Color master batch extrusion molding defect prediction method and system based on rheology model

Abstract

The present application relates to the technical field of verification optimization, in particular to a color master batch extrusion molding defect prediction method and system based on a rheology model. The method comprises the following steps: obtaining rheology test data by performing rheology test; performing parameter fitting on a rheology model based on the rheology test data to obtain a rheology prediction model; obtaining a melt flow, an extrusion temperature and a mold geometry parameter; performing fluid mechanics calculation to obtain a wall shear rate at a mold outlet; inputting the wall shear rate into the rheology prediction model to obtain a wall shear stress; monitoring a processing process to obtain multi-source data; time-aligning the wall shear rate, the wall shear stress and the multi-source data to obtain a processing time sequence; and analyzing and predicting based on the processing time sequence to obtain a roughness early warning parameter. The present application realizes the prediction and early warning of surface roughness through the rheology model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of verification optimization, in particular to a color master batch extrusion molding defect prediction method and system based on rheology model. BACKGROUND

[0002] In the field of modern engineering design and manufacturing, the deep integration and application of computer-aided design and computer-aided manufacturing technology are increasingly widespread, and have been applied to color master batch production processes.

[0003] However, the existing technology often relies on a single data source in obtaining design parameters, simulation data and manufacturing process information, which makes it difficult to meet the needs of complex calculation models and simulation prediction in terms of data precision and integrity, making it difficult to cope with dynamic changes in design requirements and high-precision simulation tasks. Relying on repeated physical prototype manufacturing and experimental testing, this approach not only has high costs, time-consuming and labor-intensive, but also makes it difficult to identify potential problems in the early stages of design, resulting in long design iteration cycles.

[0004] With the rapid development of data processing technology and computational modeling capabilities, applying advanced information processing technology to complex product design, simulation and manufacturing processes has become an important way to improve efficiency and accuracy. Especially in the design space exploration involving multiple parameters and multiple constraint conditions, the existing technology has been unable to meet the growing complexity needs.

[0005] Therefore, a color master batch extrusion molding defect prediction method and system based on rheology model are proposed. SUMMARY

[0006] The purpose of the present application is to provide a color master batch extrusion molding defect prediction method and system based on rheology model, which includes performing rheology test to obtain rheology test data; performing parameter fitting on the rheology model based on the rheology test data to obtain a rheology prediction model; obtaining the melt flow rate, extrusion temperature and mold geometry parameters; performing fluid mechanics calculation to obtain the wall shear rate at the mold outlet; inputting the wall shear rate into the rheology prediction model to obtain the wall shear stress; monitoring the processing process to obtain multi-source data; time-aligning the wall shear rate, wall shear stress and multi-source data to obtain a processing time sequence; analyzing and predicting based on the processing time sequence to obtain roughness early warning parameters.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] The color master batch extrusion molding defect prediction method based on rheology model comprises:

[0009] The rheological test is performed on the color master batch to obtain shear viscosity data of the color master batch at different shear rates at a preset processing temperature, and rheological test data is obtained;

[0010] Parameters of a preset rheological model are fitted based on the rheological test data, and a rheological prediction model is obtained;

[0011] The melt flow rate, extrusion temperature and die geometry parameters in the color master batch processing process are obtained;

[0012] Fluid mechanics calculation is performed based on the melt flow rate, extrusion temperature and die geometry parameters, and the wall shear rate at the die outlet is obtained; the wall shear rate is input into the rheological prediction model, and the wall shear stress is obtained;

[0013] The processing process is monitored, and multi-source data is obtained; the wall shear rate, wall shear stress and multi-source data are time-aligned, and a processing time sequence is obtained; analysis and prediction are performed based on the processing time sequence, and a roughness early warning parameter is obtained.

[0014] The construction of the rheological prediction model includes:

[0015] A digital rheometer is used to detect the color master batch processing process, and shear viscosity data of the color master batch at different shear rates at a preset processing temperature is measured;

[0016] Data cleaning is performed based on the processing temperature, shear rate and shear viscosity data, and rheological test data is obtained;

[0017] A power law model with temperature correction is determined as the preset rheological model, and parameter fitting is performed on the preset rheological model based on the rheological test data; an iterative optimization algorithm with multi-batch data fusion is performed in a digital simulation environment to minimize the prediction error and determine the optimal model parameters;

[0018] The optimal model parameters are applied to the preset rheological model, and a rheological prediction model is obtained.

