Injection quality closed loop control method based on MES data base

Abstract

This invention relates to the field of industrial data processing technology, specifically to a closed-loop control method for injection molding quality based on a MES data base. The method first extracts a full-dimensional contextual feature set from the MES data base and integrates the viscosity fluctuation characteristics of the screw pressure curve acquired in real time to construct an enhanced input vector. Second, it constructs a multi-fidelity Gaussian process proxy model based on production data, using historical molding features from the MES as low-fidelity input and measured part dimensions as high-fidelity calibration, establishing a mapping between process parameters and dimensional deviations through a differential Gaussian process structure. Then, it performs online optimization using Bayesian functions. Finally, it automatically distributes the optimized process parameters through the MES equipment control interface and achieves a self-evolving closed loop based on quality inspection data. This invention deeply integrates historical production data with real-time sensor data, significantly improving the injection molding process's adaptability to material fluctuations and its robustness across different machines, thereby enhancing product quality consistency.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing technology, and in particular to a method and system for closed-loop control of injection molding quality based on a MES data platform. Background Technology

[0002] Manufacturing enterprises have generally introduced MES (Manufacturing Execution System) data management systems in their production and processing plants. However, most current MES systems only cover data recording and process monitoring, and still have the following shortcomings:

[0003] The MES system lacks a real-time parameter reverse compensation mechanism for process deviations caused by batch fluctuations in plastic materials (such as changes in the proportion of recycled materials).

[0004] Existing optimization methods often deviate from the historical production context in MES (such as mold life, machine differences, and environmental temperature and humidity), resulting in optimization results lacking robustness on different machines.

[0005] Multi-fidelity models rely on simulation data and have inherent deviations from actual production. Existing multi-fidelity injection molding optimization technologies all adopt a modeling framework of "simulation data as low fidelity and experimental data as high fidelity." However, injection molding simulation models cannot fully simulate the complex physical processes in actual production (such as the nonlinear effects of mold thermal deformation and melt shear heat generation), resulting in systematic deviations between the simulation-based low-fidelity models and the real production process, thus limiting the practical effectiveness of multi-fidelity optimization.

[0006] Lacking physical constraints, existing Bayesian optimization methods are prone to generating infeasible process parameters. These methods focus solely on quality indicators and fail to consider the thermodynamic and physical boundaries of the injection molding process. The algorithms may generate process parameters that lead to melt degradation, incomplete filling, or equipment overload, posing safety hazards. These parameters require manual verification before production can begin, preventing true automation.

[0007] Open-loop control mode relies on manual intervention. Most existing technologies are still in the open-loop mode of algorithm-recommended parameters-manual confirmation-manual input of injection molding machine. The execution of optimization results depends on the operator's judgment, the response is lagging and there is human error, and it is impossible to achieve real-time closed-loop control of production fluctuations. Summary of the Invention

[0008] To address the problems in existing technologies, such as process deviations caused by material batch fluctuations, poor robustness due to optimization results being out of context of production, large discrepancies between simulation-driven multi-fidelity models and actual production, and inability to achieve closed-loop control due to reliance on manual intervention, this invention provides a closed-loop control method for injection molding quality based on MES data, comprising the following steps:

[0009] S1. Extract the full-dimensional contextual feature set of injection molding production tasks from the MES data base;

[0010] S2. Collect the real-time material viscosity fluctuation characteristics of the injection molding machine screw during the injection stage, and concatenate them with the full-dimensional context feature set to form an enhanced input vector;

[0011] S3. Construct a multi-fidelity Gaussian process proxy model based on production data, and use this model to predict product quality indicators using historical MSE data.

[0012] S4. Sample multiple sets of candidate process parameters, construct a Bayesian function with thermodynamic and physical hard constraints, combine it with the enhanced input vector and predicted product quality indicators, use the Bayesian function to iteratively optimize, and output the optimal process parameters.

[0013] Preferably, in step S1, the full-dimensional context feature set includes the rheological properties of the current material batch, the cumulative number of times the current mold has been used, the historical performance deviation characteristics of the injection molding machine, environmental temperature and humidity data, and the calculated equivalent melt flow rate deviation and mold parting surface gap compensation factor.

