Control method for medical infusion bottle cap production, electronic equipment and storage medium

Abstract

The invention provides a control method for medical infusion bottle cap production, electronic equipment and a storage medium. The control method comprises the steps that data of a plurality of sensors in an injection molding system are collected in real time; constructing a digital twin mapping system comprising a virtual mapping layer, a data layer, a simulation layer and a simulation and analysis layer; inputting the sensor data into a system to generate dynamic analogue simulation data, and constructing a first parameter prediction model through a data-driven modeling algorithm based on the two types of data; or constructing a second parameter prediction model by using a transfer learning algorithm; based on the parameter prediction model, searching and determining an optimal production parameter from an empirical case set containing historical data and model deduction data by adopting a fuzzy reasoning algorithm; and outputting the optimal parameters to the injection molding system and updating the optimal parameters in real time. Accurate mapping and dynamic simulation are realized by fusing digital twinning and sensor data, optimal production parameters are determined in combination with a parameter prediction model and a fuzzy reasoning algorithm, production accuracy, stability and efficiency are improved, and product quality is guaranteed.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent control, and in particular to intelligent control methods, electronic devices, and storage media for the production of medical infusion bottle caps. Background Technology

[0002] In the medical field, infusion bottle caps are a key component of drug packaging, and their quality directly affects the safety and stability of the drugs. In particular, the fitting precision requirements between the infusion bottle cap, the bottle body, and the rubber stopper are extremely high, with dimensional tolerances typically controlled within ±0.01mm. Furthermore, stringent requirements are placed on surface smoothness and the amount of debris. Therefore, the manufacturing process of medical infusion bottle caps must be strictly controlled.

[0003] Currently, relevant technologies have equipped visual inspection systems (such as laser rangefinders) and corresponding visual inspection methods to detect defects in each batch of products. However, most of these systems only transport products to the visual inspection system for defect screening after the products have been molded and demolded. This makes it impossible to monitor and adjust parameters in real time during the production process. Furthermore, the detection precision and accuracy of the relevant visual inspection systems and methods are insufficient, resulting in defective infusion bottle caps not being identified and still being transported to the next production process. This leads to wasted costs and may even result in irreversible consequences. Summary of the Invention

[0004] This disclosure provides an intelligent control method, electronic device, and storage medium for the production of medical infusion bottle caps, which at least solves the technical problems in the related art where parameters in the processing and production process cannot be controlled in real time, and the accuracy or precision of the related visual inspection methods is not high.

[0005] In a first aspect of this disclosure, an intelligent control method for the production of medical infusion bottle caps is provided. This method is applied to a medical infusion bottle cap injection molding system, which includes multiple sensors. The intelligent control method includes: acquiring data from multiple sensors in the medical infusion bottle cap injection molding system in real time; constructing a digital twin mapping system for the medical infusion bottle cap injection molding system, wherein the digital twin mapping system includes a virtual mapping layer, a data layer, a model layer, and a simulation and analysis layer; inputting the data from multiple sensors into the digital twin mapping system to obtain dynamic simulation data; based on the multiple sensor data and the dynamic simulation data, constructing a first parameter prediction model using a data-driven modeling algorithm; or constructing a second parameter prediction model using a transfer learning algorithm; based on the first parameter prediction model or the second parameter prediction model, determining the optimal production parameters from an experience case set using a fuzzy inference algorithm, wherein the experience case set includes historical data and model deduction data; and outputting the optimal production parameters to the medical infusion bottle cap injection molding system and updating them in real time.

[0006] In a second aspect of this disclosure, an intelligent control device for the production of medical infusion solutions is provided, which is applied to a medical infusion bottle cap injection molding system. The medical infusion bottle cap injection molding system includes multiple sensors. The intelligent control method includes: an acquisition module for acquiring data from the multiple sensors in the medical infusion bottle cap injection molding system in real time; a parameter prediction model generation module for constructing a digital twin mapping model of the medical infusion bottle cap injection molding system, inputting the multiple sensor data into the digital twin mapping system to obtain dynamic simulation data, and generating a parameter prediction model based on the multiple sensor data and the dynamic simulation data; an optimal production parameter determination module for determining the optimal production parameters based on the parameter prediction model and using a fuzzy inference algorithm to search from an experience case set, wherein the experience case set includes historical data and model inference data; and an output and update module for outputting the optimal production parameters to the medical infusion bottle cap injection molding system and updating them in real time. This intelligent control method achieves precise mapping and dynamic simulation of the medical infusion bottle cap injection molding system through the fusion of digital twin and sensor data. Combined with parameter prediction models and fuzzy inference algorithms, it can efficiently determine the optimal production parameters, helping to improve the accuracy, stability and efficiency of medical infusion bottle cap injection molding production and ensuring the quality of medical infusion bottle caps.

[0007] In a third aspect of this disclosure, an electronic device is provided, comprising: a processor and a memory for storing processor-executable instructions, wherein the processor is configured to execute the executable instructions to implement the method described above.

[0008] In a fourth aspect of this disclosure, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the method described above.

[0009] Compared with related technologies, the embodiments of this disclosure construct a digital twin mapping system corresponding to the medical infusion bottle cap injection molding system. Utilizing data from multiple sensors and dynamic simulation data generated by the digital twin mapping system, a parameter prediction model is generated. Based on this model, an optimal production parameter is determined from a set of experience cases using a fuzzy inference algorithm. This optimal production parameter is then output to the medical infusion bottle cap injection molding system and updated in real time. This method can predict defects (surface defects, chip risk) after product molding in advance and apply the optimal production parameters to the medical infusion bottle cap injection molding system in real time, improving the accuracy, stability, and efficiency of medical infusion bottle cap injection molding production and ensuring the production quality of medical infusion bottle caps.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 The sensor composition structure of a medical infusion bottle cap injection molding system provided in some embodiments of this disclosure is shown;

[0013] Figure 2 A flowchart is shown below illustrating an intelligent control method for the production of medical infusion bottle caps provided in some embodiments of this disclosure;

[0014] Figure 3 A flowchart illustrating the generation of a parameter prediction model in an intelligent control method for the production of medical infusion bottle caps provided in some embodiments of this disclosure is shown.