[0019] The melt flow rate, extrusion temperature and die geometry parameters in the color master batch processing process are realized by combining real-time sensor data acquisition and preset database calling;

[0020] The melt flow rate is estimated in real time by the screw rotation speed of the extruder and the metering section efficiency; the extrusion temperature is obtained in real time by the melt temperature sensor;

[0021] The die geometry parameters are called from the pre-stored digital model database and dynamically updated in combination with real-time die wear monitoring.

[0022] The fluid mechanics calculation is performed based on the finite difference method;

[0023] The wall shear rate and wall shear stress at the die outlet are obtained by parallel computing processing the melt flow rate, extrusion temperature and die geometry parameters.

[0024] An integrated machine learning model is trained and optimized by labeled defect information in historical processing data, so that the model learns the complex nonlinear relationship between the processing time sequence and the surface roughness defect, and obtains a roughness prediction model; the historical processing data includes historical processing time sequence data and corresponding surface roughness defect labels;

[0025] The trained roughness prediction model is applied to dynamically identify the color master batch surface roughness defect in combination with the current processing time sequence; the multi-source data includes extrusion temperature fluctuation, screw speed stability, melt pressure fluctuation, color master batch batch rheological difference factor and die geometry change amount.

[0026] The identification process of the roughness prediction model includes:

[0027] Feature engineering module: sliding window statistics, time domain and frequency domain feature extraction and correlation analysis are performed on the processing time sequence to generate a high-dimensional feature vector;

[0028] Sequence analysis module: a sequence model in deep learning is used to process long-term processing time sequence to capture the time sequence dependence, evolution trend and strong correlation pattern between parameters and defect occurrence;

[0029] Decision-making module: a shallow machine learning model is used to classify and regress the extracted high-dimensional feature vector, time sequence dependence, evolution trend and strong correlation pattern between parameters and defect occurrence to obtain the surface roughness defect; the surface roughness defect is used to calculate and warn the roughness warning parameters;

[0030] Model training and optimization: data preprocessing and expansion, multi-task learning and transfer learning are used to train and optimize the model.

[0031] The color master batch extrusion molding defect prediction system based on rheology model includes:

[0032] The simulation test module performs rheological test on the color master batch to obtain the shear viscosity data of the color master batch at different shear rates under the preset processing temperature, and obtains the rheological test data;

[0033] The model training module performs parameter fitting on the preset rheological model based on the rheological test data to obtain a rheological prediction model;

[0034] The parameter acquisition module acquires the melt flow rate, extrusion temperature and die geometry parameters in the color master batch processing process;

[0035] A parameter calculation module performs fluid mechanics calculation based on the melt flow, extrusion temperature and mold geometric parameters to obtain a wall shear rate at the mold outlet; and inputs the wall shear rate into a rheology prediction model to obtain a wall shear stress;

[0036] An analysis and early warning module monitors the processing to obtain multi-source data; aligns the wall shear rate, wall shear stress and multi-source data in time to obtain a processing time sequence; and performs analysis and prediction based on the processing time sequence to obtain a roughness early warning parameter.

[0037] Compared with the prior art, the present application has the following beneficial effects:

[0038] 1. High-precision data acquisition and data cleaning are performed by a digital rheometer, a power law model with temperature correction is used, the rheological behavior of the color master batch under different processing temperatures and shear rates can be more accurately described, and the universality of the model is enhanced; parameter fitting is performed in a digital simulation environment through an iterative optimization algorithm of multi-batch data fusion, the contingency of a single batch of data can be effectively overcome, the fitting accuracy and robustness of the model parameters are improved, and a data basis is provided for subsequent defect prediction.

[0039] 2. Fluid mechanics calculation is performed by using a finite difference method, the flow behavior of the color master batch melt in a complex mold flow channel is accurately calculated, and the wall shear rate and wall shear stress at the mold outlet are accurately obtained; further introduction of parallel computing technology significantly shortens the fluid mechanics calculation time, ensures the real-time response capability of the prediction system, and provides fast calculation support for online defect early warning.

[0040] 3. By integrating a machine learning model, the model can learn and capture the complex nonlinear relationship between the processing time sequence and the surface roughness defect, the intelligence and accuracy of defect identification are improved, and the limitations of traditional single parameter threshold judgment are overcome; meanwhile, the fusion of multi-source data provides comprehensive and rich feature information for the model, enhances the generalization ability and prediction accuracy of the roughness prediction model, and realizes dynamic identification and early warning of the surface roughness defect of the color master batch. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 Fig. 1 is a flowchart of a color master batch extrusion molding defect prediction method based on a rheology model of the present application;

[0042] Figure 2 Fig. 3 is a structural diagram of a roughness prediction model of the present application;

[0043] Figure 3 Fig. 4 is a structural diagram of a color master batch extrusion molding defect prediction system based on a rheology model of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application. EMBODIMENT

[0045] The present application proposes a color master batch extrusion molding defect prediction method based on a rheology model, and the process is shown in Figure 1 , which comprises the following steps.