[0014] Preferably, the expression for calculating the equivalent melt flow rate deviation is:

[0015] ;

[0016] ;

[0017] in, This represents the equivalent melt flow rate of the actual mixture within the hopper. The standard melt flow rate of the raw material. This represents the percentage by weight of the virgin material. The melt flow rate of the sprue. This represents the percentage by weight of the sprue material. This represents the equivalent melt flow rate deviation value. The target reference melt flow rate;

[0018] The formula for calculating the mold parting surface clearance compensation factor is:

[0019] ;

[0020] in, This is the mold parting surface clearance compensation factor. The current cumulative number of mold productions recorded by MES. The last maintenance cycle of the mold recorded by the MES system. The maximum number of standard mold maintenance cycles preset for the MES process library. The material flow sensitivity coefficient. The nonlinear acceleration index of mold wear, This is the measured increase in the parting line gap, recorded through external measurement. Adjust the weighting coefficients for the gaps.

[0021] Preferably, step S2 specifically involves real-time acquisition of the pressure-time curve of the injection molding machine screw during the injection stage, performing first-order derivative on the curve, calculating the real-time material viscosity fluctuation characteristics including the pressure rise slope and the pressure difference at the V / P switching point, and concatenating these characteristics with the full-dimensional context feature set to form an enhanced input vector of dimension (1×N).

[0022] Preferably, in step S3, the input to the multi-fidelity Gaussian process proxy model is the molding feature data of the injection molding machine from the historical data of the MES, and the model calibration label is the measured dimensional deviation of the part entered by the MES quality inspection module. The process parameters are established through the differential Gaussian process structure. The mapping relationship with the final product quality.

[0023] Preferably, the expression for the multi-fidelity Gaussian process surrogate model is:

[0024] ;

[0025] in, For high-fidelity product quality indicators predicted by the model, As a scaling factor, For low-fidelity prediction functions, This is the deviation function.

[0026] Preferably, in step S4, the thermodynamic physical hard constraint is to predict whether the current parameter will cause the product molding process to fail based on the material data and mold feature data in the MES process knowledge base before each process parameter sampling. If it will, the process parameter is determined to exceed the process safety red line, and an infinite penalty is imposed on it and it is directly discarded.

[0027] Preferably, in step S4, the objective function expression for Bayesian function optimization is:

[0028] ;

[0029] in, For the current injection molding process parameters, For the enhanced input vector, The overall cost objective function is... For the critical dimensions of injection molded parts predicted by the multi-fidelity Gaussian process proxy model, The standard target size for this injection molded part as specified in the MES product library. The allowable dimensional tolerances for this injection molded part are: To the current injection molding process parameters The predicted single injection molding cycle time, Based on the injection cycle time, , These are the weighting coefficients.

[0030] Preferably, the process further includes step S5, where the optimal process parameters are written into the injection molding machine PLC controller for production operations. After the quality inspection of the produced products is completed, the measured dimensional data is used as new MSE historical data to update the multi-fidelity Gaussian process proxy model.

[0031] The beneficial effects of this invention are as follows:

[0032] This invention utilizes a multi-fidelity Gaussian process proxy model built on a MES data foundation, eliminating data biases introduced by traditional simulation modeling and effectively improving the accuracy of product quality prediction. By integrating MFR bias, mold clearance compensation factor, and screw dynamic pressure data to form an enhanced input vector, it can adapt to various production conditions such as raw material fluctuations and mold wear in real time, resulting in a more versatile optimization scheme. A thermodynamic hard constraint verification is added before Bayesian parameter sampling to preemptively screen out unreasonable process parameters that may cause melt degradation or insufficient filling, eliminating the need for manual parameter verification. The optimized process parameters can be automatically sent to the injection molding machine PLC for production, and the actual measured data of the finished product updates the proxy model in reverse, forming a closed-loop self-optimization system that significantly reduces manual intervention. At the same time, the objective function takes into account both product dimensional accuracy and production cycle, optimizing the production process while ensuring yield, and reducing raw material loss and production costs. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention. Detailed Implementation

[0034] Example 1:

[0035] Embodiments of the present invention provide a closed-loop control method for injection molding quality based on a MES data base, such as... Figure 1 As shown, it includes the following steps:

[0036] S1. Extract the full-dimensional context feature set of the current production task from the MES data base, including: the rheological properties of the current material batch, the cumulative number of times the current mold is used, the historical performance deviation characteristics of the injection molding machine, the ambient temperature and humidity data, and the calculated equivalent melt flow rate (MFR) deviation and the mold parting surface gap compensation factor.