[0015] Figure 4 The flowchart shown is a process for constructing a second parameter prediction model in an intelligent control method for the production of medical infusion bottle caps provided in some embodiments of this disclosure;

[0016] Figure 5 The structure of an intelligent control device for the production of medical infusion bottle caps provided in some embodiments of this disclosure is shown;

[0017] Figure 6 The structure of an electronic device provided in some embodiments of this disclosure is shown. Detailed Implementation

[0018] To better understand the above technical solutions, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] It should be noted that any variation of the terms "comprising" and "having" is intended to cover inclusion that does not include other components, such as components A, B, and C, but not necessarily those explicitly listed, but may include other components not explicitly listed.

[0020] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0021] Medical infusion bottle caps are a key component of infusion containers, primarily functioning to seal medication, maintain sterility, facilitate puncture and drug administration, or connect to infusion sets. They play a crucial role in ensuring drug safety, efficacy, and ease of clinical use. Currently, defect detection in medical infusion bottle cap production largely relies on visual inspection after demolding to screen out qualified products. However, this method cannot adjust processing parameters in real-time during production, and existing visual inspection methods suffer from low precision and accuracy. Consequently, a small number of defective medical infusion bottle caps are still produced and conveyed to the next process, leading to problems such as poor sealing, leakage, and debris in subsequently produced infusion bottles. This results in wasted production costs and even irreversible consequences. Therefore, strict control over the processing of medical infusion bottle caps is necessary.

[0022] To address the aforementioned issues, this disclosure proposes an intelligent control method for the production of medical infusion bottle caps, as exemplarily described in its embodiments. This method combines the concepts of the Internet of Things (IoT) and digital twins, establishing a mapping relationship between physical entities and virtual systems. It analyzes real-time data acquired during the production process and historical data, uses a parameter prediction model to predict production conditions, and references historical experience data. When abnormal production parameters are detected (potentially leading to defects in the medical infusion bottle caps), timely feedback is provided to the production control center, enabling real-time adjustment of production parameters during injection molding and improving production yield. Defects in medical infusion bottle caps include at least surface defects, such as uneven surfaces, bubbles, burrs, shrinkage marks, and flash; and insufficient chip removal during demolding and puncture processes.

[0023] In some implementations, the intelligent control method for the production of medical infusion bottle caps can be applied to a medical infusion bottle cap injection molding system. For example... Figure 1 As shown, Figure 1The structure of a sensor assembly in an exemplary medical infusion bottle cap injection molding system is shown. This system includes multiple sensors for detecting various production data during the manufacturing process, all connected to a control center 100. Typically, the medical infusion bottle cap injection molding process comprises four main stages: filling, holding pressure, cooling, and demolding. In the filling stage, the screw moves forward, pushing molten material (melt) into the mold cavity; in the holding pressure stage, the screw maintains a holding pressure, and the mold is filled with material; in the cooling stage, the molded part is sufficiently cooled by a coolant flow; finally, in the demolding stage, the mold opens, and the molded part is ejected. To cover the detection data at each stage of the injection molding process, in some embodiments, multiple sensors may include: a piezoelectric pressure sensor 101 disposed on the nozzle / mold cavity runner for detecting the pressure of the molten material, such as a Kistler 6183A; an infrared temperature sensor 102 disposed on the nozzle outlet for detecting the melt temperature, such as an Optris CTlaser 3M; an embedded PT100 thermocouple 103 disposed on the mold core and cooling channels for detecting the mold temperature, such as an OMEGA F2020; a strain gauge sensor 104 disposed on the tie rod / moving platen for measuring the clamping force, such as an HBM U9C; and a magnetostrictive displacement sensor 105 disposed on the injection cylinder for measuring the screw position, such as an MTS Temposonics R series, etc. In addition, the injection molding system may also include an online vision inspection device 106 installed on the conveyor belt after ejection, which is used to detect information such as bottle cap size, for example, using the Keyence CV-X series; it may also be equipped with a high-speed microscopic camera and an AI image processing device 107 installed in the observation window embedded in the mold cavity, which is used to observe the surface condition of the bottle cap, chip situation, etc., for example, using the Phantom TMX 7510.

[0024] In some implementations, such as Figure 2 As shown, Figure 2 The present disclosure illustrates an intelligent control method for the production of medical infusion bottle caps, which includes steps S101 to S105.

[0025] Step S101: Real-time acquisition of data from multiple sensors in the medical infusion bottle cap injection molding system.

[0026] In some implementations, this step involves acquiring key sensor data from the medical infusion bottle cap injection molding system in real time, including machine status data such as melt pressure, melt temperature, mold temperature, clamping force, screw position, and other measured values; it may also include real-time results from an automatic vision inspection system, such as the appearance and size of the bottle cap, as well as images of bottle cap debris captured by a high-speed microscopic camera and processed data from an AI image processing device.

[0027] Step S102: Construct a digital twin mapping system for the medical infusion bottle cap injection molding system, wherein the digital twin mapping system includes a virtual mapping layer, a data layer, a model layer, and a simulation and analysis layer.

