Glass container pressing and blowing control method and system

Abstract

The invention provides a pressure-blowing control method and system for a glass container, which are applied to the technical field of glass container manufacturing, and are used for estimating a glass forming state by collecting local pressure data of a mold wall in real time, and dynamically adjusting blowing pressure according to state deviation. The real-time control of the local flow spreading behavior of the glass, the real-time perception of the glass forming process and the dynamic control of the glass forming process are realized, and the method has the beneficial effects of improving the wall thickness uniformity of the glass container and improving the stress distribution of the glass container.

Description

Technical Field

[0001] This application relates to the technical field, and in particular to a method and system for controlling the pressure blowing of glass containers. Background Technology

[0002] Traditional press-blown glass manufacturing is primarily designed for relatively simple glass containers, such as standard cylindrical or conical flasks. For these products, the regular geometry makes the flow and expansion behavior of the glass within the mold relatively predictable. Therefore, pre-set, uniform blowing pressure profiles and overall mold temperature control can typically achieve good wall thickness uniformity and reasonable stress distribution. However, with the increasing market demand for diverse appearances and functions in glass containers, complex-shaped glass containers are becoming more prevalent. These complex shapes may include those with prominent recesses, protrusions, spirals, one-piece molded handles, intricate embossed patterns, or asymmetrical contours. Manufacturing these complex-shaped products presents significant challenges to traditional press-blown glass control methods.

[0003] Especially on modern glass manufacturing production lines that pursue high-speed and high-efficiency production, the forming cycle is extremely short (maybe only a few seconds). Any tiny parameter deviation or imperceptible state change can be amplified in a short time, leading to defects such as uneven wall thickness, stress concentration, or even cracking. For example, if the initial distribution of the droplet is biased towards one side of a complex mold shape (such as a recess or convex area), it may result in too little or too much preformed glass in that area, which in turn leads to abnormal expansion rate of the glass in that area during blowing, ultimately forming a container with uneven wall thickness. The interaction between the slight temperature differences in different areas of the mold and the glass expansion rate may also lead to localized excessively fast or slow cooling, generating stress.

[0004] Therefore, in the press blow molding process of complex-shaped glass containers, facing the differences in the initial state of the molten droplets and the resulting uncertainty in the dynamic rheological behavior of the glass in the mold, how can we achieve rapid adaptive control of the glass forming process within a very short molding cycle through real-time, localized process state perception to ensure the uniformity of the final product wall thickness and the stability of stress distribution? Summary of the Invention

[0005] In view of the shortcomings of the prior art, this application provides a glass container pressure blowing control method and system, which has the advantages of improving the uniformity of glass container wall thickness and improving the stress distribution of glass container.

[0006] In a first aspect, a method for controlling the pressure blowing of a glass container, the method comprising the steps of:

[0007] S1: Real-time acquisition of local pressure data on the mold wall during the glass blowing and expansion process to obtain pressure array data;

[0008] S2: Filter and calibrate the pressure array data to obtain processed pressure data;

[0009] S3: Input the processed pressure data into the forming state estimation model to estimate the forming state parameters of the glass in the mold;

[0010] S4: Calculate the deviation between the forming state parameters and the target state, and dynamically adjust the subsequent stage parameters of the blowing pressure curve according to the deviation to obtain the adjusted blowing pressure parameters.

[0011] S5: The adjusted blowing pressure parameters are sent to the blowing system to control the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow behavior of the glass.

[0012] Furthermore, step S1 includes:

[0013] S11: Divide the critical area on the cavity wall of the blow molding die, and arrange multiple thin film pressure sensors in the critical area in an array, with each thin film pressure sensor integrating a temperature compensation circuit.

[0014] S12: Synchronously acquire pressure and temperature data from all thin-film pressure sensors to obtain pressure array data containing pressure and temperature information;

[0015] S13: Based on the pre-calibrated pressure sensor temperature-pressure calibration curve, the pressure array data is calibrated in real time using the collected temperature data to eliminate temperature drift error and obtain the calibrated pressure array data.

[0016] Furthermore, step S13 includes:

[0017] S131: Construct a standard temperature field containing multiple temperature gradient levels, and record the pressure output value of each pressure sensor under each temperature gradient level;

[0018] S132: For each pressure sensor, fit the temperature-pressure calibration curve of that pressure sensor based on the pressure output value;

[0019] S133: Calculate the pressure correction value at this temperature based on the temperature data and the temperature-pressure calibration curve;

[0020] S134: Subtract the pressure correction value from the pressure array data to eliminate temperature drift error and obtain the calibrated pressure array data.

[0021] Furthermore, step S2 includes:

[0022] S21: The pressure array data is initially filtered using a moving average filtering algorithm to obtain the pre-filtered pressure array data;

[0023] S22: Construct a multi-scale decomposition model based on wavelet transform to decompose the pre-filtered pressure array data into components with multiple different frequency scales;

[0024] S23: For components at different frequency scales, an adaptive threshold filtering method is used for refined filtering;

[0025] S24: Perform inverse wavelet transform on each frequency component after fine filtering to reconstruct the filtered pressure array data;

[0026] S25: Establish a database of static characteristic parameters of pressure sensors, which includes the sensitivity and zero-point drift parameters of each pressure sensor;

[0027] S26: Based on the pressure sensor static characteristic parameter database, the filtered pressure array data is calibrated using the least squares method to obtain the processed pressure array data.