[0046] Rheological testing is performed on the color master batch to obtain shear viscosity data of the color master batch at different shear rates at a preset processing temperature, and rheological test data is obtained.

[0047] Based on the rheological test data, parameter fitting is performed on the preset rheological model to obtain a rheological prediction model.

[0048] The melt flow, extrusion temperature and mold geometry parameters in the color master batch processing process are obtained.

[0049] Based on the melt flow, extrusion temperature and mold geometry parameters, fluid mechanics calculation is performed to obtain the wall shear rate at the mold outlet; the wall shear rate is input into the rheological prediction model to obtain the wall shear stress.

[0050] The processing process is monitored to obtain multi-source data; the wall shear rate, wall shear stress and multi-source data are time-aligned to obtain a processing time sequence; based on the processing time sequence, analysis and prediction are performed to obtain a roughness early warning parameter.

[0051] Preferably, the construction of the rheological prediction model comprises the following steps.

[0052] A digital rheometer is used to detect the color master batch, and shear viscosity data of the color master batch at different shear rates is measured at a preset processing temperature.

[0053] Based on the processing temperature, shear rate and shear viscosity data, data cleaning is performed to obtain rheological test data.

[0054] A power law model with temperature correction is determined as the preset rheological model, and parameter fitting is performed on the preset rheological model by using the rheological test data; an iterative optimization algorithm using multi-batch data fusion is performed in a digital simulation environment to minimize the prediction error and determine the optimal model parameters.

[0055] Apply the optimal model parameters to the preset rheological model to obtain a rheological prediction model.

[0056] At a preset processing temperature, such as 180-220℃, a digital rheometer such as a high-precision rotary rheometer or a capillary rheometer is used to set multiple shear rate points at logarithmic or uniform intervals in the shear rate range to , and rheological testing is performed on the color master batch to obtain shear viscosity data of the color master batch at different shear rates at the preset processing temperature, which is arranged into rheological test data.

[0057] The power-law model with temperature correction is determined as the preset rheological model, and parameter fitting is performed on the preset rheological model through the rheological test data; including using the Levenberg-Marquardt iterative optimization algorithm based on historical multi-batch data fusion in the Python programming environment (combined with the Scipy.optimize library) or the Matlab simulation environment to minimize the root mean square error and determine the optimal model parameters; and applying the optimal model parameters to the preset rheological model to obtain a rheological prediction model.

[0058] The power-law model is a simplified model for describing the relationship between the viscosity of a non-Newtonian fluid and the shear rate, especially for shear-thinning fluids.

[0059] The temperature correction part takes into account the influence of temperature; because the viscosity of a polymer solution is very sensitive to temperature, the higher the temperature, the thinner the solution will generally become, and the better the flowability. In the form of the Arrhenius equation, the sensitivity of viscosity to temperature change is quantified through a flow activation energy parameter. The greater the activation energy, the greater the influence of temperature on viscosity.

[0060] The power-law part captures the characteristics of the change in viscosity of the solution with shear rate; for most color master batch solutions, they belong to shear-thinning fluids, and the greater the shear rate, the lower the viscosity. The power-law model quantifies the degree of viscosity change through a power-law index, and represents the inherent viscosity of the solution through a consistency coefficient.

[0061] Combining the two parts together, a comprehensive model is obtained, which can more accurately reflect the change rule of the viscosity of the color master batch solution when the temperature and flow rate change in actual extrusion processing.

[0062] The present application ensures the accuracy and reliability of rheological test data by high-precision data acquisition and data cleaning through a digital rheometer. The power law model with temperature correction can more accurately describe the rheological behavior of color master batches at different processing temperatures and shear rates, enhancing the universality of the model. Through the iterative optimization algorithm of multi-batch data fusion, parameter fitting is carried out in the digital simulation environment, effectively overcoming the randomness of single batch data, improving the fitting accuracy and robustness of model parameters, and providing a more solid data and model foundation for subsequent defect prediction.