[0028] In some implementations, a digital twin mapping system will be used to help obtain predicted and optimized production parameters. It should be noted that a digital twin mapping system is a virtual representation of a physical system, capable of simulating various stages of the physical system's production process and using simulation, machine learning, and inference to assist decision-making through real-time data updates. The core features of a digital twin mapping system include three basic components: a physical object, a virtual counterpart, and a dynamic connection path that facilitates continuous interaction between the two. This can include a virtual mapping layer, a data layer, a model layer, and a simulation and analysis layer. Figure 3 As shown, the specific construction steps S1021 to S1024 of the digital twin mapping system will be described below.

[0029] In some implementations, the construction of a digital twin mapping system for a medical infusion bottle cap injection molding system may specifically include:

[0030] Step S1021: Based on the geometry and structure, physical properties, behavior and state information of the medical infusion bottle cap injection molding system, construct a virtual mapping layer corresponding to the medical infusion bottle cap injection molding system.

[0031] The core of digital twins is the accurate mapping of physical entities, requiring the complete reproduction of the static characteristics and dynamic behavior of physical objects. In some implementations, geometric and structural data can include the three-dimensional structure, dimensions, and shape of the injection molding system, such as the assembly structure of components in industrial equipment, which can typically be constructed using technologies such as CAD, laser scanning, and CT / MRI. Physical property data includes intrinsic attributes such as material properties, mechanical parameters, and chemical properties, providing basic data for simulation analysis, such as the density, elastic modulus, and heat resistance of melt materials. Behavioral and state data includes descriptions of the dynamic response of the injection molding system under different conditions, such as the curves of the injection molding system's rotational speed and temperature changing with load.

[0032] Step S1022: Construct a data layer, wherein the data layer includes at least one of the following: real-time data from multiple sensors of the medical infusion bottle cap injection molding system, historical data, and environmental data.

[0033] Digital twins rely on the real-time input and fusion of massive amounts of data, involving the collection, processing, storage, and transfer of data between the physical and virtual models. In some implementations, the data layer may include: real-time sensor data, which may include status data of the injection molding system, such as temperature, pressure, and images; historical data, which may include historical operation records, fault logs, and maintenance files of the medical infusion bottle cap injection molding system, as well as process data related to the production and processing of the injection molding system (such as production plans, supply chain information, and user operation records), which can be used for model training and trend analysis; environmental data, which may include external parameters of the environment in which the injection molding system operates (such as temperature, humidity, and air pressure), which can be used to compensate for the interference of environmental factors on production parameters; or a combination of one or more of the above data.

[0034] It should be noted that in the data layer, various acquired data need to be cleaned and fused to ensure the accuracy and consistency of the data. Existing methods can be used for data cleaning and fusion, which will not be described in detail here.

[0035] Step S1023: Obtain at least one type of data from the data layer, integrate geometric and structural, physical attributes, behavioral and state information, and rule knowledge to construct the model layer.

[0036] The model layer is the core of the digital twin mapping system. Its essence is a digital mirror image of the physical entity. It requires a comprehensive analysis of the physical entity to extract key features and parameters to provide a basis for modeling. After the model layer is completed, it can realize the abstraction of the laws of the physical entity and dynamic deduction.

[0037] In some implementations, the geometric and structural dimensions represent the spatial form of the physical entity. When constructing the model layer, it is necessary to accurately map the three-dimensional structure, assembly relationships, and key dimensions of the medical infusion bottle cap injection molding system to provide a spatial carrier for subsequent simulations. The physical attribute dimension represents physical characteristics and laws. When constructing the model layer, the material properties of the melt and the effects of physical fields such as pressure, temperature, and injection speed during the injection molding process are considered, serving as the core engine for simulation analysis. The behavioral and state information dimension represents the dynamic operation process, describing the state changes of each stage of the injection molding process over time, enabling the model layer to achieve dynamic twinning. The rule dimension represents the integration of constraints and knowledge, transforming domain knowledge, production specifications, and safety restrictions into rules that the model can recognize, thereby ensuring the rationality of the simulation results. Based on the data in the data layer, the above-mentioned geometric, physical, behavioral, and rule-based dimensions are integrated to construct a model layer that ensures the accuracy of the digital twin mapping system in quality control and production prediction.

[0038] Step S1024: Construct a simulation and analysis layer to synchronously simulate the current state of the medical infusion bottle cap injection molding system and dynamically simulate the state, behavior, or future trend of the medical infusion bottle cap injection molding system.

[0039] In some implementations, within the simulation and analysis layer, on the one hand, the current state of the medical infusion bottle cap injection molding system is accurately and synchronously simulated based on real-time data from multiple sensors, achieving a real-time mirroring of the physical system. For example, the temperature field distribution in the virtual mapping layer can be adjusted in real-time based on temperature data measured by temperature sensors; or the magnitude and distribution of pressure in the model can be dynamically changed to ensure that the virtual mapping layer maintains a high degree of consistency with the operating state of the medical infusion bottle cap injection molding system. On the other hand, based on the constructed parameter prediction model and data in the data layer, new design schemes and / or production parameters can be simulated and verified, and predictive simulations can be performed on the future trends of the injection molding system. By simulating different production parameters or their combinations, a scientific basis can be provided for production decisions.

[0040] Step S103: Input multiple sensor data into the digital twin mapping system to obtain dynamic simulation data. Based on the multiple sensor data and the dynamic simulation data, construct a first parameter prediction model using a data-driven modeling algorithm; or construct a second parameter prediction model using a transfer learning algorithm.