[0028] Furthermore, step S23 includes:

[0029] S231: Calculate the energy value of each frequency component within a preset time window and construct an energy distribution map;

[0030] S232: Determine the energy threshold of each frequency component based on the energy distribution diagram;

[0031] S233: Compare the amplitude of each frequency component with the corresponding energy threshold. If the amplitude is less than the energy threshold, it is determined to be noise, and the amplitude is set to zero. If the amplitude is greater than or equal to the energy threshold, it is determined to be a valid signal, and the amplitude is retained.

[0032] S234: Integrate the frequency components after amplitude adjustment to obtain the finely filtered pressure array data.

[0033] Furthermore, step S3 includes:

[0034] S31: Construct a forming state estimation model containing a radial basis function neural network. The input of the neural network is the processed pressure array data, and the output is the contact area between the glass and the mold, the local expansion rate, and the wall thickness unevenness trend.

[0035] S32: Train a radial basis function neural network using finite element simulation data and actual production data to obtain the trained forming state estimation model;

[0036] S33: Input the processed pressure array data into the forming state estimation model to estimate the forming state parameters of the glass in the mold. The forming state parameters include the contact area between the glass and the mold, the local expansion rate, and the wall thickness unevenness trend.

[0037] Furthermore, step S32 includes:

[0038] S321: Construct a finite element model, which includes the initial temperature, shape and position deviation of the molten droplet, to simulate the evolution of the temperature field, stress field and deformation field of the glass during the pressure blow molding process, and to obtain the pressure array data of the mold cavity wall and the corresponding glass container wall thickness distribution data under different initial states.

[0039] S322: Normalize the finite element simulation data and select a subset of samples from the normalized finite element simulation data to construct the initial training dataset;

[0040] S323: Collect pressure array data and wall thickness distribution data of different batches of glass containers during the actual production process as actual production data, and fuse the actual production data with the initial training dataset to obtain a hybrid training dataset;

[0041] S324: Divide the hybrid training dataset into K subsets, select K-1 subsets as the training set each time, and use the remaining 1 subset as the validation set to train the radial basis function neural network. Adjust the expansion speed and the number of hidden layer neurons of the neural network according to the root mean square error on the validation set to obtain the neural network model.

[0042] S325: Train the neural network model and monitor the root mean square error on the validation set. When the root mean square error on the validation set does not decrease within N consecutive epochs, stop training, restore the weights and thresholds of the neural network to the optimal state, and obtain the forming state estimation model.

[0043] Furthermore, step S4 includes:

[0044] S41: Establish a mapping relationship model, which takes the forming state parameters as input and the subsequent stage parameters of the blowing pressure curve as output;

[0045] S42: Determine the target state parameters, which include the target contact area, the target local expansion rate, and the target wall thickness uniformity.

[0046] S43: Calculate the deviation between the forming state parameters and the target state parameters to obtain the deviation vector;

[0047] S44: Input the deviation vector into the mapping relationship model, and calculate the adjustment amount of subsequent stage parameters of the blowing pressure curve through the mapping relationship model. The adjustment amount includes the blowing pressure amplitude, blowing time and blowing rate.

[0048] S45: The adjustment amount of the subsequent stage parameters of the blowing pressure curve is superimposed with the current blowing pressure curve parameters to obtain the adjusted blowing pressure parameters.

[0049] Furthermore, step S41 includes:

[0050] S411: Construct a BP neural network, which includes an input layer, a hidden layer, and an output layer. The nodes in the input layer correspond to the forming state parameters, and the nodes in the output layer correspond to the subsequent stage parameters of the blowing pressure curve.

[0051] S412: Train a BP neural network using offline experimental or simulation data, measure the forming state parameters by changing the subsequent stage parameters of the blowing pressure curve, minimize the mean square error between the network output and the target output, and obtain the mapping relationship model.

[0052] Secondly, a glass container pressure blowing control system, applied in the steps of any of the above-described methods, the system comprising:

[0053] Acquisition module: Real-time acquisition of local pressure data on the mold wall during the glass blowing and expansion process to obtain pressure array data;

[0054] Processing module: Filters and calibrates the pressure array data to obtain processed pressure data;

[0055] Estimation module: Inputs the processed pressure data into the forming state estimation model to estimate the forming state parameters of the glass in the mold;

[0056] Calculation module: Calculates the deviation between the forming state parameters and the target state, and dynamically adjusts the subsequent stage parameters of the blowing pressure curve according to the deviation to obtain the adjusted blowing pressure parameters;

[0057] Control module: Sends the adjusted blowing pressure parameters to the blowing system to control the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow behavior of the glass.