[0063] Preferably, the melt flow, extrusion temperature and mold geometry during the processing of the color master batch are realized by combining real-time sensor data acquisition and preset database calling;

[0064] The melt flow is estimated in real time by the screw rotation speed of the extruder and the efficiency of the metering section; the extrusion temperature is obtained in real time by the melt temperature sensor;

[0065] Further, the online accurate melt flow measurement method based on machine vision and differential pressure coupling supplements the screw rotation speed and metering section efficiency estimation method, and the output flow value will correct the preliminary flow value estimated by the screw rotation speed and metering section efficiency. Combined with real-time correction of melt density by machine vision, differential pressure array sensing flow rate distribution, and multi-source data fusion and self-calibration technology, online, high-precision, high-robustness measurement of color master batch melt flow is realized; the real-time and accuracy of the flow data are significantly improved, thereby providing more reliable input for subsequent fluid mechanics calculation and defect prediction.

[0066] Specifically, considering the influence of the slight change of the density of the color master batch melt at different temperatures and pressures on the flow calculation, a high-resolution industrial camera and a laser thickness measurement system are added at the outlet end of the extrusion die. The cross-sectional size of the extruded melt at the moment of leaving the die is obtained in real time, and the image of the melt cross section is captured; the actual width and geometry of the melt are accurately identified through edge detection and pixel analysis. Based on historical experimental data and a preset polymer equation of state, a machine learning model such as a multilayer perception or a support vector regression is established, which takes the real-time extrusion temperature, melt pressure and melt cross-sectional size obtained by machine vision as input, and outputs the corrected melt density in real time.

[0067] At key positions in the flow channel of the extrusion die, an array of high-sensitivity micro-differential pressure sensors is arranged along the flow direction and the cross-sectional direction; for example, at the die inlet, the middle section, and before the die outlet. The sensor array synchronously collects pressure gradient information at different positions in the flow channel, with a sampling frequency not less than 200 Hz. The real-time differential pressure data collected are input into a pre-trained inverse rheology model, which is constructed based on a finite element model or a reduced-order model simulated by computational fluid dynamics. The instantaneous melt flow rate is accurately calculated by area integration of the real-time local flow velocity distribution obtained by inverse calculation at the die outlet cross section.

[0068] The melt density data corrected by machine vision, the flow velocity distribution data obtained by inverse calculation of the differential pressure array, and the estimated value of the screw rotation speed and the metering section efficiency are converged to a central data fusion unit.

[0069] An extended Kalman filter or an unscented Kalman filter algorithm is applied to fuse the above-mentioned multi-source heterogeneous data. The filter can effectively handle sensor noise and model uncertainty, provide real-time and high-precision optimal estimates of melt flow rate, and adaptively calibrate the metering section efficiency parameter in the traditional estimation formula. The fusion algorithm simultaneously performs anomaly detection on the sensor data, and issues a sensor failure or system anomaly warning when the data deviates significantly.

[0070] The die geometric parameters are retrieved from a pre-stored digital model database and dynamically updated in combination with real-time die wear monitoring.

[0071] The die wear monitoring is performed continuously at a preset frequency by online optical detection or displacement sensors. After data cleaning and preprocessing, the collected wear data are used to update the key geometric parameters such as the die inlet diameter, the flow channel length, and the cross-sectional area in the digital model database in real time. The update strategy is as follows: when the wear amount reaches a preset threshold, the system automatically adds the wear amount to the original geometric parameters to form a dynamically updated die geometric model.

[0072] The die wear includes mechanical wear and thermal deformation.

[0073] The present application ensures the real-time accuracy of key processing parameters such as melt flow rate and extrusion temperature by combining real-time sensor data acquisition, estimation, and database calling. In particular, the die geometric parameters can be dynamically updated in combination with real-time wear monitoring, so that the die model used for fluid mechanics calculation is more close to the actual wear state, avoiding prediction deviation caused by die wear. This comprehensive and dynamic parameter acquisition mechanism significantly improves the accuracy and adaptability of fluid mechanics calculation and subsequent defect prediction, making the system more capable of responding to dynamic changes in actual production.

[0074] Preferably, the fluid mechanics calculation is performed based on the finite difference method;

[0075] By parallel computing processing the melt flow rate, extrusion temperature and mold geometry parameters, the wall shear rate and wall shear stress at the mold outlet are obtained.

[0076] Obtain the melt flow rate, extrusion temperature and key geometry parameters of the mold during the processing of the color master batch, including but not limited to the die diameter of the mold, the length of the mold flow channel, the cross-sectional area of the mold flow channel;

[0077] Based on the melt flow rate, extrusion temperature and mold geometry parameters, fluid mechanics software is used to perform fluid mechanics calculation by finite difference method, combined with power-law fluid or Bird-Carreau model, to obtain the wall shear rate at the mold outlet; the wall shear rate is input into the rheological prediction model to obtain the wall shear stress. Further, precision calibration can be performed by finite element analysis or finite volume method.