[0041] In some implementations, data-driven modeling methods are used to abstract and dynamically deduce the patterns of the medical infusion bottle cap injection molding system, and based on this, a first parameter prediction model is constructed. More specifically, various scenarios can be simulated in a digital twin mapping system, enabling the virtual mapping layer to form a comprehensive representation of the medical infusion bottle cap injection molding process, observe the production results under different production parameters, and use data-driven modeling methods (such as regression analysis) to construct a first parameter prediction model, thereby achieving real-time monitoring, fault detection, and decision support for the medical infusion bottle cap injection molding system.

[0042] In some implementations, to improve the pass rate of medical infusion bottle cap production, the first parameter prediction model includes at least a surface defect prediction model and a chip shedding risk prediction model. The surface defect prediction model is used to predict the probability of burrs, shrinkage marks, cracks, etc., while the chip shedding prediction model is used to predict the chip shedding risk in scenarios such as demolding and puncture.

[0043] Specifically, based on the dynamic characteristics and material properties of the simulated injection molding process in the digital twin system, the severity of surface defects can be predicted and calculated using the following formula:

[0044] (1)

[0045] In the above formula (1), D represents the severity of the surface defect. The holding pressure fluctuation value is T1, the melt temperature is T2, the mold temperature is t1, the cooling time is V, the injection speed is V, and MFR represents the melt flow rate. This serves as a benchmark value for pressure fluctuation during pressure holding. This serves as a reference value for the rate of temperature change. This serves as a baseline value for injection speed. The baseline values ​​for the melt flow rate are given, and these baseline values ​​can be obtained through Design of Experiments (DOE) or historical production databases. k0 to k4 are model parameters, which can all be obtained by fitting dynamic simulation data.

[0046] Understandably, when The larger the temperature difference between the melt and the mold, the more prone the bottle cap surface is to shrinkage marks after demolding. Insufficient holding pressure can also lead to regional depressions. When the temperature difference between the melt and the mold is too large, the surface is prone to stress shrinkage due to a sudden drop in temperature and rapid cooling. Insufficient cooling time can result in uneven surface hardening and depressions. Excessive injection speed can cause melt turbulence, leading to scratches or weld lines on the surface. When the MFR (Medium-to-Flat Rate) is too high, the melt can accumulate in dead corners of the mold, creating flash; when it is too low, insufficient filling leads to surface depressions. Therefore, various production parameter scenarios can be simulated in a digital twin mapping system to obtain dynamic simulation data, which can then be fitted using the least squares method to obtain the aforementioned model parameters. In some implementations, k0 can range from -2.8 to -0.3, k1=0.3, k2=0.04, k3=0.05, and k4=0.02.

[0047] Furthermore, due to the special nature of medical infusion bottle caps, the risk of debris, impurities, or detachment from the surface of the cap after demolding also needs to be considered. Debris risk prediction mainly involves anticipating the possibility of cap material debris detaching during the injection molding process and subsequent storage, transportation, and use (such as puncture and drug administration, and infusion procedures). Specifically, chipping can occur in the following ways: during injection molding, unreasonable mold structure or improper process parameter settings (such as temperature, pressure, and holding time) can cause burrs or micro-cracks on the edges and sealing surfaces of the bottle cap, leading to chipping during demolding or subsequent processing; or during storage and transportation, defects in the bottle cap material itself (such as excessive brittleness), external pressure, or friction can cause chipping; or during clinical use, when a puncture needle punctures the rubber stopper or plastic cap, inappropriate puncture angle or force, or problems with the bottle cap material or structural design can cause chipping that may mix into the medication, affecting medication safety.

[0048] Specifically, the chipping risk index can be calculated using the following formula:

[0049] (2)

[0050] In the above formula (2), F represents the chipping risk index, and P 熔体 t1 represents the melt pressure, and t2 represents the injection molding time. V represents the rate of change of melt pressure over time, reflecting how quickly the melt pressure rises or falls during injection molding; 剪切速率 The melt shear rate represents the rate of shear deformation caused by the velocity gradient when the melt flows in the screw or flow channel. It can be obtained by calculating the differential of the screw displacement or by simulating the injection flow using simulation software such as Moldflow and outputting the shear rate field. The temperature difference of the melt within the mold reflects the degree of unevenness in the temperature distribution of the mold. This is the baseline value for the rate of change of melt pressure over time. This serves as a baseline value for the melt shear rate. The baseline values ​​for the temperature difference of the melt within the mold are provided. These baseline values ​​for each parameter can be obtained through Design of Experiments (DOE) or historical production databases. α, β, and γ are model parameters, which can be determined using dynamic simulation data in conjunction with the melt material properties.

[0051] Understandably, when During rapid changes, the instantaneous release or concentration of stress within the melt can lead to localized material breakage and spalling, resulting in debris; when V 剪切速率 Excessive heat will increase friction between the melt and the screw and mold surfaces, potentially leading to overheating and decomposition of the material or surface wear, resulting in chipping from the bottle cap; when Excessive stress can lead to uneven cooling of the melt within the mold, generating internal stress that may cause the cap surface or internal material to detach and form debris due to stress concentration. Similarly, various production parameter scenarios can be simulated in a digital twin mapping system, and model parameters can be determined through multiple linear regression analysis based on the properties of the melt material. In some implementations, medical-grade polypropylene is used as an example of the melt material. =1.23, =0.089, =0.56.

[0052] In other implementations, to improve the accuracy of the prediction model, it can be fed into an artificial neural network for training, resulting in an optimized prediction model. It is understood that training an artificial neural network model requires a large amount of sample data, but constructing this data typically incurs significant human and time costs. To address this issue, transfer learning algorithms can be introduced, leveraging knowledge from relevant tasks to enhance the effectiveness and performance of model training.