[0058] Beneficial effects: The glass container pressure blowing control method and system proposed in this application estimates the glass forming state by real-time acquisition of local pressure data of the mold wall, and dynamically adjusts the blowing pressure according to the state deviation. This realizes real-time control of local glass flow behavior, real-time perception of the glass forming process, and dynamic control of the glass forming process, which has the beneficial effects of improving the uniformity of glass container wall thickness and improving the stress distribution of glass container. Attached Figure Description

[0059] Figure 1 This is a flowchart of a glass container pressure blowing control method proposed in this application.

[0060] Figure 2 This is a structural diagram of a glass container pressure blowing control system proposed in this application.

[0061] Figure 3 This is a frame diagram of a glass container pressure blowing control system proposed in this application.

[0062] Labeling Explanation: 201, Acquisition Module; 202, Processing Module; 203, Estimation Module; 204, Calculation Module; 205, Control Module. Detailed Implementation

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and marked in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0064] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0065] Please refer to Figure 1 A method for controlling the pressure blowing of a glass container, the method comprising the following steps:

[0066] S1: Real-time acquisition of local pressure data on the mold wall during the glass blowing and expansion process to obtain pressure array data;

[0067] S2: Filter and calibrate the pressure array data to obtain processed pressure data;

[0068] S3: Input the processed pressure data into the forming state estimation model to estimate the forming state parameters of the glass in the mold;

[0069] S4: Calculate the deviation between the forming state parameters and the target state, and dynamically adjust the subsequent stage parameters of the blowing pressure curve according to the deviation to obtain the adjusted blowing pressure parameters.

[0070] S5: The adjusted blowing pressure parameters are sent to the blowing system to control the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow behavior of the glass.

[0071] Specifically, during the pressure blow molding process of glass containers, the dynamic rheological behavior of glass is uncertain, which may lead to defects such as uneven wall thickness or stress concentration in the product. This method addresses this issue by constructing a real-time feedback control loop.

[0072] First, in step S1, local pressure data of the mold wall is collected in real time to obtain pressure distribution information in the area in contact with the mold during the glass expansion process. This pressure data directly reflects the local stress and expansion state of the glass within the mold, and is the basis for sensing the process status.

[0073] Next, in step S2, the acquired pressure data is filtered and calibrated to remove noise interference and correct sensor errors, ensuring that the data used in subsequent processing is accurate and reliable. Accurate data is a prerequisite for effective state estimation and control.

[0074] Then, in step S3, the processed pressure data is input into the forming state estimation model. This model uses the pressure data to infer the actual forming state of the glass within the mold, such as the contact area between the glass and the mold, the local expansion rate, and the wall thickness variation trend. This step transforms the raw sensor data into state parameters meaningful to the glass forming process, enabling an understanding of the internal process state.

[0075] Subsequently, in step S4, the deviation between the estimated forming state parameters and the preset target state is calculated, quantifying the gap between the current forming state and the ideal state. Based on this deviation, the required adjustment amount for subsequent stages of the blowing pressure curve is dynamically calculated, resulting in new blowing pressure parameters. This step is the core of the control decision, determining how to change the control input based on the difference between the actual state and the target.

[0076] Finally, in step S5, the adjusted blowing pressure parameters are sent to the blowing system. Based on the received parameters, the blowing system adjusts the pressure of the compressed gas entering the mold cavity in real time. The change in blowing pressure directly affects the glass, influencing its local expansion rate and flow direction within the mold, thereby correcting the glass forming process and bringing it closer to the target state.

[0077] Through the above steps, the method forms a closed-loop control system, including sensing, state understanding, decision adjustment and execution control. This real-time and dynamic adjustment capability enables the method to cope with the uncertainties caused by factors such as the initial state of the glass droplet and temperature distribution, thereby improving the molding quality and stability of complex-shaped glass containers.

[0078] Furthermore, step S1 includes:

[0079] S11: Divide the critical area on the cavity wall of the blow molding die, and arrange multiple thin film pressure sensors in an array in the critical area. Each thin film pressure sensor integrates a temperature compensation circuit.

[0080] S12: Synchronously acquire pressure and temperature data from all thin-film pressure sensors to obtain pressure array data containing pressure and temperature information;

[0081] S13: Based on the pre-calibrated temperature-pressure calibration curve of the pressure sensor, the pressure array data is calibrated in real time using the collected temperature data to eliminate temperature drift error and obtain the calibrated pressure array data.

[0082] Specifically, regions on the cavity wall of the blow molding die that significantly affect the forming state, such as areas with complex shapes, are selected as critical regions. Within these critical regions, multiple thin-film pressure sensors are installed at predetermined intervals and in a predetermined arrangement, such as a rectangular array. Each thin-film pressure sensor integrates a temperature compensation circuit to initially mitigate the impact of temperature changes on the sensor output.

[0083] During the blow molding process, the data acquisition system simultaneously reads the pressure signal output by each film pressure sensor and the temperature signal output by its integrated temperature sensor. These synchronously acquired data constitute pressure array data and temperature array data containing time and spatial information.