[0078] By using finite difference method for fluid mechanics calculation, the present application can accurately simulate the flow behavior of color master batch melt in complex mold flow channel, so as to accurately obtain the wall shear rate and wall shear stress at the mold outlet. Further, the introduction of parallel computing technology significantly shortens the fluid mechanics calculation time, ensures the real-time response ability of the prediction system, and provides fast calculation support for online defect early warning.

[0079] Preferably, an integrated machine learning model is trained and optimized by labeled defect information in historical processing data, so that the model learns the complex nonlinear relationship between the processing time sequence and the surface roughness defect, and obtains a roughness prediction model; the historical processing data includes historical processing time sequence data and corresponding surface roughness defect labels;

[0080] The trained roughness prediction model is applied to dynamically identify the surface roughness defect of the color master batch combined with the current processing time sequence; the multi-source data includes extrusion temperature fluctuation, screw speed stability, melt pressure fluctuation, color master batch batch rheological difference factor and mold geometry change amount.

[0081] The processing process is monitored to obtain multi-source data; the wall shear rate, wall shear stress and multi-source data are time-aligned by a high-precision timestamp-based synchronization algorithm, and linear interpolation or spline interpolation method is used to handle the inconsistent sampling frequency of different sensors or occasional data loss, to obtain the processing time sequence.

[0082] Based on the processing time sequence, an integrated machine learning model is trained and optimized by historical data for analysis and prediction to obtain roughness warning parameters; for example, the root mean square value or peak-to-valley value of the surface roughness, and the preset threshold is used for early warning.

[0083] The specific early warning includes surface roughness defects of the color master batch, expressed by the root mean square value of surface roughness, and a roughness defect threshold is set based on the production standard and industry requirements of the color master batch; the predicted surface roughness defects are identified according to the roughness defect threshold to determine the roughness early warning parameter.

[0084] The multi-source data includes: the standard deviation of the extrusion temperature per unit time, the average deviation of the screw rotation speed in the set interval, the fluctuation range of the melt pressure in the set time window, the similarity evaluation factor (for example, the correlation coefficient or the Euclidean distance) of the rheological curves of different batches of color master batches, and the geometric size change amount of the mold key size monitored in real time through online optical detection or displacement sensors. Among them, the rheological difference factor of the color master batch is evaluated by comparing the similarity of the rheological curve of the current batch of color master batch and the rheological curve of the standard batch (or the average batch).

[0085] The model integrating the convolutional neural network (CNN) in deep learning and the long short-term memory network (LSTM) is trained and optimized through the labeled defect information in the historical processing data, the Adam optimizer is used during training, the learning rate is set to 0.001, the batch size is 32, and the iteration number is 100 cycles, so that the model learns the complex nonlinear relationship between the processing time sequence and the surface roughness defect, and a roughness prediction model is obtained; the historical processing data includes historical processing time sequence data and corresponding surface roughness defect labels.

[0086] The CNN part includes two convolutional layers, the first layer has 16 convolutional kernels with a size of 3x3 and ReLU activation; the second layer has 32 convolutional kernels with a size of 3x3 and ReLU activation, followed by a maximum pooling layer. The LSTM part includes two LSTM layers, each containing 128 units. For the roughness regression task, the mean square error (MSE) is used, and for the defect classification task, the cross-entropy loss is used, and the weighted sum is used as the total loss function.

[0087] The present application integrates a machine learning model, especially uses the labeled defect information in the historical processing data for training, so that the model can autonomously learn and capture the complex nonlinear relationship between the processing time sequence and the surface roughness defect. This greatly improves the intelligence and accuracy of defect identification, and surpasses the limitations of traditional single parameter threshold judgment. The fusion of multi-source data provides more comprehensive and rich feature information for the model, further enhances the generalization ability and prediction accuracy of the roughness prediction model, and realizes dynamic identification and early warning of the surface roughness defects of the color master batch.

[0088] Preferably, the structure of the roughness prediction model is as shown in Figure 2 The identification process includes:

[0089] Feature engineering module: sliding window statistics, time and frequency domain feature extraction and correlation analysis are performed on the machining time series to generate high-dimensional feature vectors;

[0090] Sequence analysis module: sequence models in deep learning are used to process long-term machining time series to capture the time sequence dependency between parameters, evolution trend and strong association pattern between defect occurrence;

[0091] Decision-making module: shallow machine learning models are combined to classify and regress the extracted high-dimensional feature vectors, time sequence dependency, evolution trend and strong association pattern between defect occurrence to obtain surface roughness defects; roughness warning parameters are calculated and warned according to the surface roughness defects;

[0092] Model training and optimization: data preprocessing and expansion, multi-task learning and transfer learning are used to train and optimize the model.