[0053] Transfer learning techniques are employed to transfer knowledge between datasets of different components, utilizing source datasets from both software simulations and experimental tests. For example, to obtain data related to surface defects using simulations, it is necessary to establish a correlation between simulation output variables and experimental observations. Transfer learning can reduce the amount of data required to train parameter prediction models, and software simulation data can serve as a low-cost alternative to experimental data. The predicted metric values ​​obtained through transfer learning from simulations are comparable to those of untrained networks, but the amount of data required for the target dataset is significantly reduced.

[0054] It is understandable that the second-parameter prediction model constructed using transfer learning algorithms also includes at least a surface defect prediction model and a chip risk prediction model, which are determined by selecting different production parameters and their relationship with quality results. Specifically, such as Figure 4 As shown, the construction of the second parameter prediction model using the transfer learning algorithm mainly includes the following steps S1031 to S1033:

[0055] Step S1031: Use Moldex3D software to simulate the injection molding process of medical infusion bottle caps and create a source dataset. The source dataset includes an experimental dataset and a software simulation dataset. The software simulation dataset is generated through Latin hypercube sampling experiments.

[0056] In this step, the CAD model of the medical infusion bottle cap is imported into Moldex3D software. Meanwhile, in order to obtain simulation results within an acceptable time frame and further improve the accuracy of the results, some modifications can be made to each model. For example, some features in the medical infusion bottle cap that do not increase the mesh burden or extend the calculation time but do not improve the accuracy of the results can be removed.

[0057] Understandably, experimental datasets are training data from existing, well-dataned, or sufficiently learned tasks. These datasets consist of specific samples and related information collected through experiments, observations, and other methods. For example, 1000 effective injection molding records of medical-grade polypropylene injection-molded infusion bottle caps, including production parameters and defect markers.

[0058] For constructing the software simulation dataset, a mesh structure with four boundary layers can be selected for the medical infusion bottle cap to improve the simulation accuracy near the surface of the cap. Furthermore, in the simulation, considering that the Latin hypercube sampling experiment distributes sampling points almost randomly within the domain of each variable, it can minimize the overlap between experimental conditions, improve sampling efficiency, and reduce the correlation between multidimensional variables. Therefore, a Latin hypercube sampling experimental design is used to systematically change the production parameters. These parameters are selected from several key production parameters that the operator frequently adjusts during iterative machine setup, such as mold temperature, injection speed, and holding pressure. Simulation results are obtained for each key production parameter or its combination, representing the quality / performance indicators (labels) of the medical infusion bottle cap, such as the shrinkage rate. Finally, 100 data points are generated, each simulated individually, and each includes the production parameters and corresponding labels.

[0059] It should be noted that since simulation results do not directly provide experimental observations regarding, for example, whether "surface defects are visible" or "whether there is debris," a classification feature needs to be determined. In some implementations, the frozen layer ratio is used as a classification feature to assess the visibility of defects. Additionally, by conducting simulations with randomly varying production parameters, an effective FLR threshold is determined from the FLR values ​​obtained from these simulations and multiple molded medical infusion bottle caps. This threshold is used to distinguish between defective and non-defective cases. The FLR threshold is then compared with the FLR value associated with each simulation. If the sampled FLR exceeds the threshold, a surface defect is considered to exist, and the simulation result is classified as Class 1. Conversely, if the FLR value is equal to or below the threshold, the defect is considered invisible, and the simulation result is classified as Class 0. For example, the FLR threshold is 5%. This threshold is used to classify simulation results into Class 0 and Class 1.

[0060] Step S1032: Pre-train the second parameter prediction model based on the source dataset, and obtain the initial weights of the second parameter prediction model through backpropagation optimization. The second parameter prediction model is a CNN-LSTM neural network.

[0061] In predicting defects in medical infusion bottle caps, CNN convolutional layers can be used to extract local features through sliding windows. This allows for the automatic learning of implicit correlations between multiple production parameters (such as temperature-pressure coupling effects), significantly improving the accuracy of defect cause identification. Furthermore, LSTM neural networks, through gating mechanisms (forget gate, input gate, output gate), can effectively capture long-term temporal dependencies and remember parameter changes at key time points (such as pressure peaks at the end of filling and temperature drops during cooling). This allows for precise correlation between temporal trajectories and defects, which is crucial for predicting surface defects in infusion bottle caps (such as microcracks at the puncture site). Therefore, the second parameter prediction model employs a CNN-LSTM neural network, possessing both spatial correlation capture and temporal dynamic modeling capabilities. This perfectly adapts to the complexity of the medical infusion bottle cap injection molding process, efficiently utilizing limited medical data while accurately identifying subtle defects, providing more reliable predictive support for the quality control of medical infusion bottle caps.

[0062] In some implementations, to optimize the performance of artificial neural networks (ANNs), the hyperparameters of the ANN need to be determined before applying transfer learning. First, the structure of the ANN can be determined by generating loss function curves and accuracy curves, thereby selecting a network structure that minimizes the risk of overfitting or underfitting. Then, the remaining hyperparameters are determined through grid search. For example, a CNN-LSTM neural network can be an ANN with four hidden layers, each containing four neurons. It should be noted that early stopping callbacks can also be added to the ANN to further reduce the risk of overfitting. Afterward, the CNN-LSTM neural network is trained using the source dataset to obtain the initial weights through backpropagation optimization.

[0063] Step S1033: Perform secondary training on the target dataset, select a soft-start or random initialization strategy to fine-tune the initial weights, and determine the second parameter prediction model after transfer learning.

[0064] After obtaining the initial weights, the weights of the second parameter prediction model are fine-tuned to adapt the model to predict surface defects or debris risk in medical infusion bottle caps. The target dataset includes only experimental results data, specifically the production parameters used to perform the current task and their corresponding label data.