[0037] The MES material traceability module obtains the mixing ratio of new material and recycled material in the current hopper, and calculates the equivalent melt flow rate (MFR) deviation, which characterizes the fluctuation of raw material flowability. The expression is as follows:

[0038] ;

[0039] ;

[0040] in, The equivalent melt flow rate (g / 10min) of the actual mixture in the hopper. The standard melt flow rate (g / 10min) of the raw material can be obtained from the MES material library; This represents the mass percentage of the raw material, dimensionless; in this example, it is taken as 70%. The melt flow rate (g / 10min) of the sprue (recycled material) is usually greater than that of the virgin material; The percentage of sprue material by mass is dimensionless; in this example, it is taken as 30%. This is the MFR deviation value; The target reference melt flow rate is typically taken as... .

[0041] Based on the cumulative number of mold uses and maintenance records, a nonlinear wear calculation is used to characterize the mold wear level, and the mold parting surface clearance compensation factor is expressed as follows:

[0042] ;

[0043] in, This is the mold parting surface clearance compensation factor, with a value range of 0~1, which is used to compensate for the attenuation of the holding pressure in the subsequent process. The current cumulative number of mold productions recorded by MES. The mold number recorded by MES during the last maintenance / overhaul of the mold. The maximum number of standard mold maintenance cycles preset for the MES process library. This is the material flow sensitivity coefficient; the value is larger for materials with high flowability. This is the nonlinear acceleration index of mold wear, which is usually greater than or equal to 1, indicating that the wear is faster in the later stages. This is the measured increase in the parting line gap value recorded through external measurement (optional parameter). Adjust the weighting coefficients for the gaps.

[0044] The MES data foundation is an underlying support base built upon configuration center, gateway, and end-to-end components. It aggregates heterogeneous data from multiple sources in the production and processing plant, completes data storage and management, and serves as the underlying data infrastructure for microservice architecture.

[0045] S2. Real-time acquisition of the pressure-time curve of the injection molding machine screw during the injection stage. The first derivative of the screw pressure curve from 0.5s to 1.5s during the injection stage is calculated to determine the real-time material viscosity fluctuation characteristics such as the pressure rise slope and the pressure difference at the V / P switching point (the difference between the instantaneous pressure and the reference value). These characteristics are then combined with the full-dimensional context feature set to form an enhanced input vector of dimension (1×N).

[0046] S3. To address the issues of low frequency of high-fidelity dimensional measurement data acquisition and high frequency of low-fidelity machine sensor data acquisition but lack of intuitive correlation, a multi-fidelity Gaussian process proxy model based on production data is constructed. This model differs from simulation-driven multi-fidelity models; its input consists of low-cost, high-frequency low-fidelity data, specifically injection molding machine molding characteristic data (screw pressure curve, peak pressure, filling work, etc.) from MES historical data, derived from process parameters. The data is generated directly in real-time by the injection molding machine; the model calibration label consists of high-cost, low-frequency, high-fidelity data, specifically the measured dimensional deviations of parts entered by the MES quality inspection module, representing the true quality result of the final product after the combined effect of process parameters and the mold; the model establishes process parameters through a differential Gaussian process structure. The model outputs a high-fidelity product quality index with accurate predictions, based on the mapping relationship between the index and the final product quality. The expression is as follows:

[0047] ;

[0048] in, To accurately predict high-fidelity product quality indicators, such as critical dimensions of injection molded parts and product dimensional deviations; The scaling factor represents the linear scaling ratio when converting from a low-fidelity model to a high-fidelity model, reflecting the transmission loss between the pressure sensed by the injection molding machine and the actual pressure in the mold cavity. It is a low-fidelity prediction function, which is a rough prediction of the product quality index after the combined effect of process parameters and mold. It is obtained by training a large amount of historical machine molding characteristic data as samples to capture the basic physical trend of the injection molding process. The bias function is an independent residual correction Gaussian process used to capture residuals that cannot be explained by low-fidelity data (such as random fluctuations caused by uneven sprue material). and It is trained from ≥50 sets of "process parameters-context-measured dimensions" triplet data accumulated in MES.

[0049] When detected When the fluctuation exceeds a preset threshold (e.g., 5%) or the cumulative number of mold cycles exceeds 80% of the maintenance cycle, the multi-fidelity Gaussian process agent model incremental update is automatically triggered.

[0050] S4. Embed Bayesian functions with thermodynamic physical hard constraints in MSE;

[0051] Thermodynamic physical hard constraint means that before each process parameter sampling, based on the material data and mold feature data in the MES process knowledge base, the thermodynamic formula is used to predict whether the current parameter will cause the product molding process to fail (such as melt degradation or incomplete filling). If it will, the process parameter is determined to exceed the process safety red line, and an infinite penalty is imposed on it and it is directly discarded.