[0084] Before a sensor is put into use, it needs to be pre-calibrated. The sensor is placed in a standard temperature field at different temperatures, and a known standard pressure is applied at each temperature, recording the pressure value output by the sensor. Based on this calibration data, a temperature-pressure calibration curve is established for each sensor. This curve describes the deviation between the sensor's output pressure and the actual pressure at different temperatures. During actual production, the pressure correction value on the corresponding calibration curve at that temperature is consulted or calculated using real-time acquired temperature data.

[0085] The pressure correction value is subtracted from the synchronously acquired raw pressure data to eliminate measurement errors caused by temperature drift. The calibrated pressure array data more accurately reflects the actual local pressure of the glass on the mold wall.

[0086] This ensures that the pressure data collected in high-temperature environments has higher reliability.

[0087] Furthermore, step S13 includes:

[0088] S131: Construct a standard temperature field containing multiple temperature gradient levels, and record the pressure output value of each pressure sensor under each temperature gradient level;

[0089] S132: For each pressure sensor, fit the temperature-pressure calibration curve of that pressure sensor based on the pressure output value;

[0090] S133: Calculate the pressure correction value at this temperature based on the temperature data and the temperature-pressure calibration curve;

[0091] S134: The pressure correction value is subtracted from the pressure array data to eliminate temperature drift error and obtain the calibrated pressure array data.

[0092] Specifically, this technical solution addresses the problem of mold temperature changes affecting the measurement accuracy of pressure sensors during the glass container blow molding process.

[0093] By exposing pressure sensors to a series of known, stable temperature points (temperature gradient levels) in a controlled environment (e.g., a temperature calibration chamber) and recording the pressure output value of each sensor at each temperature point, performance data of the sensors at different temperatures can be obtained. For each sensor in the array, this data is used to fit a specific temperature-pressure calibration curve, which describes the pressure drift characteristics of that sensor at different temperatures. In actual pressure blow molding production, temperature data of each pressure sensor is acquired in real time.

[0094] Using the sensor's temperature-pressure calibration curve, the pressure measurement error caused by temperature drift, i.e., the pressure correction value, is calculated based on the current temperature. Subtracting this correction value from the sensor's real-time measured pressure value yields the true pressure value, unaffected by temperature. Performing this operation on all sensors in the array provides accurately temperature-calibrated pressure array data.

[0095] This ensures that the pressure data upon which subsequent forming state estimation and control depend is highly accurate, thereby improving the reliability of the control system.

[0096] Furthermore, step S2 includes:

[0097] S21: The pressure array data is initially filtered using a moving average filtering algorithm to obtain the pre-filtered pressure array data;

[0098] S22: Construct a multi-scale decomposition model based on wavelet transform to decompose the pre-filtered pressure array data into components with multiple different frequency scales;

[0099] S23: For components at different frequency scales, an adaptive threshold filtering method is used for refined filtering;

[0100] S24: Perform inverse wavelet transform on each frequency component after fine filtering to reconstruct the filtered pressure array data;

[0101] S25: Establish a database of static characteristic parameters of pressure sensors, which includes the sensitivity and zero-point drift parameters of each pressure sensor;

[0102] S26: Based on the pressure sensor static characteristic parameter database, the least squares method is used to calibrate the filtered pressure array data to obtain the processed pressure array data.

[0103] The initial filtering process employs a moving average algorithm, smoothing the raw pressure data by calculating the average value of data points within a preset time window to reduce the impact of high-frequency random noise. The refined filtering is based on multi-scale decomposition and adaptive thresholding using wavelet transform. The data is decomposed into different frequency components, and a filtering threshold is dynamically determined for each component based on its characteristics. Components below the threshold are suppressed, while those above are retained, followed by inverse wavelet transform reconstruction. The calibration process utilizes a pre-established database of sensor static characteristic parameters, which stores inherent error information such as sensitivity and zero-point drift for each sensor. The least squares method is used to correct the filtered data based on these parameters, eliminating individual sensor differences and systematic biases.

[0104] Furthermore, step S23 includes:

[0105] S231: Calculate the energy value of each frequency component within a preset time window and construct an energy distribution map;

[0106] S232: Determine the energy threshold for each frequency component based on the energy distribution diagram;

[0107] S233: Compare the amplitude of each frequency component with the corresponding energy threshold. If the amplitude is less than the energy threshold, it is determined to be noise and the amplitude is set to zero. If the amplitude is greater than or equal to the energy threshold, it is determined to be a valid signal and the amplitude is retained.

[0108] S234: Integrate the frequency components after amplitude adjustment to obtain the finely filtered pressure array data.

[0109] Specifically, after wavelet decomposition of the initially filtered pressure array data, a series of components at different frequency scales are obtained. To refine the filtering of these components, the energy value of each frequency component is first calculated within a preset time window. The energy value reflects the signal strength at that frequency scale and time period. By calculating the energy values ​​for multiple time windows, an energy distribution map can be constructed, visually displaying the energy changes of different frequency components over time. Based on this energy distribution map, an adaptive energy threshold can be determined for each frequency component or each time window. This threshold is determined based on the signal's own energy characteristics and can distinguish between low-energy noise components and high-energy effective signal components.