[0093] Feature engineering module: sliding window statistics with a size of 30 seconds and a step of 5 seconds are performed on the machining time series, such as calculating the mean, standard deviation, maximum value and minimum value in the window; time domain feature extraction is performed, such as extracting the mean, variance, skewness, kurtosis and zero-crossing rate of the sequence; and frequency domain feature extraction is performed, such as extracting the main frequency and energy distribution of the signal through fast Fourier transform, and Pearson correlation coefficient analysis is performed to generate high-dimensional feature vectors;

[0094] Sequence analysis module: a bidirectional long short-term memory network (Bi-LSTM) based on Attention mechanism is used to process long-term machining time series, which contains two layers of Bi-LSTM layers, each with 128 hidden units and ReLU activation function, to capture the time sequence dependency between parameters, evolution trend and strong association pattern between defect occurrence;

[0095] The Attention mechanism used is Bahdanau Attention, which generates attention weights by calculating the similarity between the hidden states of the Bi-LSTM layer output and a context vector; the attention weights are then used to weight the output of the Bi-LSTM layer to highlight the time step information that contributes more to the prediction result.

[0096] Decision-making module: gradient boosting decision tree models in ensemble learning are combined to classify and regress the extracted high-dimensional feature vectors, time sequence dependency, evolution trend and strong association pattern between defect occurrence to obtain surface roughness defects; roughness warning parameters are calculated and warned according to the surface roughness defects;

[0097] Model training and optimization: data preprocessing through data standardization and outlier removal; data augmentation through time series random jitter, scaling and cropping; multi-task learning by predicting roughness level and defect type simultaneously, and setting different loss function weights for different tasks; transfer learning by fine-tuning a pre-trained general machining defect prediction model to train and optimize the model.

[0098] The identification process inside the roughness prediction model includes feature engineering, sequence analysis, decision discrimination and model training and optimization; the feature engineering module can extract high-dimensional features valuable for defect prediction from the original machining time series, effectively reducing data dimension and improving feature expression ability; the sequence analysis module uses a deep learning sequence model that can deeply mine the complex time series dependence, evolution trend and strong correlation pattern between machining parameters and defects, improving the model's understanding of dynamic processes. The decision discrimination module combines shallow machine learning models for classification and regression prediction to ensure the accuracy and interpretability of the prediction results; the model training and optimization module uses advanced techniques such as data preprocessing, augmentation, multi-task learning and transfer learning to significantly improve the model's generalization ability, training efficiency and prediction accuracy, ensuring the timeliness and accuracy of roughness early warning.

[0099] The present application proposes a color master batch extrusion molding defect prediction system based on rheological model, as shown in Figure 3 The present application proposes a color master batch extrusion molding defect prediction system based on rheological model, as shown in

[0100] The simulation test module performs rheological testing on the color master batch to obtain shear viscosity data of the color master batch at different shear rates under a preset processing temperature, and obtains rheological test data;

[0101] The model training module performs parameter fitting on the preset rheological model based on the rheological test data to obtain a rheological prediction model;

[0102] The parameter acquisition module acquires the melt flow, extrusion temperature and mold geometry parameters during the processing of the color master batch;

[0103] The parameter calculation module performs fluid mechanics calculation based on the melt flow, extrusion temperature and mold geometry parameters to obtain the wall shear rate at the mold outlet; and inputs the wall shear rate into the rheological prediction model to obtain the wall shear stress;

[0104] The analysis and early warning module monitors the processing process to obtain multi-source data; aligns the wall shear rate and wall shear stress with the multi-source data in time to obtain a processing time series; and performs analysis and prediction based on the processing time series to obtain roughness early warning parameters. Embodiment

[0105] The application provides a color master batch extrusion molding defect prediction method based on a rheology model, which comprises the following steps:

[0106] Rheology test is performed on the color master batch to obtain shear viscosity data of the color master batch at different shear rates at a preset processing temperature, and rheology test data are obtained.

[0107] Parameter fitting is performed on the preset rheology model based on the rheology test data, and a rheology prediction model is obtained.

[0108] The melt flow, extrusion temperature and mold geometric parameters in the color master batch processing process are obtained.

[0109] Fluid mechanics calculation is performed based on the melt flow, extrusion temperature and mold geometric parameters, and wall shear rate at the mold outlet is obtained; the wall shear rate is input into the rheology prediction model, and wall shear stress is obtained.