[0052] By inputting process parameters that satisfy thermodynamic and physical hard constraints into a multi-fidelity Gaussian process surrogate model, product quality indicators can be predicted, and the merits of this set of process parameters can be quantified.

[0053] Based on the current enhanced input vector and predicted product quality indicators, the Bayesian function is used to continuously iterate and optimize the process parameters (injection pressure, holding time, injection rate, etc.) with the minimum comprehensive target cost as the evaluation criterion.

[0054] The specific logic for determining melt degradation is as follows:

[0055] Increasing injection pressure or injection speed causes the melt to experience intense shear heat. If the final temperature exceeds the material's thermal decomposition critical point, black spots or yellowing will occur. The formula for shear heat generation is:

[0056] ;

[0057] in, This is the actual highest temperature of the melt. The set heating temperature for the injection molding machine barrel. This refers to the estimated injection pressure difference in the process parameter combination. The density of the plastic melt. Specific heat capacity of the plastic melt;

[0058] Retrieve the critical degradation temperature of this material from the MES material library. ,if If so, it is determined that the current process parameters will lead to melt degradation.

[0059] The logic for determining incomplete filling (insufficient glue):

[0060] When the algorithm attempts to reduce pressure, speed, or increase melt viscosity, the pressure at the end of the mold cavity may drop to 0, causing the fluid to stop flowing and the product not to be fully filled.

[0061] The minimum theoretical pressure required to fill the mold cavity is calculated based on the non-Newtonian fluid flow channel pressure drop formula, and the expression is:

[0062] ;

[0063] in, This represents the minimum theoretical pressure required to fill the mold cavity under the current process parameters. The apparent viscosity of the melt is calculated based on the current shear rate and temperature, combined with the MFR deviation. The maximum flow length of the melt in the mold cavity (read from the MES mold feature library). The average wall thickness of the mold product; Injection time in the current process parameters;

[0064] if Greater than the set injection pressure in the candidate parameter combination If the melt temperature is too low or exceeds the maximum mechanical capacity of the injection molding machine, it is determined that the melt will solidify prematurely, resulting in incomplete filling.

[0065] The objective function expression for Bayesian function optimization is:

[0066] ;

[0067] in, The current injection molding process parameters, i.e. the independent variables that need to be optimized, such as injection pressure and holding time, are a set of variables; For the enhanced input vector, The overall cost objective function is... For the critical dimensions of injection molded parts predicted by the multi-fidelity Gaussian process proxy model, The standard target size for this injection molded part as specified in the MES product library. This refers to the allowable dimensional tolerances for the injection molded part (used to perform dimensionless normalization on dimensional deviations). To the current injection molding process parameters The predicted single injection molding cycle time (if the holding pressure time is extended, the cycle time will be longer). Based on the injection cycle time, , These are weighting coefficients, typically In this embodiment, , This means that, under the premise of absolutely ensuring that the product dimensions are qualified (quality first), the molding cycle should be shortened as much as possible (while also taking efficiency into account).

[0068] S5. Through the MSE equipment control interface, the optimal process parameters are written into the injection molding machine PLC controller, and the mold cavity pressure feedback after production is collected in real time. After the quality inspection of this batch of products is completed, the measured size data is used as the new high-fidelity data to update the multi-fidelity Gaussian process proxy model in step S3, so as to realize quality closed-loop control and model self-evolution.

[0069] The MES data base server stores and provides production data such as material rheological properties, mold history, machine parameters, process knowledge base and quality inspection data to perform full life cycle data management.

[0070] When this embodiment is actually applied to the injection molding production process, it uses an edge protocol conversion gateway to realize bidirectional data exchange between the injection molding machine PLC and the MES system. It communicates with the injection molding machine PLC through the OPC-UA protocol to realize real-time acquisition of screw pressure, position, and temperature data at frequencies above 50Hz, as well as automatic distribution of process parameters. A visual monitoring panel is set up to display material fluctuation curves, process parameter adjustment history, comparison of predicted and measured product quality values, and model update status in real time.

[0071] Example 2:

[0072] This embodiment will illustrate how, in the injection molding process, when the viscosity increases due to the addition of 30% recycled material, the MES using the method of Embodiment 1 detects the anomaly by monitoring the screw pressure curve and calculates the control logic for how much to increase the injection pressure and how long to extend the holding time.