[0110] Subsequently, the instantaneous amplitude of each frequency component is compared with the corresponding energy threshold. If the amplitude is below the threshold, it is considered that the amplitude is mainly caused by noise and is set to zero to suppress noise. If the amplitude is greater than or equal to the threshold, it is considered that the amplitude represents a valid signal and is retained. After amplitude adjustment, all frequency components are reconstructed through inverse wavelet transform to obtain the refined filtered pressure array data after noise removal.

[0111] Therefore, this method can adaptively adjust the filtering intensity according to the energy characteristics of the signal itself, which improves the accuracy of filtering and provides a cleaner data basis for subsequent pressure data calibration and state estimation.

[0112] Furthermore, step S3 includes:

[0113] S31: Construct a forming state estimation model containing a radial basis function neural network. The input of the neural network is the processed pressure array data, and the output is the contact area between the glass and the mold, the local expansion rate, and the wall thickness unevenness trend.

[0114] S32: Use finite element simulation data and actual production data to train a radial basis function neural network to obtain a trained forming state estimation model;

[0115] S33: Input the processed pressure array data into the forming state estimation model to estimate the forming state parameters of the glass in the mold. The forming state parameters include the contact area between the glass and the mold, the local expansion rate, and the wall thickness unevenness trend.

[0116] Specifically, in the process of press blow molding of complex-shaped glass containers, the initial state differences of molten droplets and the resulting uncertainties in the dynamic rheological behavior of the glass within the mold make it difficult to accurately and in detail perceive the forming state. This solution addresses this problem by constructing a forming state estimation model based on a radial basis function neural network.

[0117] First, local pressure data of the mold wall is acquired in real time, reflecting the interaction force between the glass and the mold. The acquired raw pressure data is filtered and calibrated to remove noise and errors, resulting in processed pressure data. This processed pressure data is then used as input to a radial basis function neural network. This pre-trained neural network is capable of establishing a nonlinear mapping relationship between the pressure distribution on the mold wall and the forming state parameters inside the glass.

[0118] The training data comes from two sources: one is finite element simulation data, which obtains a large amount of pressure data and corresponding glass forming state data by simulating the blown glass process under different initial conditions; the other is actual production data, which collects pressure data and corresponding product inspection data from the actual production process. Training with a combination of simulation and actual data improves the model's generalization ability and accuracy in practical applications.

[0119] A well-trained neural network model can output estimated glass forming state parameters in real time based on the processed input pressure data, including the contact area between the glass and the mold, the rate of local glass expansion, and the trend of glass wall thickness variation. These parameters provide detailed information about the specific forming process of the glass within the mold, such as which areas have contacted the mold, which areas expand faster or slower, and which areas may have thinner or thicker walls.

[0120] This enables accurate, real-time estimation of everything from easily measurable mold wall pressure data to difficult-to-measure internal glass forming state parameters, laying the foundation for subsequent adaptive control.

[0121] Furthermore, step S32 includes:

[0122] S321: Construct a finite element model, which includes the initial temperature, shape and position deviation of the molten droplet, to simulate the evolution of the temperature field, stress field and deformation field of the glass during the pressure blow molding process, and obtain the pressure array data of the mold cavity wall under different initial states and the corresponding glass container wall thickness distribution data.

[0123] S322: Normalize the finite element simulation data and select a subset of samples from the normalized finite element simulation data to construct the initial training dataset;

[0124] S323: Collect pressure array data and wall thickness distribution data of different batches of glass containers during actual production process as actual production data, and fuse the actual production data with the initial training dataset to obtain a hybrid training dataset;

[0125] S324: Divide the mixed training dataset into K subsets. Each time, select K-1 subsets as the training set and the remaining 1 subset as the validation set. Train the radial basis function neural network. Adjust the expansion speed and the number of hidden layer neurons of the neural network according to the root mean square error on the validation set to obtain the neural network model.

[0126] S325: Train the neural network model and monitor the root mean square error on the validation set. When the root mean square error on the validation set does not decrease within N consecutive epochs, stop training, restore the weights and thresholds of the neural network to the optimal state, and obtain the formed state estimation model.

[0127] The finite element model is constructed to simulate the physical behavior of glass during the blow molding process, including the evolution of the temperature, stress, and deformation fields over time. The model input includes potential deviations in the initial temperature, shape, and position of the molten droplet, reflecting uncertainties in actual production. Through simulation, data on the pressure distribution of the mold cavity wall under different initial conditions and the final wall thickness distribution of the glass container can be obtained. This data forms the basis for training the model.

[0128] The finite element simulation data is normalized to eliminate the influence of data with different dimensions on model training. A subset of samples is selected from the normalized data to construct an initial training dataset for preliminary training of the neural network. Actual production data is collected, including pressure array data measured during actual production and wall thickness distribution data of the corresponding glass containers.

[0129] By fusing real-world production data with simulation data, a hybrid training dataset is created. This hybrid dataset combines the diversity of simulation data with the realism of real-world data, thus improving the model's generalization ability.