[0065] In some implementations, a soft-start (SS) or random initialization strategy can be chosen to fine-tune the weights of the second parameter prediction model. When the target data is limited and the source task is highly similar to the target task, a soft-start strategy is preferred. This avoids destroying the initial weights obtained from pre-training and allows for smooth transfer to the target task. However, when the target task differs significantly from the source task (e.g., different defect definitions), a random initialization strategy can be used. This can eliminate negative transfer and allow the model to learn new rules more flexibly.

[0066] It should be noted that after obtaining the second parameter prediction model, a control network can be trained using data related to the target dataset. This control network can then be used to validate the parameter prediction model and serve as a reference for evaluating the transfer learning effect. In some implementations, the training process can be repeated by gradually increasing the number of data points in the target dataset during knowledge transfer. In each iteration, three metrics are used to evaluate the network performance: accuracy, AUC, and recall. Five-fold cross-validation is performed to assess the reproducibility of the results, and then the standard deviation of each metric is calculated. Only when the accuracy, AUC, recall, and standard deviation are all superior to those of the aforementioned control network can it be used as a prediction model for parameter prediction in the injection molding process.

[0067] The introduction of transfer learning algorithms to train the second parameter prediction model in this embodiment offers the following advantages: transfer learning reduces the need for extensive data collection and model training for the target task, enhances the robustness and generalization ability of the machine learning model, reduces the risk of overfitting, and ensures reliable operation in different production scenarios. Utilizing transfer learning simplifies the training of artificial neural networks and can be used to predict surface defects and the risk of debris shedding on medical infusion bottle caps.

[0068] Step S104: Based on the first parameter prediction model or the second parameter prediction model, use the fuzzy inference algorithm to search from the set of experience cases to determine the optimal production parameters. The set of experience cases includes historical data and model inference data.

[0069] Typically, historical data can include production plans, equipment status, process parameters, quality inspection data, and anomaly handling. Anomaly handling includes deviations from normal operating conditions during production and corresponding countermeasures. For example, it might record frequent edge flash on products, analyze the causes, and explain how to handle it. Furthermore, other simulation models can be used to generate virtual data to supplement historical data. This includes simulations of non-occurring conditions, extreme scenarios, or potential optimization areas, thus obtaining model extrapolation data. Therefore, to improve the efficiency of the digital twin mapping system in accurately predicting production parameters, this step involves constructing an experience case set using historical data combined with model extrapolation data.

[0070] In some implementations, the creation of the experience case library involves creating a variety of possible improper forming conditions. After defining all features and parameters for each defect, the significant weight of each defect is calculated, and the experience case library is created based on all failures, their causal rules, attributes, and their weights.

[0071] In some implementations, potential defects can be predicted by analyzing the relationship between different production parameters during the injection molding process and the characteristics of the injection molding results. This relationship can be determined using fuzzy inference to classify the influence and relationship between production parameters and quality results. To determine the relationship between production parameters and quality results, a fuzzy inference algorithm can be used. This algorithm takes production parameters as input and generates a member function representing the degree of relationship between the production parameters and the quality results as output.

[0072] For example, after obtaining the predicted surface defect level and predicted chip risk index of the bottle cap through formula (1) and formula (2), the probability value of the bottle cap having surface defects or chip risk is calculated through formula (3), which is in the range of 0 to 1.

[0073] (3)

[0074] Where Y represents the probability value of surface defects or chipping risk, and Z represents the degree of surface defects D or the chipping risk index P.

[0075] In some implementations, the production parameters used in the surface defect prediction model and the chip risk prediction model can be classified, for example, into temperature, pressure, time, and speed. Furthermore, five fuzzy subsets are defined, such as very weak, weak, medium, strong, and very strong, to represent the degree of influence and relationship between the aforementioned production parameters and the quality results. This degree can be represented using a membership function.

[0076] Specifically, the membership function is:

[0077] (4)

[0078] Among them, membership degree Used to represent the degree of relationship between production parameters and quality results; a, b, and c are three key parameters belonging to the universe of discourse U. The specific values ​​of these three parameters will be determined based on the relationship between the parameter types (such as temperature, pressure, time, etc.) and fuzzy sets (such as very weak and weak) obtained from the simulation of the prediction model in the simulation and analysis layer. The universe of discourse U is the entire range of possible values ​​of characteristic parameters in the actual injection molding process (such as the actual measurement range of parameters such as temperature and pressure in the injection molding process); where a corresponds to the starting point (left boundary) where the membership degree starts to rise from 0, b corresponds to the vertex (center point) where the membership degree reaches the maximum value of 1, and c corresponds to the ending point (right boundary) where the membership degree drops from 1 to 0; x represents the value of the production parameter actually measured in the injection molding process. In some implementations, the determination of a, b, and c, in addition to combining the physical range of the actual process parameters (universe of discourse U), the critical values ​​in production experience (such as the threshold for defect generation), and the optimal working point (such as the parameter value with the highest pass rate), can also be determined by inputting the production parameter x into the parameter prediction model or the parameter prediction model, and obtaining the probability value according to formula (3).