[0073] S1. Before the injection molding task begins, the system obtains the material information of the current production batch through the material management module BOM of the MES data base.

[0074] 1) Due to the altered molecular weight distribution caused by secondary shearing, recycled materials typically have a higher melt flow rate (MFR) than virgin materials. The MFR is calculated, followed by the MFR deviation value ΔMFR%; based on mold maintenance records in the MES, the parting line clearance compensation factor is calculated. .

[0075] 2) Dynamic pressure feature extraction: The MES collects data from the internal sensors of the injection molding machine screw in real time at a frequency of 50Hz through the edge gateway, including the screw pressure-time curve P(t), and extracts the pressure rise slope k. If k deviates from the reference value, it is judged as an instantaneous viscosity anomaly, which is then fused with the feature from step S1 to construct an enhanced context vector.

[0076] Pressure curve feature extraction method:

[0077] When the injection molding machine starts running, it captures the screw injection pressure every 10 milliseconds. The system then compares and analyzes the pressure curve of the current cycle with the pressure-time (PT) standard baseline curve stored in the MES:

[0078] Slope analysis: The algorithm calculates the rate of change of pressure as a function of displacement, dP / ds, during the initial injection phase (0.5s-1.5s). Monitoring revealed that the addition of 30% recycled material increased melt flow resistance, causing the pressure rise slope k to increase from the baseline of 120MPa / m to 138MPa / m (an increase of 15%).

[0079] Volume / Pressure (V / P) Switching Point Offset: The system detects an instantaneous pressure rise of 2.5 MPa when the preset switching position is reached.

[0080] Curve feature extraction: MES detected a shift in the screw's pressure-time curve (PT curve) during the injection phase. When filled to 80%, the screw pressure slope k increased by 15% compared to the baseline value, and the time to reach the set V / P switching position was delayed by 0.12 seconds.

[0081] S3. Construct a multi-fidelity Gaussian process proxy model.

[0082] S4: Bayesian adaptive optimization under physical constraints:

[0083] When the MES detects material fluctuations, it automatically triggers a process parameter recalibration process. After determining that compensation is needed, Bayesian optimization is initiated.

[0084] Based on the original model of the machine's sensors, a rapid simulation was first performed using a low-fidelity model (based on historical sensor data) in the MES. The model predicted that if the original process parameters were maintained, the pressure transmission efficiency at the end of the mold cavity would decrease by 8% due to the increase in viscosity. This would cause the dimensional shrinkage of the finished product at the point far from the gate to exceed the tolerance limit by 0.05 mm.

[0085] The algorithm attempts to reduce viscosity by increasing the injection temperature, but the MES process library reports that the material is close to its degradation critical point near the current temperature. Therefore, the temperature adjustment dimension is blocked by the physical constraint module.

[0086] The algorithm then optimizes the two dimensions of injection pressure and holding time:

[0087] Injection pressure compensation: According to the data acquisition function, in order to offset the pressure drop caused by the increase in viscosity, approximately 3% of the injection pressure needs to be compensated.

[0088] Pressure holding time compensation: Since the cooling rate of high-viscosity melt in the runner is inconsistent with that of the raw material, in order to prevent insufficient filling (shrinkage) due to excessively rapid condensation, the acquisition function recommends extending the pressure holding time by 0.5s to ensure the pressure balance of the mold cavity.

[0089] Optimal decision: Finally, find the point in the probability distribution that maximizes the expected improvement value (EI).

[0090] 4): Output of the optimal solution

[0091] After the Bayesian optimization converges in the i-th iteration, the loop ends, and the optimal compensation combination is calculated: injection pressure increased by 3% and holding time extended by 0.5s. This combination has the highest potential for improving dimensional stability in probabilistic prediction.

[0092] S5: Control command issuance and closed-loop verification

[0093] Command execution: The MES system directly sends the optimized process parameters to the PLC control register of the injection molding machine through OPC-UA or internal bus protocol, realizing parameter switching without manual intervention.

[0094] Optimized injection pressure setting: refers to the maximum pressure limit command that the hydraulic system or servo motor needs to output during the injection filling stage of the injection molding machine. 103MPa is the specific pressure value required to compensate for the increase in viscosity, calculated by the Bayesian algorithm.

[0095] The optimized holding time setting refers to the length of time the screw maintains the holding pressure to continue compensating into the mold cavity after reaching the V / P (speed / pressure) switching point. 5.5s is the extended compensation time calculated by the algorithm to compensate for uneven cooling of high-viscosity melts.