[0130] A radial basis function neural network (RBF) is trained using a K-fold cross-validation method. The mixed dataset is divided into K parts, with K-1 parts used for training and 1 part for validation in rotation. The structural parameters of the neural network, such as expansion rate and the number of hidden layer neurons, are adjusted based on the root mean square error (RMSE) on the validation set to optimize model performance. During training, the RMSE on the validation set is continuously monitored. Training is stopped when the error does not decrease within N consecutive training epochs to avoid overfitting. The weights and thresholds of the neural network are then restored to the state where the validation set error is lowest, resulting in the final shaped state estimation model.

[0131] Furthermore, step S4 includes:

[0132] S41: Establish a mapping relationship model. The mapping relationship model takes the forming state parameters as input and the subsequent stage parameters of the blowing pressure curve as output.

[0133] S42: Determine the target state parameters, which include the target contact area, the target local expansion rate, and the target wall thickness uniformity.

[0134] S43: Calculate the deviation between the forming state parameters and the target state parameters to obtain the deviation vector;

[0135] S44: Input the deviation vector into the mapping relationship model, and calculate the adjustment amount of subsequent stage parameters of the blowing pressure curve through the mapping relationship model. The adjustment amount includes the blowing pressure amplitude, blowing time and blowing rate.

[0136] S45: The adjustment amount of the subsequent stage parameters of the blowing pressure curve is superimposed with the current blowing pressure curve parameters to obtain the adjusted blowing pressure parameters.

[0137] Specifically, in the face of the differences in the initial state of molten droplets during the pressure blow molding of complex-shaped glass containers and the resulting uncertainty in the dynamic rheological behavior of the glass, this method senses the forming state of the glass in real time and compares it with the preset target state.

[0138] When a deviation is detected between the current forming state (e.g., the contact area between the glass and the mold, local expansion rate, or uneven wall thickness) and the target state, this deviation is calculated as a vector. This deviation vector is then input into a mapping model. This mapping model is trained to output the necessary adjustments to parameters of subsequent stages of the blowing pressure curve (such as the magnitude, duration, or rate of change of the blowing pressure) based on the input forming state deviation. For example, if the estimation model indicates that the glass expansion rate is too slow in a certain region, the mapping model might calculate an adjustment to increase the pressure of the corresponding blowing stage in that region.

[0139] These calculated adjustments are superimposed on the currently applied blowing pressure parameters to generate new, adjusted blowing pressure parameters. These adjusted parameters are sent to the blowing system to change the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow and expansion of the glass within the mold, correcting forming deviations, promoting the glass to develop towards the target state, and ultimately improving the uniformity of product wall thickness and the stability of stress distribution.

[0140] Furthermore, step S41 includes:

[0141] S411: Construct a BP neural network, which includes an input layer, a hidden layer, and an output layer. The nodes in the input layer correspond to the forming state parameters, and the nodes in the output layer correspond to the subsequent stage parameters of the blowing pressure curve.

[0142] S412: Train a BP neural network using offline experimental or simulation data, measure the forming state parameters by changing the subsequent stage parameters of the blowing pressure curve, minimize the mean square error between the network output and the target output, and obtain the mapping relationship model.

[0143] Specifically, in order to solve the problem of unclear mapping relationship between the deviation between the forming state and the target state and the amount of air blowing pressure adjustment, a BP neural network is used to establish the mapping model.

[0144] First, a backpropagation (BP) neural network is constructed. The number of nodes in the input layer corresponds to the number of forming state parameters (such as contact area, local expansion rate, and wall thickness unevenness trend), and the number of nodes in the output layer corresponds to the number of parameters in the subsequent stages of the blowing pressure curve (such as pressure amplitude, blowing time, and blowing rate). The number of nodes in the hidden layer can be set according to the actual situation.

[0145] Then, training data is acquired through offline experiments or simulations. In these experiments or simulations, the parameters of subsequent stages of the blowing pressure curve are systematically changed, and the resulting glass forming state parameters are recorded.

[0146] The collected forming state parameters are used as input samples for the BP network, and the corresponding subsequent stage parameters of the blowing pressure are used as the expected output samples for the network. These input-output samples are then used to train the BP neural network.

[0147] During training, the weights and thresholds of the network are adjusted using the backpropagation algorithm to minimize the mean square error between the network's output on the input sample and the expected output. After sufficient training, this BP neural network can learn the nonlinear mapping relationship from the glass forming state parameters to the subsequent stage parameter adjustments for the required blowing pressure.

[0148] Therefore, once the forming state parameters of the glass are collected in real time and the deviation from the target state is calculated, the deviation (or the forming state parameters directly) can be input into the trained BP neural network. The network will output the corresponding adjustment amount of the subsequent stage parameters of the blowing pressure curve, thereby realizing the dynamic adjustment of the blowing pressure.

[0149] Please refer to Figure 2 , Figure 3 A glass container pressure blowing control system, applied in the steps of any of the above methods, the system comprising:

[0150] Acquisition module 201: Real-time acquisition of local pressure data of the mold wall during the glass blowing expansion process to obtain pressure array data;

[0151] Processing module 202: Filters and calibrates the pressure array data to obtain processed pressure data;

[0152] Estimation module 203: Inputs the processed pressure data into the forming state estimation model to estimate the forming state parameters of the glass in the mold;

[0153] Calculation module 204: Calculates the deviation between the forming state parameters and the target state, and dynamically adjusts the subsequent stage parameters of the blowing pressure curve according to the deviation to obtain the adjusted blowing pressure parameters.