[0110] The processing process is monitored, and multi-source data are obtained; the wall shear rate, wall shear stress and multi-source data are time-aligned, and a processing time sequence is obtained; analysis and prediction are performed based on the processing time sequence, and a roughness early warning parameter is obtained.

[0111] Specifically, rheology test is performed on the target color master batch to obtain shear viscosity data thereof at a preset processing temperature.

[0112] The color master batch with polyethylene (PE) as a base material, a density of 0.92 g / cm3, titanium dioxide (TiO2) as a main pigment component, a weight content of about 30%, and a small amount of lubricant and dispersant is selected.

[0113] Before the test, the color master batch sample needs to be dried in a 80℃ vacuum oven for 4 hours to remove moisture. The test is performed at two temperature points of 190℃ and 200℃, and the temperature control accuracy is ±0.5℃; the test is performed in the shear rate range of 10s -1 to 1000s -1 Shear rate points are uniformly set in the logarithmic coordinate.

[0114] The original test data are cleaned, and transient response data in the starting stage and abnormal data points caused by sample degradation or bubbles are removed. The abnormal values are identified and removed through a three-standard-deviation criterion. The cleaned data are used as the rheology test data.

[0115] The melt flow is estimated in real time according to the screw rotation speed of the extruder and the metering section efficiency. The screw rotation speed is obtained in real time through a rotary encoder installed at the end of the extruder screw, and the sampling frequency is 100Hz. The metering section efficiency is determined through offline calibration experiments based on historical production data and the extruder model.

[0116] The collected wear data is processed by median filtering, and the corresponding die orifice diameter in the digital model database is updated in real time. When the die wear reaches the preset threshold, for example, 0.01 mm, the system automatically accumulates the wear to the original die diameter to form a dynamically updated die geometry model.

[0117] Fluid mechanics calculation: based on finite volume method.

[0118] Based on the dynamically updated die geometry parameters, a model of the die flow channel is created in ANSYS Space Claim, accurate structured hexahedral mesh is generated, and mesh encryption is performed near the die wall to capture the boundary layer flow behavior. Steady, incompressible, laminar flow model is adopted. Material properties are defined as non-Newtonian fluid, and viscosity model is selected as the above-mentioned power law model with temperature correction.

[0119] Boundary conditions are set as follows: mass flow inlet is set according to the real-time estimated melt flow. The pressure outlet is set to zero gauge pressure. The no-slip wall condition is set, and the die wall temperature is set to the real-time obtained extrusion temperature. Parallel computing is performed using multi-core CPU server, which significantly shortens the calculation time to seconds (usually <5 seconds), ensuring real-time response.

[0120] The processing process is monitored to obtain multi-source data, and the wall shear rate, wall shear stress and multi-source data are time-aligned to construct a processing time sequence, which is then analyzed and predicted to obtain roughness warning parameters.

[0121] Multi-source data monitoring:

[0122] Extrusion temperature fluctuation: the standard deviation of the melt temperature at the die inlet in the past 30 seconds is calculated; screw speed stability: the average deviation of the screw speed in the past 30 seconds is calculated; melt pressure fluctuation: by installing a melt pressure sensor before the die inlet, the peak-to-peak value (difference between maximum and minimum) of the melt pressure in the past 30 seconds is calculated; color master batch rheological difference factor: when each batch of color master batch is stored, rheological test is performed and compared with the rheological curve of the standard batch. Frechet distance is used as the similarity evaluation factor. The smaller the factor value, the closer the current batch is to the standard batch in rheological behavior. The factor is updated regularly. Die geometry change: determined by the real-time die wear monitoring system, i.e. the cumulative wear of the die orifice diameter.

[0123] A synchronization algorithm based on high-precision time stamp is used to synchronize the clocks of all sensors and computing units through the network time protocol with the server, ensuring that the time stamp accuracy is within milliseconds. For the case of inconsistent sampling frequencies of different sensors or occasional data missing, a cubic spline interpolation method is used to unify all data to a sampling frequency of 1 Hz, and the missing data is filled.

[0124] An integrated machine learning model is trained and optimized by labeled defect information in historical processing data to obtain a roughness prediction model and applied to real-time prediction.

