The hotspot risk identification module is triggered by the complete temperature field distribution data of the current period output in real time in step S1 and the temperature field change trend data of future time predicted in step S2. By superimposing the two, a preliminary future temperature field estimate is obtained. Then, a physical constraint correction process based on a simplified thermodynamic conservation equation is applied to verify and correct the physical feasibility of the preliminary future temperature field estimate, generating a final future temperature field distribution data that satisfies the simplified thermodynamic conservation equation, and identifying potential overheating risk areas and their risk levels accordingly. The adaptive feedforward composite control module is used as a feedforward signal to trigger the final future temperature field distribution data and potential overheating risk areas generated in step S3. It combines a multi-objective optimization function with the goals of minimizing temperature tracking error and smoothing control actions to solve for the optimal control parameters for each independent cooling loop. The optimal control parameters are then issued to the actuator in advance to realize adaptive feedforward control of the cooling system. At the same time, it forms a composite control strategy with traditional feedback control.

Images