[0154] Control module 205: Sends the adjusted blowing pressure parameters to the blowing system to control the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow behavior of the glass.

[0155] The acquisition module 201 is responsible for acquiring the local pressure information generated between the glass and the mold wall when the glass expands inside the mold. This information is output in the form of pressure array data, reflecting the contact distribution and stress state between the glass and the mold.

[0156] The processing module 202 receives the raw pressure array data output by the acquisition module and performs necessary preprocessing, including removing noise introduced during the acquisition process and correcting the sensor's own errors, to ensure the accuracy and reliability of the data and provide high-quality input for subsequent state estimation.

[0157] The estimation module 203 uses the processed pressure data and a pre-established forming state estimation model to infer the actual forming state of the glass in the mold, such as the contact area between the glass and the mold, the local expansion rate, and the possible uneven wall thickness trend.

[0158] The calculation module 204 receives the forming state parameters output by the estimation module, compares them with the preset ideal target state, quantifies the difference between the two, and calculates the specific adjustment amount required for the subsequent stages of the current blowing pressure curve based on the difference.

[0159] The control module 205 receives the adjusted blowing pressure parameters output by the calculation module, converts them into control signals, and sends them to the actuator of the blowing system. It adjusts the pressure of the compressed gas entering the mold cavity in real time, which directly acts on the glass and affects its local flow and expansion behavior, thereby correcting deviations in the forming process.

[0160] Specifically, the system achieves real-time closed-loop control of the glass container press blow molding process through the collaborative work of its modules. The acquisition module 201 continuously acquires local pressure data of the mold wall, which directly reflects the dynamic rheological behavior of glass within a complex-shaped mold.

[0161] The processing module 202 performs rapid and effective filtering and calibration on these raw data to eliminate interference and errors, obtaining processed data that reflects the true pressure. The processed data is then sent to the estimation module 203, where the forming state estimation model parses the current forming state parameters of the glass. These parameters provide key information about whether the glass expansion is uniform and whether it fits the mold as expected.

[0162] The calculation module 204 compares these real-time estimated state parameters with the predetermined target state, identifies deviations in the forming process, and calculates how to adjust the blowing pressure based on the deviations, such as increasing or decreasing the pressure amplitude at a certain stage, changing the blowing duration or rate, to guide the glass towards the target state.

[0163] Finally, the control module 205 sends the calculated adjustment parameters to the blowing system. The blowing system precisely adjusts the blowing pressure according to these instructions, which directly affects the local flow of the glass. This allows for rapid response to differences in the initial state and dynamic changes of the droplets within a very short forming cycle, effectively solving the problems of uneven wall thickness and stress concentration that are prone to occur in the forming of complex-shaped glass containers, and improving the stability of product quality.

[0164] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0165] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for controlling the pressure blowing of a glass container, characterized in that, The method includes the following steps: S1: Real-time acquisition of local pressure data on the mold wall during the glass blowing and expansion process to obtain pressure array data; S2: Filter and calibrate the pressure array data to obtain processed pressure data; S3: Input the processed pressure data into the forming state estimation model to estimate the forming state parameters of the glass in the mold; S4: Calculate the deviation between the forming state parameters and the target state, and dynamically adjust the subsequent stage parameters of the blowing pressure curve according to the deviation to obtain the adjusted blowing pressure parameters. S5: The adjusted blowing pressure parameters are sent to the blowing system to control the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow behavior of the glass.

2. The glass container pressure blowing control method according to claim 1, characterized in that, Step S1 includes: S11: Divide the critical area on the cavity wall of the blow molding die, and arrange multiple thin film pressure sensors in the critical area in an array, with each thin film pressure sensor integrating a temperature compensation circuit. S12: Synchronously acquire pressure and temperature data from all thin-film pressure sensors to obtain pressure array data containing pressure and temperature information; S13: Based on the pre-calibrated pressure sensor temperature-pressure calibration curve, the pressure array data is calibrated in real time using the collected temperature data to eliminate temperature drift error and obtain the calibrated pressure array data.

3. The glass container pressure blowing control method according to claim 2, characterized in that, Step S13 includes: S131: Construct a standard temperature field containing multiple temperature gradient levels, and record the pressure output value of each pressure sensor under each temperature gradient level; S132: For each pressure sensor, fit the temperature-pressure calibration curve of that pressure sensor based on the pressure output value; S133: Calculate the pressure correction value at this temperature based on the temperature data and the temperature-pressure calibration curve; S134: Subtract the pressure correction value from the pressure array data to eliminate temperature drift error and obtain the calibrated pressure array data.