[0079] Specifically, taking the surface defect prediction model as an example, assuming it is necessary to determine whether the relationship between melt temperature and the occurrence of surface defects is strong or moderate, firstly, in the surface defect prediction model, the other three values ​​besides melt temperature are set to normal values ​​in the normal processing process. Then, in the simulation and analysis layer, different scenarios are simulated to calculate the probability of surface defects occurring under simulation conditions. Then, according to the definitions of a, b, and c, the values ​​of a, b, and c in the membership function are determined. For example, when the melt temperature is less than or equal to 200℃, the material is not fully plasticized, and the probability of surface defects is close to 1; when the melt temperature is equal to 230℃, the product qualification rate is the highest, and the probability of surface defects is close to 0; when the melt temperature is greater than or equal to 250℃, the material degrades, producing scorch marks, and the probability of surface defects is also close to 1. Furthermore, simulation results show that when the temperature is between 200 and 230℃, the probability of surface defects decreases as the temperature increases (corresponding to a membership degree increasing from 0 to 1); when the temperature is between 230 and 250℃, the probability of surface defects increases as the temperature increases (corresponding to a membership degree decreasing from 1 to 0). Therefore, we can determine that a is 200℃, b is 230℃, and c is 250℃. If the monitored melt temperature is 220℃, then the membership function calculation yields 0.67, indicating that there is a certain probability of minor defects occurring at this temperature.

[0080] Then, the fuzzy membership is transformed into a clear value by defuzzification formula (5). By analogy, the clear value of each production parameter is calculated. Finally, the fuzzy weight of each production parameter is calculated by weighted average method, which is obtained by formula (6).

[0081] (5)

[0082] Where C() represents deblurring, It represents a fuzzy quantity, used to represent a less clear or ambiguous relationship (such as the relationship between melt temperature and the occurrence of surface defects).

[0083] (6)

[0084] Among them, W fi It is the fuzzy weight of the i-th feature, FO fi The weight of the i-th feature represents the probability that a large surface defect caused by abnormal melt temperature occurs in historical faults. It is determined based on production data. It is a clear value representing the relationship between the j-th parameter and the relevant quality result i, PR ji This refers to the correlation term between the j-th parameter and the relevant quality result i.

[0085] Following the above process, the production parameter with the highest weight in the risk of surface defects or chipping is determined as the feature parameter. This feature parameter is then used to search the experience case set to output the optimal production parameter based on solutions for partially or completely similar situations. If no similar situation is found in the experience case set, the predicted parameter output by either the first or second parameter prediction model is used as the optimal production parameter.

[0086] Step S104: Output the optimal production parameters to the medical infusion bottle cap injection molding system and update them in real time.

[0087] In this step, the determined optimal production parameters are fed back to the medical infusion bottle cap injection molding system. These parameters are adjusted promptly, and any predicted abnormal parameters are corrected in a timely manner to prevent the production of defective caps, thereby reducing production costs. It is understood that these optimal production parameters are updated in real time to allow for timely adjustments to the production process.

[0088] This disclosure presents an intelligent control method for the production of medical infusion bottle caps, based on a digital twin of the injection molding system. It aims to improve production efficiency and product quality by integrating AI-driven detection methods. This digital twin utilizes data from the injection molding process to predict and optimize product and production performance in real time. By employing a twin mapping system for simulation prediction, it can identify production parameters that need adjustment to improve efficiency and product quality, ensuring continuous optimization. Furthermore, this intelligent control method also references historical production data based on case-based reasoning and fuzzy weights, applying previous experience to adjust production parameters for the current task, further improving the accuracy of predicting and preventing production defects.

[0089] Based on the same inventive concept, this disclosure also provides an intelligent control device for the production of medical infusion bottle caps, such as... Figure 5 As shown, Figure 5 An intelligent control device 200 for the production of medical infusion bottle caps is shown in some embodiments of this disclosure. The control device includes: an acquisition module 201, used to acquire data from multiple sensors in the medical infusion bottle cap injection molding system in real time; a parameter prediction model generation module 202, used to construct a digital twin mapping model of the medical infusion bottle cap injection molding system, input the multiple sensor data into the digital twin mapping system to obtain dynamic simulation data, and generate a parameter prediction model based on the multiple sensor data and the dynamic simulation data; an optimal production parameter determination module 203, used to determine the optimal production parameters based on the parameter prediction model and using a fuzzy inference algorithm to search from an experience case set, wherein the experience case set includes historical data and model inference data; and an output and update module 204, used to output the optimal production parameters to the medical infusion bottle cap injection molding system and update them in real time.

[0090] The specific functions of each functional module in the above-mentioned device have been described in detail in the intelligent control method for the production of medical infusion bottle caps provided in some embodiments of this disclosure, and will not be elaborated here.

[0091] Based on the same inventive concept, this disclosure also provides an electronic device 300, which is manifested in the form of a general-purpose computing device. For example... Figure 6 As shown, the electronic device includes: a processor 301; and a memory 302 for storing executable instructions of the processor 301, wherein the processor 301 is configured to execute executable instructions to implement the intelligent control method for the production of medical infusion bottle caps as described in any of the above embodiments.

[0092] Furthermore, this electronic device can also communicate via a bus interface. For example... Figure 6 As shown, bus interface 303 provides an interface between the bus and receiver 304 and transmitter 305. In some embodiments, receiver 304 and transmitter 305 may also be the same element, i.e., a transceiver, to provide a unit for communicating with various other devices over a transmission medium. Processor 301 is responsible for managing the general processing of the bus, while memory 302 may be used to store data used by processor 301 during operation.

[0093] In exemplary embodiments of this disclosure, based on the same inventive concept, this disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the intelligent control method for the production of medical infusion bottle caps as described in any of the above embodiments. In some possible embodiments, various aspects of this disclosure can also be implemented in the form of a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps according to the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0094] The program product for implementing the above-described method according to embodiments of this disclosure can store program code in a portable compact disc read-only memory (CD-ROM) and can run on a terminal device, such as a personal computer. Of course, the program product of this disclosure is not limited thereto. In embodiments of this disclosure, the readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc., and the storage medium may also include combinations of the above types of memory.

[0095] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0096] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should 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 processor, 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 and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 Devices that specify the functions in one or more boxes.