[0096] Feedback Correction: After the product of this batch is produced, the MES quality inspection module enters the actual size of the part. At this time, the algorithm compares the actual size (high-fidelity data) with the previous molding prediction (low-fidelity data). If the measured size recovers to within ∓0.01mm, the effectiveness of the compensation logic is verified. The successful "context-parameter-result" triple is stored in the data base, and the surrogate model is updated using the residual correction function of the Gaussian process to ensure that the prediction of the next batch is more accurate.

[0097] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A method for closed-loop control of injection molding quality based on MES data platform, characterized in that, Includes the following steps: S1. Extract the full-dimensional contextual feature set of injection molding production tasks from the MES data base; S2. Collect the real-time material viscosity fluctuation characteristics of the injection molding machine screw during the injection stage, and concatenate them with the full-dimensional context feature set to form an enhanced input vector; S3. Construct a multi-fidelity Gaussian process proxy model based on production data, and use this model to predict product quality indicators using historical MSE data. S4. Sample multiple sets of candidate process parameters, construct a Bayesian function with thermodynamic and physical hard constraints, combine it with the enhanced input vector and predicted product quality indicators, use the Bayesian function to iteratively optimize, and output the optimal process parameters.

2. The injection molding quality closed-loop control method based on MES data base according to claim 1, characterized in that, In step S1, the full-dimensional context feature set includes the rheological properties of the current material batch, the cumulative number of times the current mold has been used, the historical performance deviation characteristics of the injection molding machine, environmental temperature and humidity data, and the calculated equivalent melt flow rate deviation and mold parting surface gap compensation factor.

3. The injection molding quality closed-loop control method based on MES data base according to claim 2, characterized in that, The expression for calculating the equivalent melt flow rate deviation is as follows: ； ； in, This represents the equivalent melt flow rate of the actual mixture within the hopper. The standard melt flow rate of the raw material. This represents the percentage by weight of the virgin material. The melt flow rate of the sprue. This represents the percentage by weight of the sprue material. This represents the equivalent melt flow rate deviation value. The target reference melt flow rate; The formula for calculating the mold parting surface clearance compensation factor is: ； in, This is the mold parting surface clearance compensation factor. The current cumulative number of mold productions recorded by MES. The last maintenance cycle of the mold recorded by the MES system. The maximum number of standard mold maintenance cycles preset for the MES process library. The material flow sensitivity coefficient. The nonlinear acceleration index of mold wear, This is the measured increase in the parting line gap, recorded through external measurement. Adjust the weighting coefficients for the gaps.

4. The injection molding quality closed-loop control method based on MES data base according to claim 1, characterized in that, Step S2 specifically involves real-time acquisition of the pressure-time curve of the injection molding machine screw during the injection stage, performing first-order derivative on the curve, calculating the real-time material viscosity fluctuation characteristics including the pressure rise slope and the pressure difference at the V / P switching point, and then concatenating these characteristics with the full-dimensional context feature set to form an enhanced input vector of dimension (1×N).

5. The injection molding quality closed-loop control method based on MES data base according to claim 1, characterized in that, In step S3, the input to the multi-fidelity Gaussian process proxy model is the molding feature data of the injection molding machine from the historical data of the MES. The model calibration label is the measured dimensional deviation of the part entered by the MES quality inspection module. Process parameters are established through the differential Gaussian process structure. The mapping relationship with the final product quality.

6. The injection molding quality closed-loop control method based on MES data base according to claim 5, characterized in that, The expression for the multi-fidelity Gaussian process proxy model is: ； in, For high-fidelity product quality indicators predicted by the model, As a scaling factor, For low-fidelity prediction functions, This is the deviation function.

7. The injection molding quality closed-loop control method based on MES data base according to claim 1, characterized in that, In step S4, the thermodynamic physical hard constraint is to predict whether the current parameter will cause the product molding process to fail based on the material data and mold feature data in the MES process knowledge base before each process parameter sampling. If it will, the process parameter is determined to exceed the process safety red line, and an infinite penalty is imposed on it and it is directly discarded.

8. The injection molding quality closed-loop control method based on MES data base according to claim 1, characterized in that, In step S4, the objective function expression for Bayesian function optimization is as follows.

9. The injection molding quality closed-loop control method based on MES data base according to claim 1, characterized in that, It also includes step S5, which writes the optimal process parameters into the injection molding machine PLC controller for production operation. After the quality inspection of the produced products is completed, the measured size data is used as new MSE historical data to update the multi-fidelity Gaussian process proxy model.

Images