4. The glass container pressure blowing control method according to claim 1, characterized in that, Step S2 includes: S21: The pressure array data is initially filtered using a moving average filtering algorithm to obtain the pre-filtered pressure array data; S22: Construct a multi-scale decomposition model based on wavelet transform to decompose the pre-filtered pressure array data into components with multiple different frequency scales; S23: For components at different frequency scales, an adaptive threshold filtering method is used for refined filtering; S24: Perform inverse wavelet transform on each frequency component after fine filtering to reconstruct the filtered pressure array data; S25: Establish a database of static characteristic parameters of pressure sensors, which includes the sensitivity and zero-point drift parameters of each pressure sensor; S26: Based on the pressure sensor static characteristic parameter database, the filtered pressure array data is calibrated using the least squares method to obtain the processed pressure array data.

5. The glass container pressure blowing control method according to claim 4, characterized in that, Step S23 includes: S231: Calculate the energy value of each frequency component within a preset time window and construct an energy distribution map; S232: Determine the energy threshold of each frequency component based on the energy distribution diagram; S233: Compare the amplitude of each frequency component with the corresponding energy threshold. If the amplitude is less than the energy threshold, it is determined to be noise, and the amplitude is set to zero. If the amplitude is greater than or equal to the energy threshold, it is determined to be a valid signal, and the amplitude is retained. S234: Integrate the frequency components after amplitude adjustment to obtain the finely filtered pressure array data.

6. The glass container pressure blowing control method according to claim 1, characterized in that, Step S3 includes: S31: Construct a forming state estimation model containing a radial basis function neural network. The input of the neural network is the processed pressure array data, and the output is the contact area between the glass and the mold, the local expansion rate, and the wall thickness unevenness trend. S32: Train a radial basis function neural network using finite element simulation data and actual production data to obtain the trained forming state estimation model; S33: Input the processed pressure array data into the forming state estimation model to estimate the forming state parameters of the glass in the mold. The forming state parameters include the contact area between the glass and the mold, the local expansion rate, and the wall thickness unevenness trend.

7. The glass container pressure blowing control method according to claim 6, characterized in that, Step S32 includes: S321: Construct a finite element model, which includes the initial temperature, shape and position deviation of the molten droplet, to simulate the evolution of the temperature field, stress field and deformation field of the glass during the pressure blow molding process, and to obtain the pressure array data of the mold cavity wall and the corresponding glass container wall thickness distribution data under different initial states. S322: Normalize the finite element simulation data and select a subset of samples from the normalized finite element simulation data to construct the initial training dataset; S323: Collect pressure array data and wall thickness distribution data of different batches of glass containers during the actual production process as actual production data, and fuse the actual production data with the initial training dataset to obtain a hybrid training dataset; S324: Divide the hybrid training dataset into K subsets, select K-1 subsets as the training set each time, and use the remaining 1 subset as the validation set to train the radial basis function neural network. Adjust the expansion speed and the number of hidden layer neurons of the neural network according to the root mean square error on the validation set to obtain the neural network model. S325: Train the neural network model and monitor the root mean square error on the validation set. When the root mean square error on the validation set does not decrease within N consecutive epochs, stop training, restore the weights and thresholds of the neural network to the optimal state, and obtain the forming state estimation model.

8. The glass container pressure blowing control method according to claim 1, characterized in that, Step S4 includes: S41: Establish a mapping relationship model, which takes the forming state parameters as input and the subsequent stage parameters of the blowing pressure curve as output; S42: Determine the target state parameters, which include the target contact area, the target local expansion rate, and the target wall thickness uniformity. S43: Calculate the deviation between the forming state parameters and the target state parameters to obtain the deviation vector; S44: Input the deviation vector into the mapping relationship model, and calculate the adjustment amount of subsequent stage parameters of the blowing pressure curve through the mapping relationship model. The adjustment amount includes the blowing pressure amplitude, blowing time and blowing rate. S45: The adjustment amount of the subsequent stage parameters of the blowing pressure curve is superimposed with the current blowing pressure curve parameters to obtain the adjusted blowing pressure parameters.

9. The glass container pressure blowing control method according to claim 8, characterized in that, Step S41 includes: S411: Construct a BP neural network, which includes an input layer, a hidden layer, and an output layer. The nodes in the input layer correspond to the forming state parameters, and the nodes in the output layer correspond to the subsequent stage parameters of the blowing pressure curve. S412: Train a BP neural network using offline experimental or simulation data, measure the forming state parameters by changing the subsequent stage parameters of the blowing pressure curve, minimize the mean square error between the network output and the target output, and obtain the mapping relationship model.

10. A glass container pressure blowing control system, characterized in that, The system, applied in the steps of the method according to any one of claims 1-9, comprises: Acquisition module: Real-time acquisition of local pressure data on the mold wall during the glass blowing and expansion process to obtain pressure array data; Processing module: Filters and calibrates the pressure array data to obtain processed pressure data; Estimation module: Inputs the processed pressure data into the forming state estimation model to estimate the forming state parameters of the glass in the mold; Calculation module: Calculates the deviation between the forming state parameters and the target state, and dynamically adjusts the subsequent stage parameters of the blowing pressure curve according to the deviation to obtain the adjusted blowing pressure parameters; Control module: Sends the adjusted blowing pressure parameters to the blowing system to control the pressure of the compressed gas entering the mold cavity in real time, thereby affecting the local flow behavior of the glass.

Images