[0125] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for predicting defects in masterbatch extrusion molding based on a rheological model, characterized in that, include: Rheological tests were conducted on the masterbatch to obtain the shear viscosity data of the masterbatch at different shear rates under the preset processing temperature, and thus the rheological test data were obtained. Based on rheological test data, the parameters of a pre-set rheological model are fitted to obtain a rheological prediction model. Obtain the melt flow rate, extrusion temperature, and die geometry parameters during the color masterbatch processing; Based on the melt flow rate, extrusion temperature, and die geometry parameters, fluid dynamics calculations are performed to obtain the wall shear rate at the die outlet; the wall shear rate is then input into a rheological prediction model to obtain the wall shear stress. The processing process is monitored to obtain multi-source data; the wall shear rate and wall shear stress are time-aligned with the multi-source data to obtain a processing time series; based on the processing time series, analysis and prediction are performed to obtain roughness warning parameters.

2. The method for predicting defects in masterbatch extrusion molding based on a rheological model according to claim 1, characterized in that: The construction of the rheological prediction model includes: The color masterbatch was tested using a digital rheometer, and the shear viscosity data of the color masterbatch at different shear rates was measured at a preset processing temperature. Data cleaning was performed based on processing temperature, shear rate, and shear viscosity data to obtain rheological test data. A power-law model with temperature correction is selected as the preset rheological model. The parameters of the preset rheological model are fitted using rheological test data. This includes an iterative optimization algorithm using multi-batch data fusion in a digital simulation environment to minimize prediction errors and determine the optimal model parameters. The optimal model parameters are applied to a preset rheological model to obtain a rheological prediction model.

3. The method for predicting defects in masterbatch extrusion molding based on a rheological model according to claim 1, characterized in that: The melt flow rate, extrusion temperature, and die geometry parameters during the masterbatch processing are achieved through a combination of real-time sensor data acquisition and pre-set database access. The melt flow rate is estimated in real time by the extruder screw speed and metering section efficiency; the extrusion temperature is acquired in real time by a melt temperature sensor. The mold geometry parameters are retrieved from a pre-stored digital model database and dynamically updated in conjunction with real-time mold wear monitoring.

4. The method for predicting defects in masterbatch extrusion molding based on a rheological model according to claim 1, characterized in that: The fluid dynamics calculations are performed based on the finite difference method. By processing the melt flow rate, extrusion temperature, and die geometry parameters in parallel, the wall shear rate at the die exit is obtained, and the wall shear stress is further obtained.

5. The method for predicting defects in masterbatch extrusion molding based on a rheological model according to claim 1, characterized in that: An integrated machine learning model is trained and optimized using labeled defect information from historical processing data, enabling the model to learn the complex nonlinear relationship between processing time series and surface roughness defects, thus obtaining a roughness prediction model; the historical processing data includes historical processing time series data and corresponding surface roughness defect labels; The trained roughness prediction model is applied to dynamically identify surface roughness defects of masterbatch by combining the current processing time series; the multi-source data includes extrusion temperature fluctuation, screw speed stability, melt pressure fluctuation, batch rheological difference factor of masterbatch, and die geometric change.

6. The method for predicting defects in masterbatch extrusion molding based on a rheological model according to claim 5, characterized in that: The identification process of the roughness prediction model includes: Feature engineering module: Performs sliding window statistics, time-domain and frequency-domain feature extraction, and correlation analysis on the processing time series to generate high-dimensional feature vectors; Sequence Analysis Module: Employs sequence models from deep learning to process long-term processing time series, capturing the temporal dependencies, evolution trends, and strong correlation patterns between parameters and defect occurrence; Decision-making module: Combining a shallow machine learning model, the module classifies and regresses the extracted high-dimensional feature vectors, temporal dependencies, evolution trends, and strong correlation patterns with defect occurrence to obtain surface roughness defects; and calculates and issues warnings based on the surface roughness defects to provide roughness warning parameters. Model training and optimization: The model is trained and optimized through data preprocessing and augmentation, multi-task learning, and transfer learning.

7. A color masterbatch extrusion molding defect prediction system based on a rheological model, characterized in that, include: The simulation test module performs rheological tests on the masterbatch to obtain shear viscosity data of the masterbatch at different shear rates under a preset processing temperature, thus obtaining rheological test data. The model training module fits the parameters of a preset rheological model based on rheological test data to obtain a rheological prediction model. The parameter acquisition module acquires the melt flow rate, extrusion temperature, and die geometry parameters during the masterbatch processing. The parameter calculation module performs fluid dynamics calculations based on the melt flow rate, extrusion temperature, and die geometry parameters to obtain the wall shear rate at the die outlet; the wall shear rate is then input into the rheological prediction model to obtain the wall shear stress. The analysis and early warning module monitors the processing process and obtains multi-source data; it aligns the wall shear rate and wall shear stress with the multi-source data in time to obtain a processing time series; and it analyzes and predicts based on the processing time series to obtain roughness early warning parameters.

Images