[0097] Although preferred embodiments of this disclosure 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 the preferred embodiments as well as all changes and modifications falling within the scope of this disclosure.

[0098] Obviously, those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and practicing the technical content disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or effective technical means in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope of this disclosure is indicated by the claims.

[0099] The above description is merely an embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. It should be understood that various modifications, equivalent substitutions, and improvements can be made to this disclosure without departing from the scope of the above embodiments, and all such modifications, substitutions, and improvements should be included within the scope of the claims of this disclosure.

Claims

1. A control method for the production of medical infusion bottle caps, characterized in that, An injection molding system for medical infusion bottle caps, the system comprising multiple sensors, and the control method comprising: Real-time acquisition of data from the multiple sensors in the medical infusion bottle cap injection molding system; A digital twin mapping system for the medical infusion bottle cap injection molding system is constructed, wherein the digital twin mapping system includes a virtual mapping layer, a data layer, a model layer, and a simulation and analysis layer; The data from the multiple sensors are input into the digital twin mapping system to obtain dynamic simulation data. Based on the data from the multiple sensors and the dynamic simulation data, a first parameter prediction model is constructed using a data-driven modeling algorithm; or a second parameter prediction model is constructed using a transfer learning algorithm. Based on the first parameter prediction model or the second parameter prediction model, the optimal production parameters are determined by searching from the set of experience cases using a fuzzy inference algorithm. The set of experience cases includes at least historical data and model inference data. The optimal production parameters are output to the medical infusion bottle cap injection molding system and updated in real time.

2. The control method according to claim 1, characterized in that, The digital twin mapping system for constructing the medical infusion bottle cap injection molding system includes: Based on the geometric and structural, physical properties, behavior and state information of the medical infusion bottle cap injection molding system, a virtual mapping layer corresponding to the medical infusion bottle cap injection molding system is constructed. The data layer is constructed, wherein the data layer includes at least one of the following: real-time acquired data from the multiple sensors of the medical infusion bottle cap injection molding system, historical data, and environmental data; Obtain at least one type of data from the data layer, fuse the geometry and structure, the physical properties, the behavior and state information, and rule knowledge to construct the model layer; The simulation and analysis layer is constructed to synchronously simulate the current state of the medical infusion bottle cap injection molding system and to dynamically simulate the state, behavior, or future trend of the medical infusion bottle cap injection molding system.

3. The control method according to claim 1, characterized in that, The first parameter prediction model includes at least a surface defect prediction model and a chip risk prediction model.

4. The control method according to claim 3, characterized in that, The surface defect prediction model is used to predict the severity of surface defects in the medical infusion bottle cap, and the severity is calculated using the following formula: ； Where D represents the severity of the surface defect, The holding pressure fluctuation value is T1, the melt temperature is T2, the mold temperature is t1, the cooling time is V, the injection speed is V, and MFR represents the melt flow rate. This serves as a benchmark value for pressure fluctuation during pressure holding. This serves as a reference value for the rate of temperature change. This serves as a baseline value for injection speed. k0 represents the baseline value for the solution flow rate; k0 to k4 are model parameters.

5. The control method according to claim 3, characterized in that, The debris risk prediction model is used to predict the debris risk index of the medical infusion bottle cap during demolding or puncture, wherein the debris risk index is calculated by the following formula: Where F represents the chipping risk index, P 熔体 t1 represents the melt pressure, and t2 represents the injection molding time. V represents the rate of change of melt pressure over time. 剪切速率 Indicates the melt shear rate; This refers to the temperature difference of the melt within the mold; This is the baseline value for the rate of change of melt pressure over time; This is the baseline value for the melt shear rate; This serves as a baseline value for the temperature difference of the melt within the mold. These are the model parameters.

6. The control method according to claim 1, characterized in that, The method of constructing the second parameter prediction model using the transfer learning algorithm includes: The injection molding process of the medical infusion bottle cap was simulated using Moldex3D software to create a source dataset, which includes an experimental dataset and a software simulation dataset. The software simulation dataset was generated through Latin hypercube sampling experiments. The second parameter prediction model is trained based on the source dataset, and the initial weights of the second parameter prediction model are obtained through backpropagation optimization, wherein the second parameter prediction model is a CNN-LSTM neural network. Secondary training is performed on the target dataset, and the initial weights are fine-tuned by selecting a soft-start or random initialization strategy to determine the second parameter prediction model after transfer learning.

7. The control method according to claim 1, characterized in that, Based on the first parameter prediction model or the second parameter prediction model, the optimal production parameters are determined from the set of empirical cases using a fuzzy inference algorithm, including: The degree of relationship between each production parameter and the quality result is determined by the membership function, and the production parameter with the highest weight in the risk of surface defects or chipping is determined as the feature parameter by the defuzzification and weighted average method. The optimal production parameters are determined by searching the set of empirical cases using the feature parameters, based on the production parameters corresponding to cases where partial or complete similarity occurs.

8. The control method according to claim 7, characterized in that, The membership function is: ； Membership degree Used to represent the degree of relationship between the production parameters and the quality results; a corresponds to the starting point where the membership degree begins to rise from 0, b corresponds to the peak where the membership degree reaches the maximum value of 1, and c corresponds to the end point where the membership degree falls from 1 to 0. x represents the value of the production parameter measured during the injection molding process of the medical infusion bottle cap.

9. An electronic device, characterized in that, The device includes a processor and a memory for storing processor-executable instructions, wherein the processor is configured to execute the executable instructions to implement the control method for producing medical infusion bottle caps as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the control method for the production of medical infusion bottle caps as described in any one of claims 1-8.

Images